JPH06102908A - Demand prediction device - Google Patents

Abstract

PURPOSE:To predict heat demand with high prediction precision and to improve the economy of the operation of a heat source unit. CONSTITUTION:A long term demand prediction means 100 predicts the change of heat demand for the next whole day. An operation plan generation means 400 generates an operation schedule 107 based on the predicated result. A short term demand prediction means 200 executes prediction based on the temperature of the day and an actual heat demand value, decides the operation schedule 107 at every hour and minutely corrects it. A control means 500 controls a plant X based on a corrected result. Thus, heat demand prediction precision and the economy of the operation of the heat source unit can be improved.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a heat demand predicting device for predicting heat demand in advance.

[0002]

2. Description of the Related Art Regarding prediction of heat demand and operation control method of heat source equipment based on the prediction result, see Air Conditioning and Sanitary Engineering, Volume 64, No. 10 (1990), pp. 23-27. It is discussed in "Regional Cooling Control System" (hereinafter referred to as "Prior Art 1"), JP-A-4-15441 (hereinafter referred to as "Prior Art 2"), and the like.

The prediction of heat demand in the above-mentioned prior art 1 is as follows.
This is done by a method of multiplying a factor that affects demand by a coefficient and using the sum as a predicted value. Estimated time is 24
There are two types of time and one hour ahead.

The prediction of the heat demand in the prior art 2 is a method of inputting factors affecting heat demand such as forecast data such as weather and temperature of the next day and days of the week to the neural network and outputting the heat demand pattern of the next day. There is.

[0005]

However, in the above-mentioned prior art 1, since the same prediction formula is used in the long-term prediction and the short-term prediction, the data that cannot be obtained at the time of performing the long-term prediction. It was not possible to make a more accurate forecast by newly incorporating the in the short-term forecast.

Further, in the prior art 2, since there is only one type of prediction period, the heat demand forecast based on the data obtained on or before the previous day is used to calculate the actual heat demand on the day and the temporal change of the outside air condition. However, the prediction accuracy was poor. Therefore, it is difficult to automatically optimize the operation of the heat source device according to the heat demand.

An object of the present invention is to provide a demand forecasting device with high forecasting accuracy.

An object of the present invention is to provide a device operation control device capable of automatically optimizing the operation of a heat source device.

[0009]

SUMMARY OF THE INVENTION The present invention has been made to achieve the above object, and as one mode thereof, a predetermined long-term prediction target period in a demand forecasting device for forecasting fluctuations in heat demand. Long-term forecasting means for predicting fluctuations in heat demand (hereinafter referred to as "long-term forecasting") and forecasting fluctuations in heat demand within a short-term forecasting period shorter than the above long-term forecasting period (hereafter "short-term forecasting") And a short-term prediction means, the short-term prediction means performing the short-term prediction using data including the result of the long-term prediction. .

The above-mentioned long-term forecasting means carries out the above-mentioned long-term forecasting by using at least one of a forecast of weather, a forecasted minimum temperature, and a forecasted maximum temperature for a time region subject to the long-term forecast. Is also good. The above-mentioned short-term prediction means is the actual temperature at the time of executing the short-term prediction (hereinafter referred to as “current temperature”), the difference between the temperature at the time of executing the previous short-term prediction and the current temperature, the actual humidity at the time of executing the short-term prediction (hereinafter "Current humidity"), the difference between the humidity at the time of executing the previous short-term prediction and the current humidity, the actual heat demand value at the time of performing the prediction (hereinafter referred to as "current demand value"), and the short-term prediction execution time Short-term prediction may be performed using at least one of the prediction result of the long-term prediction means, the difference between the prediction result and the current demand value, and the prediction execution time.

The long-term predicting means and the short-term predicting means are
At least one of them may be configured to include a neural network learned by using teacher data including the relationship between past heat demand fluctuations and weather conditions.

In a control device for controlling the operation of equipment to be controlled according to heat demand, long-term prediction means for predicting fluctuations in heat demand within a long-term prediction target period (hereinafter referred to as "long-term prediction"), and a book As another aspect of the invention, by using the data including the result of the long-term prediction, the fluctuation of the heat demand in the short-term prediction target period shorter than the long-term prediction target period is predicted (hereinafter referred to as “short-term prediction”). ) A short-term prediction means, and an operation plan creation means for creating an operation plan of the equipment based on the result of the long-term prediction and modifying the operation plan based on the result of the short-term prediction. An operation control device is provided.

As another aspect of the present invention, in a plant apparatus capable of controlling the operating state of equipment according to heat demand, the fluctuation of demand within a long-term prediction target period is predicted (hereinafter referred to as "long-term prediction"). Predict the fluctuation of heat demand within the short-term forecast period shorter than the long-term forecast period by using the forecast unit and the data including the results of the long-term forecast unit (hereinafter referred to as "short-term forecast") Prediction means, equipment to be controlled, based on the result of the long-term prediction, while creating an operation plan of the device, operation plan creation means for modifying the operation plan based on the result of the short-term prediction, A plant apparatus comprising: a control unit that controls the device based on the operation plan.

[0014]

[Operation] Using at least one of the weather forecast, the predicted minimum temperature, and the predicted maximum temperature for the time region that is the target of the long-term prediction (for example, the day after the next day), the heat Predict demand fluctuations. For example, the predicted heat demand value for each hour on the next day is predicted.

On the other hand, the above-mentioned short-term forecasting means, present temperature, the difference between the temperature at the time of executing the previous short-term forecast and the present temperature,
Current humidity, the difference between the humidity at the time of executing the short-term prediction last time and the current humidity, the current demand value, the prediction result of the long-term prediction means for the short-term prediction execution time, the difference between the prediction result and the current demand value , And the prediction execution time are used to predict the heat demand within the short-term prediction period.

The operation plan preparation means prepares an operation plan based on the result of the long-term prediction, and corrects the operation plan based on the result of the short-term prediction. The control means controls the device according to the operation plan.

Thus, by correcting the long-term demand forecast with the demand forecast in a short time, it is possible to make a highly accurate heat demand forecast, and it is possible to perform optimum automatic operation of the heat source equipment.

[0018]

An embodiment of the present invention will be described with reference to the drawings.

FIG. 1 shows the basic configuration of the operation control device for the heat source device of this embodiment.

This embodiment comprises a long-term demand forecasting means 100, a short-term demand forecasting means 200, a storage means 300, an operation plan creating means 400, and a control means 500.

Before describing the details of each section, first, an outline of the overall operation of this embodiment will be described.

The long-term demand predicting means 100 changes the daily heat demand on an hourly basis on the basis of information 101 provided by an operator such as a factor affecting the heat demand (for example, the contents of a weather forecast). The heat demand forecast value 103
And is stored in the storage means 300. Although not used in the prediction processing in this embodiment,
Plant data such as information showing fluctuations in heat demand the day before 10
The prediction may be performed in consideration of the number 4 as well.

The operation plan creating means 400 is the storage means 3
The heat demand forecast value 103 'is acquired from 00, and the operation schedule 107 for the next day of the heat source device is created.

On the same day, the short-term demand forecasting means 200 outputs the heat demand forecasting value 105 for every one hour and stores it in the storage means 300 in order to correct the heat demand forecasting value 103 '. Then, the operation plan creating means 400 acquires the heat demand forecast value 105 'and corrects the operation schedule established based on the heat demand forecast value 103'.

The control means 500 outputs a control command 108 such as a start / stop command and a load command in order to control each heat source device constituting the plant X according to the correction result.

Next, details of the long-term demand forecasting means 100 will be described with reference to FIGS. 2, 3, and 4. Long-term demand forecasting means 1
00 includes an input interface 120, a neural network 130, and an output interface 140.

The input interface 120 indicates the state at the time of predicting the information (expected weather, predicted maximum temperature, predicted minimum temperature) 101 given by the operator, and the plant data 102 such as the temperature and the humidity, from 0 to 1. Converted to a numerical value in the range of (Hereinafter, a certain numerical value, state, etc.,
Converting to a numerical value in the range of and expressing it is called "normalization". ) And output as the normalized input information 109.

For example, when the expected weather is fine, the node 131a corresponding to "fine" is set to 1, and the nodes 131b and c corresponding to "rainy" and "cloudy" are set to 0. The converted input information 109 is output. For temperature,
It is linearly converted and output within a preset constant temperature range. For example, when the temperature range is set to 0 to 40 ° C, the temperature of 20 ° C is converted to 0.5 and the temperature of 30 ° C is converted to 0.75.

The neural network 130 is composed of an input layer 131, an intermediate layer 132, and an output layer 133, as shown in FIG. The input layer 131 has a total of 5 nodes. The node 131a is for inputting the normalized input information 109 for the expected weather "clear". Similarly, the node 131b is "cloudy" and the node 131c is "cloudy".
The node 131 is provided corresponding to the predicted maximum temperature, and the node 131e is provided corresponding to the predicted minimum temperature. The number of nodes in the intermediate layer 132 is arbitrary.
The number of nodes in the output layer 133 is 24. The number of 24 pieces corresponds to predicting the change in the heat demand of one day by dividing it into 24 sections, that is, every hour. However, the type of input data, the number of nodes, etc. are not limited to these, and the length of the period to be predicted, the number of period divisions, the method of division, and the required prediction accuracy are required. It will be changed accordingly. Needless to say, when the prediction is performed in consideration of the plant data 104, a node for inputting the data is provided.

In each node of the neural network 130,
A weight matrix, which is a coefficient matrix between neurons, is given in advance. This weight matrix is formed by giving the weather, the maximum temperature, and the minimum temperature of a certain past day as an input of the neural network 130 and the actual heat demand value of the day as an output of the neural network 130. It is a thing. The formation of the weight matrix is usually called "learning". Needless to say, learning is performed using data for a plurality of days. The present invention does not have any particular features regarding the "learning" method itself, and will not be described further.

The neural network 130 outputs the normalized forecast demand 110 corresponding to the inputted normalized input information 109 (note: needless to say, the normalized forecast demand 1
10 is also a normalized number, that is, in the range of 0 to 1. ). The normalized forecast demand 110 is converted into an engineering value by the output interface 140 and stored in the storage means 300 as the long-term heat demand forecast value 103. An example of the long-term predicted heat demand value 103 is shown in FIG. As described above, in this embodiment, the day is divided into hourly intervals, and the predicted value of the heat demand for each section is output as the long-term predicted heat demand value 103.

Details of the short-term demand forecasting means 200 will be described with reference to FIGS. 5, 6 and 7. Short-term demand forecasting means 200
Is the input interface 210 and the neural network 2
20 and an output interface 230. It has a function of outputting a short-term heat demand forecast value 105 obtained by correcting the long-term heat demand forecast value 103 based on the plant data 102 and the heat demand forecast value 103 ″.

The input interface 220 has a function of normalizing the input data and outputting it as the normalized input information 209. As input data, plant data 102
(Temperature at current time, humidity at current time, current time, current actual heat demand value) and long-term demand forecasting means 1
00 and stored in the storage means 300 (1 hour, 2 hours, 3 hours after the current time)
There is a heat demand forecast value 103 ". Further, the input interface 120 is based on these input data, and the difference between the temperature at the present time and the temperature one hour ago (hereinafter simply referred to as" temperature deviation "). , The difference between the humidity at the current time and the humidity one hour ago (hereinafter simply referred to as "humidity deviation"), the difference between the heat demand forecast value 103 "and the current actual heat demand value (hereafter simply" demand forecast value ") Deviation of) is calculated and normalized. Then, these are also output as the normalized input information 209. The contents of the plant data 102 and the normalized input information 209 are not limited to this, and may be any other information. However, when changing the information, it is necessary to change the configuration and weight of the neural network 230.

The neural network 230 comprises an input layer 231, an intermediate layer 232, and an output layer 233, as shown in FIG. In the input layer 231, a total of seven pieces of information corresponding to the above-mentioned normalized input information 209 (current temperature, current temperature, temperature deviation, humidity deviation, current heat demand value, deviation from demand forecast value, time) are provided. It has a node. Middle layer 232
The number of nodes is arbitrary. The output layer 233 has three nodes. The number of nodes in the output layer 233 is three because the short-term prediction outputs the predicted value of heat demand one hour, two hours, and three hours after the prediction is performed. It is a thing.

However, the type of input data, the number of nodes, etc. are not limited to these.

As with the neural network 130 described above, the neural network 230 is also provided with a weight matrix, which is a coefficient matrix between neurons, in advance for each node. The learning of the neural network 230 is the same as that of the neural network 130 except that the data used for learning is different.

The forecast period of the short-term demand forecasting means 200 is three hours after the present time. The prediction format is such that the prediction period is divided every hour and the heat demand prediction value for each section is output. The predicted execution time is every hour. In the neural network 220, the long-term demand forecasting means 1
Similar to 00, a weight matrix which is a coefficient matrix between neurons is formed before performing prediction.

The neural network 220 is
Normalized forecast demand 210 based on the normalized input information 209
(Note: it goes without saying that the normalized forecast demand 210 is also normalized, that is, in the range of 0 to 1,
It is a numerical value. ). The normalized forecast demand 210 is converted into an engineering value by the output interface 240, and the demand forecast value after one hour, two hours, and three hours from the current time (hereinafter, "
Storage means 300 as short-term demand forecast value 105 ")
Stored again in.

An example of the short-term heat demand forecast value 105 is shown in FIG. In the example of this figure, it is shown that the prediction is executed at the current time of 11 o'clock and the predicted values at 12 o'clock, 13 o'clock and 14 o'clock are obtained.

The short-term heat demand forecast value 105 may be output as a difference from the current heat demand value.

The operation plan creating means 400 will be described.

The operation plan creating means 400 is an operation schedule.
The process for creating the rule 107 and its correction process are performed. The operation plan creating means 400 is not limited in its hardware and software configurations as long as the following functions can be realized.

An outline of the operation plan creation processing based on the long-term heat demand forecast value 103 'will be described with reference to FIG.

The operation plan preparing means 400 first prepares an operation schedule plan for each hour of start / stop so that the operation cost is minimized (step 6).
01). The operation schedule plan can be created using mathematical programming, which is a known technique. Also, any other method may be used.

Next, a situation in which each heat source device is operated according to the operation schedule proposal is simulated in consideration of plant dynamic characteristics (step 602).
The plant dynamic characteristics include the characteristics of piping, the time delay of heat conduction, and the effect on the plant when a large number of devices operate.

Then, based on the result of this simulation, the operating characteristics are evaluated from the viewpoint of, for example, frequent start and stop of the specific equipment (step 6).
03). As the evaluation function in this case, the number of times of start / stop, operating cost, limitation on fuel such as electric power and gas, etc. can be considered. The convergence condition value for each evaluation function is set in advance.

It is determined whether the results of the evaluation performed in step 603 have converged, that is, whether all the conditions are satisfied (step 604). If it does not converge, that is, if there is a condition that is not satisfied, the operation schedule proposal is corrected (step 605), and then the flow returns to step 602 to perform the simulation again. In addition, the correction of this schedule,
Although it is executed by knowledge processing, fuzzy inference, etc., its specific content is not related to the essence of the present invention, and therefore detailed description is omitted here. On the other hand, step 6
When all the convergence conditions are satisfied in 04,
The operation schedule 107 is output to the control means 500.

The process shown in FIG. 8 is executed only once a day in order to predict the heat demand for the next day.

FIG. 9 shows an example of the result of creating the operation schedule 107. In the example of this figure, the operating / stopped states of devices 1 to N and the load factor are output every hour. Equipment 1
Is started at 7 o'clock and is operated at 50% load until 9 o'clock.
Then, after 9 o'clock, the load is raised to 100%.
The device 2 is started at 8 o'clock and operated at 10% load up to 10 o'clock. Then, after 10:00, the load is increased to 100%. The device N is started at 11:00 and the load is 50%.
It will be operated until 2:00. Then, at 12:00, the load becomes 1
Raise it to 00%. After that, the load is reduced to 50% at 14:00 and stopped at 15:00.

A process for correcting the operation schedule 107 based on the short-term heat demand forecast value 105 will be described with reference to FIG. It is assumed that the N devices making up the plant are each given an identification number from 1 to N in advance.

First, 1 is substituted into the variable i in order to initialize the variable i for specifying the device to be modified (step 651).

The short-term heat demand forecast value 105 is compared with the current operation schedule 107, and it is determined whether the load is sufficient as it is, in other words, whether the heat demand can be sufficiently satisfied (step 652). However, only the short-term demand forecast value 105 after one hour is used in the determination. The data after 2 hours and after 3 hours are merely for obtaining the prospect of excess or deficiency of heat when the operation according to the contents of the operation schedule 107 is performed thereafter. By obtaining such an outlook, if the warm-up operation of the heat source device that requires a certain amount of time to start is performed in advance, it is possible to take prompt action even when the operation is actually corrected after 2 hours or 3 hours. it can.

If the result of the determination in step 652 is that the load is insufficient, it is confirmed that there is an uncorrected device (step 653), and then the load of the device i is increased to generate heat. Increase the amount (step 654).
The amount of increase in this case naturally corresponds to the insufficient amount of heat. Note that the load of the device i is not always 100
There is no need to increase it to%. For example, when the operating state with a load of 90% is the most preferable from the viewpoint of operating cost, the load of the device is increased to 90%. Then, as will be described later, the further shortage is dealt with by increasing the load of the next device (in the order of the identification numbers).

Then, in step 655, the variable i
Add 1 to the next device to be corrected. And
Returning to step 652, it is determined whether the heat is still insufficient.

In step 653, the variable i =
If it is N + 1, the process is terminated.

On the other hand, if it is determined in step 652 that the load is not insufficient, this time, conversely, it is determined whether or not the load can be reduced, in other words, whether or not there is too much heat. (Step 656). However, as in the case of step 652, only the short-term demand forecast value 105 after one hour is used in the determination.

If the result of determination in step 656 is that the load can be reduced, after confirming that there is still a device that has not been modified (step 657),
The load of the device i is reduced to reduce the heat generation amount (step 658). The amount of decrease in this case naturally corresponds to the amount of surplus heat. It is not always necessary to reduce the load of the device i to 0% (that is, stop). For example, if it is preferable to keep the system running even if it is small in order to avoid a situation where the system is repeatedly stopped and restarted, the load on the device is reduced only to a certain level. Further, as will be described later, the excess is dealt with by reducing the load on the next device (in the order of the identification numbers).

Then, in step 659, the variable i
Is incremented by 1 and the next device is targeted for correction (step 659). Then, returning to step 652, it is determined whether or not the heat is still insufficient.

In step 657, the variable i =
If it is N + 1, the process is terminated.

The correction process described above does not require a process that requires a long time like the simulation, so that the operation schedule can be corrected quickly. However, the correction process is not limited to this.
For example, the same processing as the processing shown in FIG. 8 may be executed every time.

The forecasting period of the long-term demand forecasting means and the short-term demand forecasting means is not limited to those shown in the above embodiment. The forecast period of the long-term demand forecasting means may be one day or more or one day or less. Similarly, the prediction interval is not limited to one hour. Similarly, in the short-term predicting means, the prediction period, the prediction interval, and the prediction execution cycle may be within 1 hour. Furthermore, in the above embodiment, both the long-term demand forecasting means and the short-term demand forecasting means,
Although it is configured to include the neural network, it is not always necessary to use the neural network. If the short-term demand forecasting means is capable of forecasting the demand with higher accuracy by using the forecasting result of the long-term demand forecasting means, its specific configuration is not limited at all.

As described above, in the above embodiment, the long-term (one day or more) demand forecast result is corrected by the short-term demand forecast, so the forecast accuracy is improved. As a result, it is possible to automatically optimize the operation control of the heat source device in accordance with the fluctuation of the heat demand.

In the above embodiment, only the prediction of the heat demand is shown, but it goes without saying that the present invention can be applied to the demand prediction other than this and the operation of the equipment based on the prediction.

The specific structure of the present invention is not limited to the above embodiment.

[0065]

According to the present invention, the accuracy of heat demand prediction is improved. In addition, it is possible to automate efficient operation control of the heat source device that follows changes in heat demand.

[Brief description of drawings]

FIG. 1 is a basic configuration diagram of an embodiment of the present invention.

FIG. 2 is an internal configuration diagram of long-term demand forecasting means.

FIG. 3 is an explanatory diagram showing details of a neural network 130.

FIG. 4 is an explanatory diagram showing an example of a long-term heat demand prediction result.

FIG. 5 is an internal configuration diagram of short-term demand forecasting means.

FIG. 6 is an explanatory diagram showing details of a neural network 230.

FIG. 7 is an explanatory diagram showing an example of a short-term heat demand forecast value.

FIG. 8 is a flow chart showing an outline of an operation schedule creation process based on a long-term predicted heat demand value.

FIG. 9 is an explanatory diagram showing an example of an operation schedule.

FIG. 10 is a flow chart showing an outline of an operation schedule correction process based on a short-term heat demand forecast value.

[Explanation of symbols]

100 ... Long-term demand forecasting means, 101 ... Information given by operator, 102 ... Plant information, 103
...... Heat demand forecast value, 105 …… Heat demand forecast value, 10
7 ... Driving schedule, 108 ... Control command, 1
09 ... Normalized input information, 110 ... Normalized forecast demand,
120 ... Input interface, 130 ... New
Lar network, 131 ... Input layer, 132 ...
Middle layer, 133 ... Output layer, 140 ... Output interface, 200 ... Short-term demand forecasting means, 209 ... Normalized input information, 210 ... Normalized forecast demand, 220
...... Input interface, 230 ・ ・ ・ Neural network, 231 ・ ・ ・ Input layer, 232 ・ ・ ・ Intermediate layer,
233 ... Output layer, 240 ... Output interface
300, database, 400, operation plan creating means, 500, control means

Front Page Continuation (72) Inventor Akihiko Yamada 4026 Kuji Town, Hitachi City, Ibaraki Prefecture Hitachi Research Laboratory, Hitachi, Ltd.

Claims (6)

[Claims] 1. A demand forecasting device for forecasting fluctuations in heat demand, and a long-term forecasting means for forecasting fluctuations in heat demand (hereinafter referred to as "long-term forecasting") within a predetermined long-term forecast period, A short-term prediction means for predicting fluctuations in heat demand within a short-term prediction target period shorter than the prediction target period (hereinafter referred to as "short-term prediction"), and the short-term prediction means includes the result of the long-term prediction. A heat demand forecasting device characterized by performing the above short-term forecast using data. 2. The long-term prediction means performs the long-term prediction by using at least one of a weather forecast, a predicted minimum temperature, and a predicted maximum temperature for a time region targeted for the long-term prediction. The heat demand forecasting apparatus according to claim 1, wherein: 3. The short-term prediction means is an actual temperature (hereinafter "current temperature") at the time of executing the short-term prediction.
), The difference between the temperature at the time of executing the previous short-term prediction and the current temperature, the actual humidity at the time of executing the short-term prediction (hereinafter referred to as "current humidity"), the humidity and the current humidity at the time of executing the previous short-term prediction. , The actual heat demand value at the time of prediction execution (hereinafter referred to as “current demand value”), the prediction result of the long-term prediction means for the short-term prediction execution time, the difference between the prediction result and the current demand value, The heat demand prediction apparatus according to claim 1, wherein short-term prediction is performed using at least one of the above and a prediction execution time. 4. The long-term predicting means and the short-term predicting means are:
At least one of them is configured to include a neural network learned by using teacher data including a relationship between past heat demand fluctuations and meteorological conditions. The heat demand forecasting device described. 5. A control device for controlling the operation of equipment to be controlled according to heat demand, and a long-term predicting means for predicting fluctuations in heat demand within a long-term prediction target period (hereinafter referred to as "long-term prediction"). , Short-term prediction means for predicting fluctuations in heat demand within a short-term prediction target period shorter than the long-term prediction target period (hereinafter referred to as "short-term prediction") using the data including the result of the long-term prediction An operation control device comprising: an operation plan creation unit that creates an operation plan of the equipment based on the result of the long-term prediction and that corrects the operation plan based on the result of the short-term prediction. 6. A plant apparatus capable of controlling an operating state of equipment according to heat demand, and a long-term forecasting means for forecasting fluctuations in demand within a long-term forecasting period (hereinafter referred to as "long-term forecast"); Using the data including the results of the long-term forecasting means, predict the fluctuation of heat demand within the short-term forecasting period shorter than the long-term forecasting period (hereinafter referred to as "short-term forecast")
Short-term prediction means, equipment to be controlled, and an operation plan creation means for creating an operation plan of the equipment based on the result of the long-term prediction and modifying the operation plan based on the result of the short-term prediction. And a control unit that controls the device based on the operation plan.