JP2904051B2 - System control unit - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Industrial applications]

[0001] The present invention relates to a system control device for controlling output from a system, and in particular, it can always perform optimal control at low cost and with high accuracy, and can also collect and optimize data during product development. The present invention relates to a system control device capable of reducing development man-hours required for design to substantially zero.

[Prior art]

[0002] Conventionally, methods using artificial intelligence technology have been widely used for the purpose of system control. Typical examples are a control method using an expert system based on rules that experts have gained empirically, a control method using model-based inference based on knowledge in system design,
There are control methods using fuzzy and neural networks.

However, the expert system cannot cope with an inexperienced situation where no rule exists, and can cope with an inexperienced situation on a model basis. To that end, a deep understanding of the operation mechanism of a controlled object (model) Is indispensable (“Inference based on cases and knowledge acquisition from cases,” Shigenobu Kobayashi,
Journal of the Japan Fuzzy Society, Vol. 4 No. 4P646-P6
55 (1992)).

[0004] Fuzzy and neural networks have excellent features such as being able to cope with complicated relationships between inputs and outputs. On the other hand, fuzzy techniques require engineers to perform tuning by trial and error. An engineer must prepare appropriate teacher data in advance and execute a large number (and therefore a long time) of learning cycles, usually from several thousand to several tens of thousands.

[0005] Therefore, it is not usually possible to cope with a system in which the relationship between the input and the output taken by tuning or learning changes. In other words, it is necessary to reconfigure the control system such as re-tuning the membership function in fuzzy and re-creating teacher data and re-learning in neuro, and realizing optimal control of the controlled system in real time is not possible. It was extremely difficult.

[0006] Therefore, as an inference method that does not depend only on the rules empirically obtained by the expert and does not require a deep understanding of the operation mechanism of the controlled object, an inference method based on past cases (Case Based Reason) is used.
ing = CBR) (JP-A-5-150).
989). In this CBR, a case closest to the current situation is searched from past cases, and an inference about the current situation can be made based on the case.

[0007] However, the conventional CBR is mainly applied to strategy support and the like, and there are few examples of using it for system control.
Because of the way CBR is used, it is usually necessary to store a large amount of comprehensive data as an example in advance, which requires a large number of development steps.

[0008] In addition, all the elemental technologies such as storage of cases, retrieval of cases, inference from cases, and additional learning of new cases, which are the elemental technologies of CBR, are still in the research stages, and improvements are needed for practical use. Is needed. ("Research issues of case-based reasoning" Shigenobu Kobayashi, Artificial Intelligence 75-4 (1991.3.6) P29-38)

[0009] Specifically, first, in the conventional CBR, in the similar case search, the similarity judgment does not have a unique quantitative criterion at present. Further, in the conventional CBR, an inference method has not been established when there is no past case sufficiently similar to the current situation, and the only option is to make uncertain inferences based on insufficient cases.

[0010] Unlike conventional fuzzy / neuro networks, the conventional CBR has a great feature that knowledge can be grown by learning a new case at any time, but the criterion for determining whether to learn a new case is clear. Rather, complicated work is required, such as the need for human intervention to reconstruct a new cluster. Moreover, even if there is a change over time between the old case and the new case (even if the old case is no longer useful), it cannot be distinguished, so there are many problems when used for system control.

[0011]

SUMMARY OF THE INVENTION The present invention is based on the disadvantages of the prior art described above, namely, rule-based inference represented by an expert system, case-based inference, model-based inference, and control methods using fuzzy and neural networks. The purpose of the present invention is to eliminate the drawbacks of the prior art, and the purpose is to perform control suitable for the situation while significantly reducing the number of development steps.

Further, the present invention has the following objects. The control device itself can automatically and selectively learn necessary information so that the control error falls within the allowable error range, thereby improving control performance. In order to further enhance control accuracy, control can be performed using a control rule most suitable for the system situation. When there are a plurality of control rules obtained under a situation close to the current situation, a more suitable control rule is created by effectively combining them. Easily extract unambiguous quantitative control rules. The degree of conformity between the current state of the system and past cases is uniquely and quantitatively determined. If the storage capacity for additional storage of a new control case becomes insufficient as a result of repeatedly storing (learning) additional control cases as necessary, the least important case is accurately selected and deleted. Be able to remember the latest more important cases. The configuration is such that the storage capacity for storing the control case is kept low while making full use of the information possessed by the control case. When the system is restarted, the effect of a change in the state of the system before and after the restart is reduced, and high-precision control can be performed promptly from the beginning of the restart.

The present invention has been made in order to solve the above problems.
In the invention according to claim 1, an operation amount output unit that supplies an operation amount to a control device, and the operation amount and a control amount output from the controlled device corresponding to the operation amount and the controlled device. Control case storage means for storing a set of state quantities affecting the control amount as a control case, and creating a cluster by grouping similar state quantities, and a function to be controlled for each of the clusters Model extraction
For the control rule extracting means to be performed and the function model extracted by the control rule extracting means, the degree of conformity to the immediately preceding control case is obtained, each function model is weighted according to the degree of conformity and averaged, and the result is obtained. An operation amount calculating means for obtaining an operation amount for making the control amount equal to a target value using a combination rule, and outputting the operation amount to the operation amount output means.

Further , according to the present invention, an operation amount output means for giving an operation amount to the controlled device, the operation amount and the control amount output from the controlled device corresponding thereto. A control case storing means for storing a set of state quantities affecting the control amount of the controlled device as a control case, and creating a cluster by grouping similar state amounts together; To the control target
For a control rule extracting means for extracting a number of models and a function model extracted by the control rule extracting means, a degree of conformity to the immediately preceding control case is obtained, and each function model is obtained.
Averaged by weighting according to the goodness of fit for Le, using the result obtained combining rule obtains a manipulated variable to match the target value of the control amount, the operation of outputting the operation amount to the operation amount output means When the control amount exceeds the permissible error range, the control case storage unit stores the control content in a corresponding cluster as a new control case, and stores the control rule extraction unit. Is characterized in that a new function model is extracted from the cluster including the control case.

Further, in the invention according to claim 4,
A set of manipulated variable output means for providing a manipulated variable to the controlled device, and a set of the manipulated variable, a control variable correspondingly output from the controlled device, and a state variable affecting the controlled variable of the controlled device; And memorize it as a control case,
A control case storage unit that creates clusters by grouping similar state quantities; and, for each cluster, each control in an (n + 1) -dimensional space composed of a control quantity and n related operation quantities. Extraction of function model of controlled object for each controlled variable as n-dimensional plane by least square error method of case coordinate points
And a function model extracted by the control rule extracting means, for which the degree of conformity to the immediately preceding control case is obtained, weighted according to the degree of conformity for each function model , averaged, and the result obtained. An operation amount calculating means for obtaining an operation amount for causing the control amount to coincide with the target value using the synthesis rule obtained, and outputting the operation amount to the operation amount output means. in the coordinate space function model is described, Seki
The reciprocity of the distance between the n-dimensional plane indicating the numerical model and the coordinate point indicating the immediately preceding control case is standardized for each function model to determine the degree of conformity.

Further , according to the twelfth aspect of the present invention,
A manipulated variable output means for providing a manipulated variable to the controlled device;
The operation amount and the control amount output from the controlled device corresponding thereto and the state amount affecting the control amount of the controlled device are stored as a set as a set of control cases, and the state amounts are similar. a control case memory means for creating a cluster collectively what is, the control to each of said clusters
Control rule extracting means for extracting a target function model ,
For function model <br/> extracted by the control rule extracting means calculates a goodness of fit for the immediate control result, each Seki
The number of models is weighted according to the degree of conformity and averaged, an operation amount for matching the control amount to the target value is obtained by using the resultant synthesis rule, and the operation amount is output to the operation amount output means. Operation amount calculation means,
When the controlled device or the control operation is restarted after the controlled device is stopped, the operation amount calculation unit determines the state amount before and after the restart with respect to the operation amount in the control case immediately before the stop .
A conversion for reducing the influence of the change is performed, the manipulated variable after the conversion is output to the manipulated variable output means, and the degree of conformity of each function model to the control result obtained thereby is obtained.

[0017]

[0018]

[0019]

[0020]

[0021]

[0022]

[0023]

[0024]

[0025]

[0026]

[0027]

[0028]

[0029]

[0030]

[0031]

According to the first aspect of the present invention, since the manipulated variable calculating means obtains the degree of conformity of each control rule, and weights the control rules to synthesize each control rule, the situation at the time of control and each control rule are synthesized. Quantitative conformity judgment with control rules can be performed.
High control accuracy is obtained. This makes it possible to infer a control rule corresponding to a situation experienced for the first time without having experienced a situation similar to the situation at the time of control in the past. That is, for example, a new control rule that can be expected to be adapted at the medium temperature can be created from the control rules at the time of high temperature and at the time of low temperature experienced in the past.

According to the second aspect of the present invention, when the control amount (system output) exceeds the set allowable error range, the control content is additionally stored as a control case to be newly stored. For the next or subsequent control, a control rule is automatically extracted from a control case group including the additionally stored control case. Therefore, it is possible to perform the control adapted to the change of the situation or the like, and the convergence is automatically reduced to the accuracy according to the set allowable error range.

In the invention according to claim 3 , the control rule is that the least squares of each control case coordinate point in an n + 1-dimensional space composed of a control amount and n operation amounts related thereto. Since each control amount is extracted as an n-dimensional plane by the error method, errors are reduced by statistical processing. In addition, since the control rules are obtained purely mathematically (as an n-dimensional plane using the least squares error method), it is not necessary to create a physical model for the controlled system. That is, there is an effect that the controlled system can be handled as a black box. Therefore, it is not always necessary to include a physical quantity related to the control amount as an element of the control case, so that a physical quantity sensor becomes unnecessary. That is, the set value of the operation amount, which is a more direct amount in the control, can be substituted.

According to the fourth aspect of the present invention, the automatic extraction of the control rules is performed for each cluster composed of control cases under similar conditions, and at the time of execution of control, the fitness of each cluster is determined. Weighted average of the control rules of the clusters according to the fitness. Then, the operation amount is calculated based on the control rule obtained as a result. In this case, the control rule is extracted as an n-dimensional plane by the least square error method of each control case coordinate point in an n + 1-dimensional space composed of each control amount and related n operation amounts for each control amount. And the degree of conformity between the state at the time of control execution and each cluster is determined by n of each cluster with respect to the same operation amount as in the previous control execution.
The reciprocal of the difference (distance in the coordinate space) between the point on the dimensional plane (control amount) and the control amount (control result) at the time of the previous control execution is
This is standardized over the entire similar cluster.

According to the fifth aspect of the present invention, when the capacity of the control case storage unit becomes insufficient, the oldest control case is erased, so that the storage area can be used effectively.

[0037] In claim 6, when the single cluster generation is complete, erases the instance or case group is a component of that cluster from the control case memory means. Therefore, only the control rule corresponding to the cluster is stored, and the amount of data to be stored can be significantly reduced.

According to the seventh and eighth aspects, the control case storage means erases the oldest cluster when the storage capacity becomes insufficient, so that a new cluster or control case can be reliably stored.

According to the ninth aspect , since the control cases used in the determination of the manipulated variable with a small number of times, that is, the control cases with low importance are deleted, the storage area can be used effectively, and the important data can be stored forever. It has the advantage of being preserved.

According to the tenth aspect , the cluster used less frequently for determining the operation amount, that is, the cluster of low importance is deleted, so that the storage area can be used effectively and the important cluster is stored forever. The advantage is that

According to the eleventh aspect of the present invention, clusters with a small cumulative fitness, that is, clusters with low importance are deleted, so that the storage area can be used effectively and important Clusters are kept forever.

According to the twelfth aspect of the invention, the degree of conformity of each control rule after the restart is determined by specifying an operation amount obtained by performing a predetermined conversion on the operation amount in the control case immediately before the stop. It is calculated based on the obtained control case, and the control rule after the restart is synthesized by the weighted average according to the degree of adaptation. The predetermined conversion includes various modes such as, for example, averaging or weighted averaging with a predetermined standard operation amount, averaging with an average value of the operation amount in each control case before stop, multiplication of a coefficient, and the like.

As described above, the system control device according to the present invention can perform real-time learning for maintaining necessary accuracy, and can automatically extract an optimal control rule even if a change over time occurs. . Therefore, by further utilizing this effect, it is possible to eliminate the need for the optimization work at the time of developing the system control device, which has been conventionally performed.

[0044]

【Example】

First Embodiment A: Configuration of First Embodiment Hereinafter, a first embodiment of the present invention will be described with reference to the drawings. This embodiment is an example of a general-purpose control system capable of controlling various objects. First,
1 shown in FIG. 1 is a controlled system, which has an actuator unit that changes a control target according to a supplied operation amount. Further, the controlled system 1 detects the state of the control target and outputs it as a control amount. For example, control target 1
Is a laser printer, the operation amount is an instruction value of the charging voltage of the photosensitive member or an instruction value of the laser power. If the control target is the print density, the detected value is the control amount. However, when the controlled system 1 uses the control target itself as the system output, the system output is the control amount.

Generally speaking, as the manipulated variable, various set values of factors (voltage set value, pressure set value, rotation angle of the adjustment volume, etc.) that can be set for adjusting the output of the controlled system are adopted. Then, as the control amount, the output or substitute amount of the controlled system is used. The output value is
It may be a physical quantity or a psychophysical quantity (good or bad score, etc.).

In this embodiment, for the sake of simplicity, the operation amount is set to a voltage set value, and the output value is set to a dimensionless number. Further, the controlled system 1 outputs a state quantity reflecting a situation where the apparatus is placed. In this case, any factor that affects the output of the controlled system 1 can be taken as the state quantity. That is, the situation in which the controlled system 1 is placed (temperature, humidity, pressure, time-dependent fluctuation of the installation location, load magnitude, etc.)
, And a plurality of types of state quantities may be used.

The state quantity is not a direct physical quantity,
A substitute may be used. For example, if mechanical wear occurs in proportion to the operation time, and the amount of mechanical wear is an important physical factor affecting the output of the controlled system 1, the accumulated operation time is used instead of the amount of wear. Can be substituted as quantity. In short, any quantity can be used as long as it is possible to grasp the control case under what condition from the viewpoint of the system output.

By the way, in the present embodiment, the case sampling time is used as the state quantity from the above viewpoint. This means that in the present embodiment, the intrinsic state quantity of the controlled system 1 varies over a wide range, such as various environmental conditions and deterioration over time, but is substantially constant over a limited time range. This is because it is assumed that what can be regarded as. For example, although it is different between morning and evening, the object is such that it can be considered that the state is substantially the same 10 minutes before and 10 minutes later.

Next, the system controller 2 will be described. First, the comparison unit 10 determines whether the control amount is within the allowable error, and outputs a signal Y to the storage control unit 11 if the control amount is within the allowable error, and outputs a signal N if the control amount is outside the allowable error. If the output signal of the comparison unit 10 is “N”, the storage control unit 11
The control amount, the state amount, and the operation amount at that time are combined into a set, and the set is stored in the case storage unit 15 as a new case. On the other hand, if the output signal of the comparison unit 10 is “Y”, the control amount at that time is discarded.

The correction rule creating unit 12 includes a case storage unit 15
A correction rule is created from the cases stored in the storage unit and supplied to the manipulated variable correction calculation unit 16. Operation amount correction calculator 16
Calculates the operation amount for realizing the control amount target value set by the user or the like with reference to the correction rule in the correction rule creation unit 12. And the obtained operation amount is
It is output to the controlled system 1 via the manipulated variable set value output unit 17.

The controlled system 1 outputs time information as a state quantity, which is supplied to the correction rule creation unit 12 and the storage control unit 11 via the state quantity recognition unit 19. In the case of this embodiment, the time information is used as the case collection time, as described later.
By the way, when the case collection time is adopted as the state quantity, time information may be generated inside the system control device 2 and the value may be used.

B: Operation of First Embodiment (1) Initial Setting Operation (Start-Up Operation) Next, the operation of this embodiment having the above-described configuration will be described. To operate this embodiment, a technician must:
Perform the startup work described below. First, a technician manually sets an arbitrary operation amount, operates the controlled system 1 under the setting, and stores the operation amount, control amount, and state amount (that is, sampling time) at that time in the system controller. 2 is stored in the case storage unit 15. Next, the technician repeats the above operation n + 1 times or more while arbitrarily changing the operation amount within an allowable range, and stores n + 1 or more control cases in the case storage unit 15 in the system control device 2. .

Here, n is the number of types of manipulated variables.
Therefore, in the case of the present embodiment in which only the voltage set value is used as the manipulated variable, n = 1, so that the case where the voltage set value is set to a different value may be collected twice or more. Further, in the case of using two of the voltage set value and the pressure set value as the manipulated variables, since n = 2, the technician sets the power set value and the pressure set value to different values, and The work may be repeated three or more times.

Table 1 shows a state in which three cases are taken when the operation amount is only the voltage set value. here,
Time information is shown as a state quantity. For example, “40525093015” in case 1 is 1
9:30:15 on May 25, 994.

[0055]

[Table 1]

(2) Operation During Operation Next, the operation after the start-up work is completed will be described. First, when the system is turned on and the control amount target output is instructed, the system control device 2 calls the latest control case, that is, the control case immediately before the last time the switch is turned off, and based on this control case, All cases with similar state quantities are searched. In this case, since it is immediately after the start-up, “case 3” in Table 1 corresponding to the last control case of the start-up work corresponds to this case. Therefore, the correction rule creating unit 12 reads “case 3” from the case storage unit 15.

In this embodiment, the case collection time is ± 1.
Those within 0 minutes are considered to be control cases collected under similar conditions, and as a result, case 1 in Table 1
Case 2 is determined to be similar to case 3.

Then, the correction rule creating unit 12 plots similar control cases in an n + 1-dimensional space composed of the operation amount and the control amount. That is, in this embodiment, since the operation amount is only the voltage set value (n = 1), the operation amount and the control amount of each control case are plotted on the two-dimensional plane.

When there are m types of control amounts, m
Are plotted in the n + 1-dimensional space, but may be plotted in the same space as long as the control amount can be represented by the same coordinate axis. Then, the relationship between the control amount and the operation amount in the n + 1-dimensional space is derived as m control rules.

In the case of this embodiment, since n = 1 and m = 1, the relationship between the output score plotted on the two-dimensional plane and the set voltage value is grasped by one control rule. . More specifically, in this embodiment, this control rule is calculated as a first-order approximation line by the least squares method. This is shown in FIG. That is, the correction rule creation unit 12 obtains coefficients a and b that can approximate output score = a × voltage set value + b.

As described above, when the control rule that captures the relationship between the control amount and the operation amount is extracted, the operation amount correction for realizing the specified output target value can be easily calculated.
That is, the operation amount correction calculation unit 16 can easily calculate the operation amount based on the control rule (the straight line shown in FIG. 2) created by the correction rule creation unit 12. For example, when the output target value is set to the output score = 100, it can be inferred from FIG. 2 that the operation amount should be set to 66V.

Next, the manipulated variable correction calculator 16 transfers the calculated manipulated variable to the manipulated variable set value output unit 17 and sets it as a new manipulated value. The controlled system 1 is controlled by the manipulated variable set value thus obtained.

(3) Correction of Control Rule and Creation of Cluster Next, after performing the above-described control operation, the system control device 2 verifies whether the actually obtained system output is equal to the target value. That is, in the case of the above example, it is verified whether or not the output score is 100. here,
In this embodiment, the allowable output error is set to ± 2 or less. Therefore, in the above example, if the output score is 98 or more and 102 or less, it means that the control can be executed with high accuracy. If the control can be executed with high accuracy, it can be interpreted that the control rule used for the current control is appropriate, and the information included in the current control content is the information already obtained (in this case, cases 1 to 3). ), It is determined that there is no need to additionally store.

In other words, if the current control content is plotted on FIG. 2 as a new control case, the control rules (more specifically, the coefficients a and b Is substantially unchanged and does not need to be modified. For such a reason, when the system output achieves the target value within the allowable error, the control is shifted to the next control without performing any special processing operation.

On the other hand, if the system output exceeds the allowable error range and deviates from the target value, it can be interpreted that the control rule used for inferring the manipulated variable was not appropriate. Therefore, the contents of this control are worth memorizing as a new control case that is more in line with the current system situation. In other words, it is necessary to modify the control rule used in the current control or create a new control rule using the current control content as a new control case. Whether the control case is newly stored is determined by the storage control unit 11.

In this embodiment, when the system output exceeds the allowable error range and deviates from the target value, the state quantity becomes
It is similar to the case where the state quantity of the control case used for control rule extraction (control case X is described for the sake of explanation. In the above example, cases 1 to 3 correspond to control case group X). In cases where there is no control rule, the control rule is modified and newly created.

In the case of correcting the control rule In this embodiment, if the time at which the current control operation is performed is within 10 minutes from the collection time of the control case group X, the current control case is replaced with the control case group X. And is used to modify the control rules used for the next control. Since this is a case in a similar state quantity, by increasing the number of cases, the accuracy of data is statistically increased, and correction is performed so that the control rule can be applied better. More specifically, the correction rule creating unit 12 recalculates the coefficients a and b using four cases obtained by adding the present control case to the cases 1 to 3.

When creating a cluster On the other hand, if the new control case differs from the collection time by more than 10 minutes, the state in which the control system is placed has changed. The derived control rules can be interpreted as no longer applicable. In this case, it is necessary to re-collect a control case under the new state quantity and extract a new control rule.

That is, a new rule must be created instead of modifying the control rule. Therefore, in such a case, n + 1 or more control cases are additionally stored through n + 1 or more control operations, starting from the time when the system output exceeds the allowable error range and deviates from the target value. Here, the reason why the number is set to n + 1 or more is the same as at the time of startup. In addition, regarding the transient control while the (n + 1) or more control cases are additionally stored, the above-described control rule is corrected, and the control error is reduced.

The above control operation is repeatedly executed.
When new necessary control cases are additionally stored, control cases such as those shown in Table 2 are accumulated in the case storage unit. The state as shown in Table 2 is stored in the normal case storage unit 15 of the system control device according to the present invention.
It is a state of.

[0071]

[Table 2]

Here, in this embodiment, clusters are created by grouping control cases having similar state quantities. Table 3 shows this state. However, Table 3 is for convenience of explanation of the operation. In practice, when a cluster is completed, individual control cases included in the cluster are deleted from the case storage unit 15. This is because information contained in each control case is all included in the control rules (coefficients a and b in this embodiment) extracted from the cluster, so that there is no need to store and save. This is because the necessary capacity is not unnecessarily increased. Therefore,
The case storage unit 15 stores only recently collected control cases (case 3 in the illustrated example) that have not yet been completed as a cluster. On the other hand, the control rule corresponding to the completed cluster is stored in the correction rule creating unit 12.
Is stored within. That is, as shown in Table 4, examples and control rules (control rules corresponding to completed clusters)
Are stored separately.

[0073]

[Table 3]

(4) Control by degree of conformity Next, under the conditions as shown in Table 4, that is,
A control operation of the present system control device during a general operation (when a large number of clusters are formed), not at the time of startup or immediately after startup, will be described. Now, assuming that the user presses the start button in order to obtain the output of the controlled system 1, the correction rule creating unit 12 in the system control device 2 temporarily stores the immediately preceding operation amount set value and the By looking at the actual system output at that time, the degree of fitness with each cluster is calculated. Here, the degree of conformity is obtained by calculating the difference between each control amount inference value obtained by applying the immediately preceding operation amount setting value to the control rule of each cluster and the actual system output, and normalizing the reciprocal thereof (total 1).

[0075]

[Table 4]

In this embodiment, only five clusters are used as similar clusters in ascending order of the difference between the inferred control value and the actual system output. When the number of clusters existing in the storage unit is five or less, the same operation is performed using only the existing clusters. For this reason, the same operation method can be used even when only one cluster exists, such as immediately after startup.

The fitness obtained in this way is referred to as “weight”.
, Weighted and averaged the control rules of each similar cluster, and combine these to create a new control rule. For example, the coefficient a is weighted and averaged to obtain a ', and the coefficient b' is determined so that an approximate straight line based on the coefficient a 'intersects the actual system output. That is, a control rule of control amount = a ′ × operation amount set value + b ′ is created.

Next, the manipulated variable correction calculation unit 16 obtains a manipulated variable set value such that the controlled variable becomes the output target value from the control rules obtained as described above, and the manipulated variable set value storage unit 17 , The control operation is executed by setting the controlled system 1 via the. Thereafter, as in the case after the above-described start-up, a new control result (system output) is compared with an allowable error to determine whether the current control content should be additionally stored. If the rule synthesized using the degree of conformity cannot be suppressed within the allowable error, the control rule is modified or newly created as needed, and further preparations are made for the next control.

C: Effect of Embodiment In this embodiment, the following operation and effect can be obtained. (1) First, since the operation can be performed only by the start-up work as described above, prior data collection and optimization work which are indispensable in the conventional apparatus of this type are not required at all. That is, in the past, in order to optimize the system in consideration of the environment and changes over time, it was necessary to collect a large amount of data by setting the environment in various ways and to perform a long-term running test. In addition, costs such as equipment costs and labor costs were very high, such as the necessity of an environmental laboratory for that purpose. However, in this embodiment, there is a great effect that the cost can be substantially reduced to almost zero. In addition, since there is no need for the conventional work requiring the specialized knowledge of optimization work, there is an effect that it is easy to secure development engineers.

The start-up work only needs to be performed once in order to make the system control device according to the present invention operable, and the quality of this work depends on whether the system control device in actual operation is good or bad. Since it has nothing to do with performance, it does not require the skill of a technician. That is, anyone can start up in a single operation and maximize the performance of the system control device.

Further, the system control device according to the present invention can determine whether the executed control result is good or bad by the control device itself, and can additionally store necessary control cases by a method.
There is an effect that the control performance can be automatically and independently improved. In other words, in the conventional fuzzy control method and neural network method, if the fuzzy rules and teacher signals to be learned in advance are not optimal, the performance in actual operation remains insufficient. The technicians depended heavily on their abilities, such as developing them over a long period of time by trial and error, and requiring more preliminary research to determine appropriate teacher signals. According to this, the effect is obtained that anyone can develop the best control device.

Further, according to the present embodiment, the learning only needs to store the control case in the memory, and the control rule can be executed instantaneously by a simple arithmetic calculation based on the control case.
A great effect is obtained that there is no need for a trial and error tuning by an engineer such as the conventional fuzzy control or a long learning operation such as a neural network. That is, according to the present embodiment, a significant effect that fully automatic real-time learning and tuning, which cannot be performed by the related art, can be performed is obtained.

(2) Next, according to the present embodiment, since the control operation can be performed based only on past control cases, there is no need to model the controlled system. That is, in the present invention, even if the controlled object is a black box, it can be controlled as it is. Therefore, even if the controller development engineer does not have a deep understanding of the detailed physical mechanisms of the controlled system,
A control device with sufficient performance can be developed, and the number of development steps can be reduced.

(3) Since the manipulated variable set value can be used as it is as an element of the control case, it is not necessary to sense a physical quantity which is an element of the control case. For this reason, sensor cost can be reduced. By the way, in the past, since a suitable sensor was not available (because the cost was too high or there was no one with sufficient performance), it could not be realized even with the most favorable control method in terms of performance. However, in the present embodiment, it is possible to freely select a method without restriction from the sensor, and a great effect of improving the performance by selecting the optimum method can be obtained.

(4) Next, according to the present embodiment, even if there is no case sufficiently similar to the situation at the time of operation in past cases, a plurality of past control rules are determined using the degree of conformity. Since they can be used in appropriate combination, there is an effect that accurate system control can be executed. Moreover, according to the present embodiment, if the desired control accuracy cannot be achieved by such a method, the system controller automatically stores additional control cases and independently improves control performance. can do.

[0086] With this configuration, the system can be operated optimally even when the user operates the system under a situation that was not assumed by the engineer at the time of development. Excellent effects can be obtained as long as the device itself can learn. For example, in the past, environmental experiments and the like were previously performed as described above, but it is impossible to collect data under all conditions in reality due to time and manpower constraints or experimental equipment, It is just a selection of some typical conditions. That is, if it is temperature and humidity, there are two points of high temperature and high humidity and low temperature and low humidity environment. However, if the optimization work is performed based only on these two points of data, if the user operates the system in another environment, sufficient control accuracy cannot be obtained. In addition, in cases where optimization was made for temperature and humidity, but the effects of changes in air pressure were not taken into account at the time of development, performance would be reduced if used in places with different air pressures, such as at high altitudes or in airplanes. Would.

However, according to the present embodiment, even in such a case, since the apparatus itself learns to obtain sufficient control accuracy in its use environment, the controlled system is always the highest. Can demonstrate performance. Further, according to the present embodiment, in the same manner as described above, even when the controlled system deteriorates with time, the control can be performed while always learning the influence of the deterioration, so that the best control is always possible. . Further, even when a worn part is replaced with a new one, the adjustment can be performed by the automatic learning according to the present invention without the need for a user or a serviceman to perform readjustment or the like manually.

(5) According to the present embodiment, since only the control case with the lowest importance can be properly selected and deleted from the storage unit, the newly required control case or cluster becomes insufficient in memory. There is no problem that the memory cannot be stored. Since the memory capacity can be designed to be a necessary minimum, the cost can be reduced, and a memory having a limited capacity can be used most effectively.

In particular, as in the present embodiment, when time is adopted as a state quantity, and when a cluster is completed, a case or a group of cases, which is a component of the cluster, is collectively deleted from the storage unit. The memory capacity for case storage can be minimized.

D: Modifications (1) The method of deriving the control rule may be other than the first-order approximation by the least squares method. That is, any relationship may be used as long as the relationship between the m types of control amounts and the n types of operation amounts in the n + 1-dimensional space can be regarded as m quantitative relationships. Therefore, a higher-order approximation calculation may be performed, or a method of causing a neural network or the like to learn may be used. However,
Since learning of the neural network takes a certain amount of time, it may not be suitable for a system control device that requires a fast response speed.

It is needless to say that when m kinds of independent control amounts are simply controlled using n kinds of independent operation amounts, n = m. If there is a relational expression defined separately, or if m types of control amounts are not independent, n ≠ m.

(2) In the case of a system control device in which the manipulated variables are of two types, a voltage set value and a pressure set value, and the controlled variable is composed of output scores A and B,
Each case is plotted in a three-dimensional space as shown in FIG. In this space, two planes, a plane A indicating the relationship between the voltage set value and the pressure set value for the output score A and a plane B indicating the relationship between the voltage set value and the pressure set value for the output score B, are Calculated by the least squares method. That is, these two planes A and B represent the control rules. If the output scores A and B are physical quantities having different dimensions, it goes without saying that two n + 1 dimensional spaces having different vertical axes are used.

In the case of n = 2 and m = 2 shown in FIG. 3, when obtaining the operation amount, as shown in FIG.
First, a straight line at which the score A intersects with the score A target plane and a straight line at which score B intersects with the score B target plane are determined. These straight lines serve as target achievement lines for score A and score B. Then, in order to simultaneously set both the score A and the score B to the respective target values, the respective target realization lines are projected onto a plane formed by the voltage set value axis and the pressure set value axis, and the voltage set at the intersection is set. Values and pressure settings may be used. The voltage value and the pressure value thus obtained are the corrected operation amounts.

Here, a description will be given of an example of calculating the degree of conformity when a control rule for a plane is created in a three-dimensional space with reference to FIG. FIG. 5 shows a case where the planes of the cluster A and the cluster B are formed for the score A, and the newly plotted point B5 is not located on any plane. At this time, the distance between the point indicating the current control content in the coordinate space, that is, the point B5 and each case plane is calculated. Then, the reciprocal thereof is obtained and normalized. That is, the sum of the reciprocals of the distance from each control case plane is set to 1. The value thus standardized has the same meaning as the fitness (fitness in the case of a straight line) in the above-described first embodiment, and the gradient in each coordinate axis direction of each case plane is weighted by this fitness and summed. I do. Then, the total amount is defined as the inclination in the direction of each coordinate axis of the new control case plane that can be adapted to the current situation, and is further adjusted to a height (intercept of the control amount axis) that includes the current control content on the plane. , Create compound rules.

(3) In a system in which the state quantity does not change and its detection and storage are unnecessary, control can be performed using only the operation amount and the control amount. In this case, there is an advantage that the device configuration is simplified.

(4) In the above-described embodiment, control is performed by obtaining the degree of fitness of a cluster. However, a cluster having the closest state quantity is determined, and control is performed using a control rule extracted from this cluster. Is also good.

(5) The definition of the degree of conformity used in this embodiment is as follows:
This is just an example, and any definition method may be used as long as the method can objectively and uniquely indicate the degree of matching for each cluster.

(6) In the present embodiment, the permissible error of the control amount is determined, and control is performed according to the result. However, when the change range of the control amount is known in advance and the range is narrow, Alternatively, the determination of the allowable error may be omitted and all the control cases may be stored.

(7) Instead of or in addition to the memory management in this embodiment, the following memory management may be performed. When the storage capacity is insufficient in the case storage unit 15, the oldest control case is deleted. When the storage capacity becomes insufficient, the case storage unit 15 deletes the oldest cluster. The case storage unit 15 stores the occurrence time and the number of times used for determining the operation amount as elements of the control case. When the storage capacity becomes insufficient, the case storage unit 15 stores the storage time before a predetermined time.
In addition, the control case with the least number of uses is deleted. The case storage 15 stores creation time,
And the number of times used for determining the operation amount is stored, and when a storage capacity shortage occurs, it is stored before a predetermined time, and
Erase the least used cluster. The case storage unit 15 stores, as elements of the cluster, the creation time and the cumulative fitness when used for determining the operation amount, and when the storage capacity is insufficient, the storage is performed before a predetermined time, and , The cluster with the smallest cumulative fitness is eliminated.

(Second Embodiment) A: Configuration of Second Embodiment Next, a second embodiment of the present invention will be described with reference to FIGS. Here, FIG. 6 is a schematic configuration diagram illustrating a configuration of a robot arm according to a second embodiment, and FIG. 7 is a block diagram illustrating a configuration of a control unit according to the second embodiment. FIG.
In FIG. 7, the same reference numerals are given to the portions corresponding to the respective portions in FIG.

This embodiment is an example in which the present invention is used to control a motor whose load changes arbitrarily. More specifically, various works W are quickly conveyed and placed precisely at a desired place. 4 is an example of drive control of a robot arm to be performed.
Further, in this example, it is required that the robot arm be moved at a speed as high as possible irrespective of the weight of the article, and that electric energy must be supplied to the motor such that the robot arm can be stopped at a predetermined position accurately within an allowable error. Is imposed.

Now, referring to FIG.
A hand 51 for gripping the work W is provided at the end of the work W. Further, a weight sensor 52 for detecting the weight of the work W and a stop position sensor 53 are provided. The robot arm 50 is driven by a robot drive unit 60 having a motor and a brake mechanism, and the robot drive unit 60 is controlled by a control device 61. In this case, the robot arm 50 grips the work W conveyed on the belt conveyor 55 by the hand 51 and places it on the target stop position P1 or P2 set on the turntable 65. The robot driving unit 6
The brake mechanism within 0 always performs a constant deceleration operation.

Next, as shown in FIG. 7, the control device 61 has the same configuration as the system control device 2 of the first embodiment (see FIG. 1). The controlled system in this case is
The robot system includes a robot driving unit 60, a robot arm 50, and the like. The output signal of the stop position sensor 53 is a control amount, and the output signal of the weight sensor 52 is a state amount.

Here, the error amount of the actual stop position with respect to the target stop position is (-) for an excessively short position and (+) for an excessively large amount.
And the amount is evaluated in millimeters. Further, an error within ± 3 mm is set as an allowable error. By the way, in order to move the robot arm 50 at a high speed, it is sufficient to supply a larger energy to a heavier article, but since the robot arm having a heavier article has a large inertia, it stops instantaneously and accurately at a target stop position. It becomes difficult to do. Therefore, it is important to determine the optimal amount of electric energy supplied to the motor in order to achieve both the transfer speed and the stop position accuracy.

Further, the relationship between the conveying speed and the stop position accuracy with respect to the weight of the article is nonlinear and complicated, and in order to perform optimal control in the prior art, an experiment is performed using various articles in advance, and the feedback gain is determined. The technician had to prepare in advance through many development steps such as optimization of the feedback and switching of the feedback gain according to the weight of the article. In the present embodiment, such advance preparation is not required, and optimization control can be easily performed.

B: Operation of Second Embodiment (1) Initial Setting Operation Next, the operation of this embodiment having the above configuration will be described.
First, the initial setting operation will be described. In this embodiment, as in the case of the first embodiment, a start-up operation as described below is required. The start-up work needs to be performed only once in order to bring the present embodiment into operation, and the quality of the work has nothing to do with the performance in actual operation.

First, a technician must select a work W having an appropriate weight.
Is manually held by the robot hand 51, an appropriate motor supply current value (operating amount) is manually set, and the robot control unit 60 is operated under the setting. Thereby, the robot arm 50 turns, and the work W is placed on the target position P1 or P2. The stop position deviation amount (control amount) at this time is detected by the stop position sensor 53,
It is supplied to the control device 61. The weight (state quantity) of the work W is detected by the weight sensor 52 and supplied to the control device 61. Then, the first case (the combination of the operation amount, the control amount, and the state amount), which is the result of the above processing, is stored in the case storage unit 15.

Next, the engineer repeats the same processing as described above n + 1 times or more while changing the motor supply current value, which is the operation amount, within an allowable range.
Is stored in n + 1 or more cases. At this time, the weight of the work W, which is the state quantity, may be kept constant.

Here, n is the type of the manipulated variable. In this embodiment, n = 1 (only the current value supplied to the motor). Therefore, the engineer changes the current value supplied to the motor to 2
More than one case should be collected.

The control cases at the time of startup obtained in this way are stored in the case storage unit 15 in a form as shown in Table 5, for example. Then, the case storage unit 15 compares the state quantities of the stored control cases, and forms a cluster in which control cases having similar state quantities are put together.

[0111]

[Table 5]

In this embodiment, the difference in weight is ± 10
The cases up to% are regarded as cases of state quantities similar to each other. This is because it is known that the correspondence between the motor supply current value and the stop position accuracy is substantially constant if the change in weight is up to ± 10%. Therefore, the cases at startup shown in Table 5 are classified into the same cluster.
The average value of the state quantities of the control cases in the cluster is set as the state quantity of the cluster.

Next, all the control cases belonging to the cluster obtained as described above are plotted in an n + 1-dimensional space composed of the operation amount and the control amount. In this embodiment, the operation amount is only the current set value supplied to the motor (that is, n = 1).
Therefore, each control case is plotted on a two-dimensional plane.

Then, the correction rule creating section 12 sets n + 1
The relationship between the control amount and the operation amount in the dimensional space is extracted as a control rule. More specifically, in this embodiment, this control rule is calculated as a first-order approximation line by the least squares method. FIG. 8 shows the control rules thus obtained. As can be easily understood from FIG. 8, the coefficients a and b which can approximate the displacement amount = a × the motor current set value + b are obtained. Here, the derivation method of the control rule generally describes the relationship between m types of control amounts and n types of operation amounts in the n + 1-dimensional space,
Any type may be used as long as it can be regarded as m quantitative relationships. For these, the first
This is as described in the embodiment. With the above processing, the start-up work is completed.

(2) Operation at the time of actual operation Control based on the fitness of the cluster Next, the operation at the time of actual operation of the robot after the start-up work is completed will be described. First, when the robot starts operation and grips the work W, a weight signal indicating the weight of the work W is output from the weight sensor 52. The correction rule creator 12 compares the supplied weight signal with the state quantity of each cluster to determine the degree of conformity.

That is, the difference between the work W gripped by the robot arm 50 and the state quantity (average weight) of each cluster is calculated, and the reciprocal thereof normalized for all clusters is calculated as the degree of conformity. In this case, since there is only a cluster consisting of two control cases at the time of startup, this cluster has a fitness of 100%. Then, when the fitness of each cluster is obtained, a weighted average according to the fitness of the control rules extracted in each cluster is made, and a combined control rule is calculated. Then, a control operation is performed using the synthesis control rule.

In this case, since the degree of conformity of the cluster at the time of startup is 100%, the control rules extracted from the cluster at the time of startup are directly provided to the control operation as a combined control rule. Next, in the manipulated variable correction calculation unit 16, a motor current set value such that the displacement amount becomes zero is obtained from the synthesis control rule. Then, a current value corresponding to the motor current set value is set to the arm rotation motor 6.
0a, and the first control operation is performed.

Next, as a result of the work transfer operation performed as described above, it is determined whether or not the work W is accurately placed on the target position P1 or P2. This determination is made by the comparing unit 10 based on the output signal of the stop position sensor 53. If the displacement amount is ± 3 mm or more, the current set of the weight (state amount) of the workpiece W, the motor current set value (operation amount), and the displacement amount (control amount) is set as a new control example. It is stored in the case storage unit 15.

On the other hand, if the displacement is less than ± 3 mm, the case obtained by the current control is extracted from the already obtained case (that is, extracted from the cluster consisting of cases 1 and 2 at the time of startup). Control rule), it can be determined that additional storage is not necessary. Therefore, the control is shifted to the next control without performing any special processing.

Modification of Control Rule and Generation of New Cluster In this embodiment, a newly added and stored control case is not similar to the case where the state quantity is similar to the state quantity of the existing cluster. The processing is divided into cases. That is, if the state quantity of the newly stored control case (the weight of the work conveyed this time) is within ± 10% of the state quantity of the existing cluster, the cluster is additionally classified into the cluster, and the cluster is additionally classified. Used to modify the control rules extracted in. This is because the accuracy of the data is statistically increased by increasing the number of cases, and the data is modified so that the control rules can be applied more effectively.

On the other hand, if the difference is ± 10% or more, the control case is not classified into an existing cluster, but a new cluster (for example, cluster B) is created and classified there. Then, when a control case classified into the cluster B occurs after the subsequent control operation, a control rule in the cluster B is extracted and provided for the subsequent control.

When the above-described control is repeatedly executed and new control cases are additionally stored, control cases such as those shown in Table 6 are accumulated in the control case storage unit 15. The state shown in Table 6 is a general state of the case storage unit 15 in the present embodiment.

[0123]

[Table 6]

Control in General Operating State Next, under the situation as shown in Table 6, ie,
A description will be given of a control operation in a general operation state not immediately after startup but after a certain degree of control has been executed.
In such a situation, when the robot arm 50 grips the workpiece W, the weight is detected by the weight sensor 52, and is compared with the state quantity of each cluster, and the fitness is calculated.

When the degree of conformity of each cluster is obtained, a weighted average corresponding to the degree of conformity is made to the control rules extracted in each cluster, and a composite control rule is calculated.
Then, based on the synthesis control rule, a motor current value such that the displacement amount becomes zero is calculated, and the arm rotation motor 60a is controlled. If the amount of displacement is equal to or more than the allowable error (± 3 mm), the current control case is stored in the case storage unit 15. If positioning is performed accurately, the case is not stored and the next control case is not performed. Move to

The above is the control operation in a general operation condition. The operation procedure itself is exactly the same as the control operation immediately after starting. However, the only difference is that since there are many clusters, many control rules are extracted from the clusters. As is evident from the above, in the present embodiment, the processing contents are completely the same even immediately after startup or when control is performed for a long period of time, and there is no need to identify both. Therefore, immediately after the start-up, it can be used in the same manner for a long period of time.

C: Effects of the Embodiment In this embodiment, the following effects can be obtained. (1) Conventionally, this type of equipment eliminates the need for prior data collection and optimization work that are indispensable at all.
Development man-hours can be greatly reduced. In addition, specialized knowledge regarding the mechanism and control technology of the robot is not required, and skill of a development engineer is not required.

(2) Since the control device itself determines the quality of the control result (stop position accuracy) and decides whether or not to add a control case, the control performance can be automatically and independently improved. This greatly surpasses fuzzy control, the neural network system, and the like, as described in the first embodiment.

(3) Since the set value can be used as it is as the operation amount which is an element of the control case, it is not necessary to sense the motor supply current as in the related art. For this reason, the cost of a sensor such as an ammeter can be reduced, and at the same time, a decrease in control performance due to a sensing error can be avoided.

(4) Even in a situation that was not assumed by the engineer at the time of development, that is, in the case of transporting a work that is heavier or lighter than expected, the control device itself autonomously learns and performs the best control. Performance is improved so that Similarly, even when the controlled system deteriorates with time, the control is always performed while learning the influence of the deterioration, so that the best control is always possible. For example, even if the braking performance deteriorates with time, it is possible to avoid a situation in which the stopping accuracy is reduced.

(5) Even if a worn part is replaced with a new one, automatic adjustment is performed by the automatic learning function according to the present invention without the need for a user or serviceman to manually perform readjustment or the like.

D: Modifications Various modifications are also possible in the second embodiment. However, since most of the modifications described in the first embodiment can be applied, description thereof will be omitted to avoid duplication.

(Third Embodiment) A: Background of Third Embodiment First, the background of the third embodiment will be described. The first mentioned
In the embodiment, the case collection time is adopted as the state quantity. This is because the factors that affect the state of the controlled system 1 are:
This is because it can be regarded as almost constant in a limited limited time range, although it is diversified such as various environmental conditions and changes with time. For example, it can be considered that there is substantially no change in state 10 minutes before and 10 minutes later, but the temperature and humidity may be significantly different between morning and evening, yesterday and today, and the like.

Therefore, when the system control device 2 or the controlled system 1 is temporarily stopped and then restarted, the controlled system 1 is stopped during the stop period, that is, while the control cases including the state quantity are not collected. It is also assumed that the physical state is significantly different.

In general, this is often found in various control systems. For example, even if the external environment is constant, the controlled system, which has been operated once and the temperature has increased, is controlled under the increased temperature. And
There are cases where the state of the controlled system 1 is completely different.

In such a case, in the first embodiment,
When the controlled system 1 is activated again and the target output of the system 1 is instructed, the system control device 2 calls the control case immediately before, that is, immediately before the switch is turned off, and sets the operation amount in the control case. The degree of conformity of each control rule is obtained from the value, and each control rule is weighted and averaged by the obtained degree of conformity to create (synthesize) a new control rule.

However, if the state immediately before the switch is turned off is greatly different from the current state, the control amount obtained for the previous operation amount is far from the control amount target value. The fitness is calculated and a new control rule is created. Of course, if the resulting control error exceeds the allowable value, by adding a control case as described above,
Although the control accuracy after the next time is improved, if an initial error is large, it takes a considerable time until an acceptable accuracy is obtained.

Therefore, in the third embodiment, in order to obtain the first control case at the time of restart, an average value of the operation amount set in the control case immediately before the switch is turned off and a predetermined standard operation amount is calculated. Then, this is set as an initial operation amount. As a result, even if the state immediately before the switch is turned off is greatly different from the state at the time of restarting, it is possible to prevent the initial error from becoming extremely large.

B: Configuration of the Third Embodiment The third embodiment will be described below with reference to FIGS. In this embodiment, the present invention is applied to a laser printer in which the above-described state change at the time of restart is expected to occur. In FIG. 9, reference numeral 100 denotes an image output unit of the laser printer, which is a controlled system in the present embodiment. A control unit 120 controls the laser output of the image output unit 100 so that the development density matches the target density.

Reference numeral 121 denotes a density adjustment dial.
The operator sets a value according to the desired density. The set value of the density adjustment dial 121 is converted into an output of the development density sensor 110 by the converter 122 (for example, “0”).
To “255”). The target density output from the converter 122 is stored in the control amount memory 123. In this case, the control amount memory 123 also stores an allowable error.

On the other hand, the output signal of the developing density sensor 110 and the output signal of the memory 123 are compared with a density comparator 124.
Are compared. In this comparison, the allowable error stored in the memory 123 is referred to. The output signal of the development density sensor 110 is supplied to the control rule search unit 130 if the difference between the two is within the allowable value, and is supplied to the control case memory 125 if the difference is equal to or larger than the allowable value. However, when the system is restarted, which will be described later, the degree of conformity of each control rule (see the first embodiment) is calculated. It is supplied to the search unit 130.

The control case memory 125 is a memory for storing control cases, and stores a set of three kinds of amounts, ie, a state amount, an operation amount, and a control amount. The reason why control cases are stored in this way is to perform various controls based on control cases stored in the past, as described in the first embodiment.

Here, the state quantities stored in the control case memory 125 are the values of temperature and humidity that have a dominant effect on the electrophotographic process, and the manipulated variables are the laser power for changing the development density of the laser printer. Set value (hereinafter, LP
The control amount is referred to as a set value.
0 output signal.

The state quantity comparator 126, the cluster memory 127, and the control rule calculator 128 extract the control rule with reference to the control case stored in the control case memory 125.

The control rule memory 129 is a memory for storing a plurality of control rules calculated by the control rule calculator 128. When a request is received from the control rule searcher 130, the control rule corresponding to the request is returned. . In this case, the control rule retrieving unit 130 controls the concentration comparator 12
The control rule memory 129 requests a control rule corresponding to the density difference supplied from the control amount 4 and the operation amount (ie, the LP set value) supplied from the operation amount memory 132.

The control rule search unit 130 applies the operation amount supplied from the operation amount memory 132 to each control rule stored in the control rule memory 129 when the system is restarted, which will be described later. Sensor 11
The degree of conformity of each control rule to the detection result of 0 is calculated. Then, the control rule search unit 130 creates (combines) a new control rule by weighting and averaging each control rule using the calculated degree of conformity as “weight”.

The manipulated variable correction value calculator 131 obtains a correction value of the manipulated variable using the control rule retrieved or created by the control rule searcher 130, and stores the obtained correction value in the manipulated variable memory 132. Supply. Thereby, the operation amount memory 132 supplies the LP setting value corresponding to the operation amount correction value to the light amount controller 116.

The standard operation amount memory 133 stores the operation amount (hereinafter, referred to as “operation amount”) to be set when the laser printer is in a so-called standard state (for example, a state of medium temperature and medium humidity which is neither high temperature and high humidity nor low temperature and low humidity). (Referred to as a standard operation amount). The standard operation amount stored in the memory 133 is output to the control rule search unit 130 when the laser printer is restarted (for example, once the power switch is turned off and the power is turned on again). . The operations of the control rule search unit 130 and the operation amount correction value calculator 131 in this case will be described later. The above-mentioned control case memory 125, control rule memory 129, and standard operation amount memory 133 are all configured by non-volatile memory elements.

The reference patch signal generator 142 outputs a calibration reference patch signal to the image output unit 100. Here, the calibration reference patch signal is the development density sensor 11
For the purpose of detecting the development density based on 0, the input image is output as a dummy to generate a patch in a predetermined area of the photoreceptor where the input image is not exposed. Reference patch signal generator 1
The operation timing of 42 is synchronized with the control unit 120 by the synchronization circuit 141. Further, the timer 140 supplies a time clock signal to the synchronization circuit 141 and the standard operation amount memory 133 described above.

C: Operation of Third Embodiment Next, the operation of the third embodiment will be described. In the following, the power supply (the temperature and the humidity) of the laser printer is assumed to greatly change. The operation at the time of restart after the switch is turned off will be described as an example.

First, after the power switch is once turned off and the control unit 120 is restarted by turning on the power switch again, the control rule search unit 130 reads the standard operation amount from the standard operation amount memory 133. Read and capture this. In addition, the control rule search unit 130 reads the operation amount immediately before the switch is turned off from the control case memory 133 from the control case memory 125, and supplies this and the standard operation amount to the operation amount correction value calculator 131. The manipulated variable correction value calculator 131 calculates the average value of the manipulated variable supplied from the control rule searcher 130 immediately before the switch is turned off and the standard manipulated variable, and outputs the result to the manipulated variable memory 132. Accordingly, the operation amount memory 132 supplies the LP setting value corresponding to the average value to the light amount controller 116.

On the other hand, a reference patch signal for calibration is output from the reference patch signal generator 142 in synchronization with the operation of the control section 120, whereby a reference patch is generated on the photosensitive member. The development density at this time is determined by the sensor 1
10 and the output signal is supplied to the control rule searcher 130. Then, the control rule search unit 13
A value of 0 applies the operation amount stored in the operation amount memory 132 to each control rule, calculates the degree of conformity of each control rule to the detected development density, and thereby creates (combines) a new control rule.

Thereafter, by comparing the new control result (that is, the density difference) with the allowable error, it is determined whether or not the current control content should be additionally stored. It is created and prepared for the next control.

Here, referring to FIGS. 10 to 13, the case where the previous operation amount set value (that is, before stop) is used as the operation amount after the restart as it is and the case where the previous operation amount set value is used as in this embodiment. The relationship between the operation amount (LP set value) and the control amount (image density) before and after the restart when the average value of the reference operation amount and the standard operation amount is the operation amount after the restart is compared. FIG. 10 shows the former case, and FIGS. 11 to 13 show the latter case.

As shown in FIG. 10, when the manipulated variable set value “105” corresponding to the previous high temperature and high humidity is used as the manipulated variable after the restart, if the state after the restart is also a high temperature and high humidity, the concentration is changed. An error hardly occurs, but when the state after the restart is low temperature and low humidity, the error becomes extremely large (maximum density difference “0.8”).

On the other hand, as shown in FIG. 11, the manipulated variable set value “105” corresponding to the previous high temperature and high humidity and the standard manipulated variable “1”
When the average value “105” with “25” is set as the manipulated variable after the restart, even if the state after the restart is low temperature and low humidity, the error is not as large as in the former case (maximum density difference “0.6”). .
As shown in FIG. 12, when the average value “125” of the operation amount set value “125” corresponding to the previous medium temperature and humidity and the standard operation amount “125” is the operation amount after the restart, the maximum The density difference is “0.4”, and as shown in FIG. 13, the average value “135” of the previous operation amount set value “145” and the standard operation amount “125” corresponding to the low temperature and low humidity is obtained after the restart. When the operation amount is used, the maximum density difference is “0.6”, and the error is not as large as in the former case.

As described above, according to the present embodiment, in order to obtain the first control case at the time of restart, the average value of the operation amount set in the control case immediately before the switch is turned off and the standard operation amount is calculated. Since it is calculated and set as the initial operation amount after the restart, even if the state immediately before the switch is turned off and the state at the time of the restart greatly differ, the initial error does not significantly increase, Control rules can be selected and synthesized with an appropriate degree of conformity, and high-precision control can be performed quickly from the beginning of the restart.

D: Modification Example (1) As another mode, an average value of the manipulated variable set values in past control cases may be used instead of the standard manipulated variable set value. This eliminates the need to determine the standard manipulated variable in advance through a preliminary experiment or the like. Also, if the user's usage environment is not a standard environment for general users but is user specific,
The operation amount can be set in consideration of the usage environment unique to the user.

(2) As yet another mode, instead of the average value of the previous manipulated variable set value and the standard manipulated variable, the average value of the weighted value of the previous manipulated variable set value and the standard manipulated variable is calculated. May be used. In this case, the weighting is defined by the length of the suspension time of the system. That is, if the stop time is short, the weight is increased, and if the stop time is long, the weight is reduced. As a result, when the stop time is short and the state change occurring during the period is stochastically small, the value becomes close to the previous set value. Conversely, when the stop time is long and it is difficult to predict the state change occurring during the period. Is close to the standard setting.

(3) The present invention is not limited to the above-described embodiment, and various other conversions may be performed, such as multiplying a previous operation amount set value by a predetermined coefficient. In short, it is only necessary to reduce the influence of the state change before and after the restart by performing some conversion without using the previous operation amount set value as it is.

(4) In this embodiment, the restart when the power switch is turned off has been described. However, the present invention is not limited to this, and it is assumed that the system state greatly changes, such as restart by a reset switch. If so, another type of restart may be used. (5) Further, in the present embodiment, the control of the laser printer has been described. However, the present invention is not limited to this and can be applied to the control of other devices.

As described above, the inventions according to the first to fifteenth aspects have the effect that the best control performance can always be realized without any data collection or optimization work by a technician. For this reason, the development cost can be reduced, and sufficient performance can be obtained without a skilled technician.

Further, according to the present invention, it is possible to realize fully automatic real-time learning and tuning which cannot be performed by conventional techniques such as fuzzy and neural networks. Furthermore, according to the present invention, it is not necessary to deeply understand and model the controlled system, and it is possible to control the system with sufficient accuracy even if it is a black box. Further, according to the present invention, it is possible to reduce sensors (cost) and to select a free control method which is free from restrictions of sensors. Moreover, according to the present invention, the optimum adjustment is automatically performed even when the controlled system deteriorates with time or when components are replaced.

Furthermore, according to the second and fourth aspects of the present invention, even if there is no case sufficiently similar to the situation at the time of operation in past cases, new control rules can be obtained from a plurality of past control rules. Create rules.

Further, in the invention according to claim 2 ,
Since the control device itself can independently determine the necessity and additionally store the control case, the control performance can be independently improved.

Further, according to the inventions set forth in claims 5 to 11 , the various effects described above can be realized with a small amount of memory, and the effect that memory shortage does not occur can be obtained.

According to the twelfth aspect of the present invention,
When the system is restarted, the influence of a change in the state of the system before and after the restart can be reduced, and highly accurate control can be performed promptly from the beginning of the restart.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an overall configuration of a first embodiment of the present invention.

FIG. 2 is a conceptual diagram showing an example plane at the time of startup of the embodiment.

FIG. 3 is a diagram showing a control case space when two operation amounts and two control amounts are employed in the embodiment.

FIG. 4 is a diagram for explaining how to obtain a correction value of an operation amount in the case shown in FIG. 3;

FIG. 5 is a conceptual diagram for explaining a method of calculating a degree of conformity when a control plane is used.

FIG. 6 is a schematic configuration diagram showing an overall configuration of a second embodiment of the present invention.

FIG. 7 is a block diagram illustrating a configuration of a control device according to a second embodiment.

FIG. 8 is a diagram for explaining a control rule at the time of startup in the embodiment.

FIG. 9 is a block diagram showing an overall configuration of a third embodiment of the present invention.

FIG. 10 is a diagram for explaining a relationship between an operation amount and a control amount before and after restart when the operation amount set value before stop is used as the operation amount after restart.

FIG. 11 is a diagram for explaining a relationship between an operation amount and a control amount before and after restart when an average value of the operation amount set value and the standard operation amount before stop is set as the operation amount after restart.

FIG. 12 is a diagram for explaining the relationship shown in FIG. 11 under another condition.

FIG. 13 is a diagram for explaining the relationship shown in FIG. 11 under another condition.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Controlled system 12 Correction rule preparation part (control rule extraction means) 15 Case storage part (control case storage means) 16 Operation amount correction operation part (operation amount operation part) 17 Operation amount set value output part (operation amount output means)

Continuation of the front page (56) References JP-A-3-123903 (JP, A) JP-A-5-143107 (JP, A) JP-A-2-77741 (JP, A) JP-A-4-1499663 (JP JP-A-4-236612 (JP, A) JP-A-7-191708 (JP, A) JP-A-6-250707 (JP, A) JP-A-5-11809 (JP, A) 5-173605 (JP, A) JP-A-2-89102 (JP, A) Suzuki Kazunori, et al., "Fuzzy Modeling of Activated Sludge Process", Proc. Of the 7th Fuzzy System Symposium, Japan Fuzzy Academic Society, June 12, 1991, p. 331-334 (58) Field surveyed (Int. Cl. ⁶ , DB name) G05B 13/00-13/04 G06F 9/44 580 JICST file (JOIS)

Claims (15)

(57) [Claims] An operation amount output means for applying an operation amount to a controlled device, wherein the operation amount and a control amount correspondingly output from the controlled device and a control amount of the controlled device are affected. A control case storage unit that stores a set of state quantities as a control case and creates a cluster by grouping similar state quantities together, and extracts a function model of a control target for each cluster. br /> control rule extracting means for function model <br/> extracted by the control rule extracting means calculates a goodness of fit for the immediate control case, each Seki
The number of models is weighted according to the degree of conformity and averaged, an operation amount for matching the control amount to the target value is obtained by using the resultant synthesis rule, and the operation amount is output to the operation amount output means. A system control device comprising: an operation amount calculating unit. 2. An operation amount output means for applying an operation amount to a controlled device, wherein the operation amount and a control amount correspondingly output from the controlled device and a control amount of the controlled device are affected. A control case storage unit that stores a set of state quantities as a control case and creates a cluster by grouping similar state quantities together, and extracts a function model of a control target for each cluster. br /> control rule extracting means for function model <br/> extracted by the control rule extracting means calculates a goodness of fit for the immediate control case, each Seki
The number of models is weighted according to the degree of conformity and averaged, an operation amount for matching the control amount to the target value is obtained by using the resultant synthesis rule, and the operation amount is output to the operation amount output means. When the control amount exceeds the allowable error range, the control case storage unit stores the control content as a new control case in a corresponding cluster, and stores the control rule extraction. The system control device extracts the new function model from the cluster including the control case. 3. The function model is defined as an n-dimensional plane based on a least squares error method of each control case coordinate point in an n + 1-dimensional space composed of a control amount and its associated n operation amounts. 3. The system control device according to claim 1, wherein the system control device is extracted every time. 4. An operation amount output means for applying an operation amount to a controlled device, wherein the operation amount and a control amount correspondingly output from the controlled device and a control amount of the controlled device are affected. A control case storage means for storing a set of state quantities as a control case and creating clusters by grouping similar state quantities together, and for each cluster, a control quantity and n related A control rule extracting means for extracting a function model of a control target for each control amount as an n-dimensional plane according to a least square error method of each control case coordinate point in an n + 1-dimensional space formed by the operation amount; for function model <br/> extracted by the extracting means, obtains the goodness of fit for the immediate control case, each Seki
The number of models is weighted according to the degree of conformity and averaged, an operation amount for matching the control amount to the target value is obtained by using the resultant synthesis rule, and the operation amount is output to the operation amount output means. And a manipulated variable computing means, wherein the manipulated variable computing means comprises a coordinate space between the n-dimensional plane representing the function model and a coordinate point representing the immediately preceding control case in a coordinate space in which each function model is described. A system control device, wherein a reciprocal of a distance is normalized for each function model to obtain the degree of conformity. 5. When insufficient memory capacity is generated in the control case memory means, the system control apparatus according to claim 1, characterized in that erasing the oldest control case. Wherein said control case memory means, at the time the cluster is completed, according to claim 1 and characterized in that erasing the control case which is a component of that cluster
5. The system control device according to any one of 4 . Wherein the control case storage means, the storage capacity when the shortage occurs, the system control device according to claim 1, 2 and 4, characterized in that erasing the oldest cluster. Wherein said control case memory means, when the storage capacity shortage has occurred, the system control device according to claim 6, characterized in that erasing the oldest cluster. 9. The control case storage means stores the time of occurrence and the number of times used for determining the operation amount as elements of the control case. stored, and claims 1, 2 and characterized by erasing the least control case of the number of times that was used
5. The system control device according to any one of 4 . Wherein said control case memory means as an element of the cluster, stores the number of times used creation time, and the operation amount determination, when the storage capacity shortage occurs, is stored in a previous predetermined time and claims, characterized in that erasing the smallest cluster number which used the 1,2 and 4
The system control device according to any one of the above. 11. The control case storage means stores a creation time and a cumulative fitness when used for determining an operation amount as elements of a cluster.
Stored in the previous predetermined time, and, according to claim 1, characterized in that erasing the smallest cluster cumulative adaptability
5. The system control device according to any one of 4 . 12. An operation amount output means for applying an operation amount to a controlled device, wherein said operation amount and a control amount correspondingly output from said controlled device and a control amount of said controlled device are affected. A control case storage unit that stores a set of state quantities as a control case and creates a cluster by grouping similar state quantities together, and extracts a function model of a control target for each cluster. br /> control rule extracting means for function model <br/> extracted by the control rule extracting means calculates a goodness of fit for the immediate control result, each Seki
The number of models is weighted according to the degree of conformity and averaged, an operation amount for matching the control amount to the target value is obtained by using the resultant synthesis rule, and the operation amount is output to the operation amount output means. An operation amount calculating means, wherein the operation amount calculating means restarts the operation amount in the control case immediately before the stop when the controlled device or the control operation is stopped and then restarted. Before and after
A system control device which performs conversion for reducing the influence of a change, outputs the manipulated variable after the conversion to the manipulated variable output means, and obtains a degree of conformity of each function model to a control result obtained thereby. 13. The effect of a change in a state quantity before and after the restart.
13. The system control device according to claim 12 , wherein the conversion for reducing is a conversion for obtaining an average of a predetermined standard operation amount and an operation amount in the control case immediately before the stop. 14. The effect of a change in a state quantity before and after the restart.
Reducing the conversion, the average value of the operation amount of the past control cases
13. The system control device according to claim 12, wherein the conversion is a conversion for obtaining an average with an operation amount in the control case immediately before the stop . 15. The effect of a change in the state quantity before and after the restart is determined.
The conversion to reduce is the one that weighted the previous operation amount
13. The system control device according to claim 12, wherein the conversion is a conversion for obtaining an average of the standard operation amount and a predetermined standard operation amount .