KR101261199B1 - Method for determining pre-cooling time of building by using intelligent control algorithm with neural network model - Google Patents

Abstract

PURPOSE: A method for controlling the reserved cooling of a building using an intelligent control algorithm by a neural network model is provided to control air conditioning equipment by the intelligent control algorithm by the neural network model, thereby maintaining an indoor space of the building pleasant simultaneously with saving energy utilized for air conditioning. CONSTITUTION: A method for controlling a decrease of indoor temperature of a building using an intelligent control algorithm by a neural network model is as follows. The neural network model having input variables of indoor temperature, outdoor temperature, an inclination of the indoor and outdoor temperature and an output variable of pre-cooling time is fetched(S101). Measuring variables including the indoor temperature and outdoor temperature are measured at a predetermined time interval(S102). A circulation variable including the inclinations of the indoor temperature and outdoor temperature is calculated based on the measuring variables(S104). The pre-cooling time at measurement timing is calculated by inputting the input variables into the neural network model(S132). A difference between the pre-cooling time and target setting time is calculated and if the difference is within a set value, the operation is performed according to a preset operation process(S133,S134,S135). The inside of the building is night-purged by inducing the outdoor air before the pre-cooling using the neural network model(S121,S122,S123,S124). [Reference numerals] (AA) Start; (BB,DD,FF) No; (CC,EE,GG) Yes; (HH) End; (S101) Fetch an adapted neural network model; (S102) Measure the indoor temperature, the indoor humidity, the outdoor temperature, and the outdoor humidity every one minute; (S103) No noise included in the measured data?; (S104) Store the indoor temperature and humidity and the outdoor temperature and humidity, and calculate and store the inclinations of the indoor and outdoor temperatures; (S121) Calculate enthalpy of the indoor and outdoor air; (S122) Enthalpy of the indoor air is greater than that of the outdoor air?; (S123) Operate an intake fan and an exhaust fan; (S124) Stop the operations of the intake fan and the exhaust fan; (S131) Input the stored data to the adapted neural network model; (S132) Calculate a pre-cooling time at each measurement time through recollection in neural networks; (S133) Calculate the difference between the calculated pre-cooling time and a target time; (S134) Difference between the calculated pre-cooling time and a target time is within a preset value?; (S135) Activate a cooling facility; (S136) Start the full power operations of a freezer, an air conditioner, and an FCU; (S137) Store the time, the indoor temperature, the inclination of indoor temperature, the outdoor temperature, and the inclination of outdoor temperature when starting the operations; (S138) Store the reaching time when reaching a set temperature; (S139) Relearn an adapted backpropagation neural network model by inputting learning data collected through driving; (S140) Store learning information on the backpropagation neural network model

Description

Method for determining pre-cooling time of building by using intelligent control algorithm with neural network model}

The present invention relates to a control method for controlling an air conditioning system in a building using an optimal control technique. More specifically, by controlling a air conditioning system using a neural network model, the air conditioner can be maintained while maintaining a comfortable environment inside the building. The present invention relates to a method of pre-cooling control of a building using an intelligent control algorithm based on a neural network model to save energy used in a building.

The air conditioning equipment is installed and applied in various buildings in a wide range of fields including office buildings, factories, hotels, restaurants, hospitals, warehouse shops, department stores, and the like. The operation of the conventional air conditioning equipment generally wastes energy, such as being operated for a long time or even all day until the administrator judges empirically and stops from starting. The energy consumed by the air conditioning system is dependent on various factors such as the structure of the building envelope, the characteristics of the space used, temperature, humidity, heat radiation through the windows and the density of the rooms, and is particularly affected by the operating time.

On the other hand, in the case of air conditioning, energy consumption for heat source and power is consumed. Therefore, in order to minimize air conditioning time, general office buildings are air-conditioned during working hours. I never do that.

In the case of air conditioning according to the occupants, air conditioning should be started before the start of the room so that the room temperature can reach the set temperature when the morning room starts. It is difficult to calculate whether it can.

In particular, during summer, air conditioning is determined based on the hottest weather. Since air conditioning is often started without consideration of climatic conditions and indoor environment, air conditioning starts too early or air is too late. Cases may occur. As a result, the indoor air conditioning may not be comfortable at the start of the room, or energy waste may occur. Therefore, there is a problem in that it is necessary to determine when operating the air conditioning equipment is an optimal time of operation in terms of comfortable indoor conditions and / or energy.

Republic of Korea Patent Publication 10-2011-0100895

The present invention provides an intelligent control algorithm using a neural network model to control the air conditioning facilities by using an intelligent control algorithm using a neural network model, thereby maintaining the interior of the building in a comfortable environment and saving energy used for the air conditioning. An object of the present invention is to provide a method of controlling pre-cooling of a used building.

The reservation cooling control method of a building using an intelligent control algorithm based on a neural network model according to the present invention includes learning the indoor temperature, the indoor temperature gradient, the outdoor temperature, and the outdoor temperature gradient as input variables and precooling time as output variables. Importing the adapted neural network model; Measuring a measurement variable including the indoor temperature and the outside temperature at a set time interval; Calculating a calculation variable including the indoor temperature slope and the outside temperature slope from the measured variable; Calculating the precooling time at a measurement time point by inputting the input variable to the neural network model; And calculating a difference between the precooling time and a target setting time and performing an operation according to a set operating procedure when the difference is within a setting value.

In the present invention, the method may further include a night purge of the inside of the building by introducing outside air before the reservation cooling through the neural network model.

According to another aspect of the present invention, a method of controlling cooling of a building using an intelligent control algorithm based on a neural network model includes an indoor temperature, an indoor temperature gradient, an outdoor temperature, and an outdoor temperature gradient as input variables, and a precooling time as an output variable. Establishing a learning material database; Determining an optimized neural network model using the learning database; Determining the neural network model adapted to the target building by learning the neural network model using the learning data collected through the operation according to the operating procedure set in the air conditioning facility of the target building; Night purging the interior of the building by introducing outside air prior to the reservation cooling through the neural network model; Operating an air conditioning system of the building to be applied according to the adapted neural network model; And re-learning the adapted neural network model for inputting the learning data collected through the operation of the air conditioning system to update the neural network model.

The reservation cooling control method of a building using an intelligent control algorithm based on a neural network model according to the present invention controls the air conditioning facilities by an intelligent control algorithm using a neural network model, thereby maintaining a pleasant environment inside the building. It can save the energy used.

1 is a flowchart illustrating a method for controlling reservation cooling of a building using an intelligent control algorithm by a neural network model according to an exemplary embodiment of the present invention.
FIG. 2 is a flowchart schematically illustrating a step of determining an optimized backpropagation neural network model in a method of pre-cooling control of a building using an intelligent control algorithm based on the neural network model of FIG. 1.
FIG. 3 is a flowchart schematically illustrating a step of determining a neural network model adapted to an application target building in a method of pre-cooling control of a building using an intelligent control algorithm based on the neural network model of FIG. 1.
FIG. 4 is an embodiment of a system schematic diagram of an air conditioning facility to which a reservation cooling control method of a building using an intelligent control algorithm based on the neural network model of FIG. 1 is applied.
FIG. 5 is an embodiment of a control and information collection system diagram of the air conditioning equipment of FIG. 4.
FIG. 6 is a table schematically illustrating factors and levels influencing precooling time in a method of pre-cooling control of a building using an intelligent control algorithm based on the neural network model of FIG. 1.
FIG. 7 is a diagram illustrating embodiments of three levels of adjacent conditions in factors and levels influencing the precooling time of FIG. 6.
FIG. 8 is a diagram illustrating embodiments of three levels of wall structure in factors and levels influencing the precooling time of FIG. 6.
FIG. 9 is a diagram schematically illustrating a structure of a back propagation neural network model in a method of pre-cooling control of a building using an intelligent control algorithm based on the neural network model of FIG. 1.

Hereinafter, the present invention will be described in detail with reference to the drawings. Detailed descriptions of well-known functions and constructions that may unnecessarily obscure the gist of the present invention will be omitted.

In the present invention will be described taking a method of controlling the reservation cooling of the building using an intelligent control algorithm by the neural network model shown in the drawing as an example. However, the present invention is to start the cooling at the optimum reservation cooling time, to minimize the energy consumption at the start of re-occupation time to maintain a comfortable indoor state, using an intelligent control algorithm based on the neural network model shown in the figure It is not limited to the method of controlling reservation cooling of a building.

FIG. 1 is a flowchart illustrating a method for controlling reservation cooling of a building using an intelligent control algorithm based on a neural network model according to an embodiment of the present invention. FIG. 2 is a flowchart illustrating a step of determining an optimized backpropagation neural network model in a method of pre-cooling control of a building using an intelligent control algorithm based on the neural network model of FIG. 1. FIG. 3 is a flowchart illustrating a step of determining a neural network model adapted to an application target building in a method of pre-cooling control of a building using an intelligent control algorithm based on the neural network model of FIG. 1.

Referring to the drawings, a method of pre-cooling control of a building using an intelligent control algorithm based on a neural network model includes: building a learning data database and determining a neural network model (S200); Determining a neural network model adapted to the target building (S300); And it may include the air conditioning operation and neural network model update step (S100). At this time, the training material database construction and neural network model determination step (S200) may include a training material database construction step (S201 ~ S203) and neural network model determination step (S204 ~ S210). The air conditioning facility operation and neural network model updating step (S100) may include an air conditioning facility operation preparing step (S101 ˜ S104), an air conditioning facility operating step (S131 ˜ S135), and a neural network model updating step (S136 ˜ S140).

In the learning material database construction step (S201 ~ S203) it is possible to build a learning material database. In this case, the learning data database may be learning data having an indoor temperature, an indoor temperature gradient, an outside temperature, and an outside temperature gradient as input variables, and a precooling time as an output variable. In the neural network model determination step (S204 ~ S210) it is possible to determine the optimized backpropagation neural network model using the training material database.

In the step S300 of determining the neural network model adapted to the target building, the neural network model is adapted to the target building by learning the neural network model using the training data collected through the operation according to the operating procedure set in the air conditioning facility of the target building. You can decide.

In the air conditioner operation preparation step (S101 ˜ S104), an adaptive neural network model may be called to prepare for operation of the air conditioner accordingly. In the air conditioning facility operation step (S131 ˜ S135), the air conditioning facility of the building to be applied may be operated according to the adapted neural network model. In the neural network model updating step (S136 ˜ S140), the neural network model may be updated by re-learning an adapted neural network model for inputting training data collected through the operation of the air conditioning system.

According to the present invention, by controlling the air conditioning facilities by an intelligent control algorithm using a neural network model, it is possible to maintain the interior of the building in a comfortable environment and at the same time save energy used for air conditioning.

On the other hand, in the case of air conditioning, energy for a heat source and / or power is consumed. Therefore, in order to minimize air conditioning time, air conditioning is carried out during working hours for air conditioning of the building, and may not be performed when it is not during working hours or when there are no occupants in the office. At least, the air conditioning may be performed in different modes by dividing the occupancy time of the user in the building and the vacancy time of the user.

In addition, it may take considerable time to start air conditioning, for example cooling, to reach a comfortable temperature for the user. Therefore, when the air conditioning starts at the starting time of the occupancy time, it may not be comfortable for the user for some time after the occupancy starting time and may not meet the indoor conditions.

In this case, when air conditioning is performed in accordance with the occupants, air conditioning should be started before the room starting time so that the room temperature can reach the set temperature when the morning room starting time is reached. However, it is difficult to calculate when the air conditioning should reach the set temperature at the start of the room.

In particular, during summer, air conditioning is determined based on the hottest weather. Since air conditioning is often started without consideration of climatic conditions and indoor environment, air conditioning starts too early or air is too late. Cases may occur. Accordingly, the indoor air conditioning may not be comfortable at the time of occupancy, or energy waste may occur.

Therefore, in the present invention, by controlling the air conditioning equipment by an intelligent control algorithm using a neural network model, it is possible to save the energy used for air conditioning while maintaining the interior of the building in a comfortable environment.

On the other hand, the present specification mainly describes the optimum pre-cooling time to minimize the energy consumption and achieve the comfortable conditions inside the building. In this case, the precooling time may be a time for starting cooling before the set time. However, the present invention is not limited thereto, and may be any one of a reservation heating time for starting heating in advance of a preset time, a reservation stop time for stopping heating in advance of a preset time or pre-cooling before a set time. have.

In the air conditioner operation preparation step (S101 ˜ S104), an adaptive neural network model may be called to prepare for operation of the air conditioner accordingly. The air conditioner operation preparation step (S101 ˜ S104) may include a step (S101) of bringing up an adaptive neural network model, a measurement variable measurement step (S102), a noise removal step (S103), and a calculation variable calculation step (S104).

In the measuring variable measuring step S102, a measuring variable including at least one of an indoor temperature and an outside temperature may be measured at a set time interval. At this time, in the present embodiment, the room temperature and the outside temperature may be measured as measurement variables.

In this case, the measurement variable may be measured at a set time interval, and the precooling time may be calculated at the same time interval. Thus, the system can be efficiently controlled. However, the present invention is not limited thereto, and the measurement cycle of the measurement variable measurement cycle and the precooling time may be different for efficient control.

In this embodiment, the room temperature and the outside temperature can be measured at a time interval of 1 minute. That is, the measurement variable can be measured at a time interval of one minute, and the precooling time can be calculated at a time interval of one minute. In this case, the air conditioner can be operated more efficiently by reflecting the human feeling. However, the present invention is not limited thereto, and by controlling the condition differently and / or adaptively according to the state of the building, the value of the measurement variable, or the amount of change thereof, the air conditioner can be controlled more efficiently.

In the noise removing step (S103), the noise may be removed by determining whether the measured measurement variable is noisy. Therefore, by excluding the case where noise is included in the measurement variable when calculating the precooling time, erroneous control can be prevented.

At this time, it is determined whether there is noise in the measured data (S103), and if it is determined that there is no noise, the calculation variable calculation step (S104) is performed; when it is determined that there is noise, the measurement variable measurement step (S102) Can be measured again. In this case, the reference value range may be set, and when the measurement variable is out of the reference value range, it may be determined that the measurement variable includes noise.

In the calculation variable calculation step (S104), an indoor temperature and an outside temperature may be stored, and a calculation variable including at least one of an indoor temperature gradient and an outside temperature gradient may be calculated from the measured variable. In this embodiment, the indoor temperature slope and the outdoor temperature slope may be calculated as calculation variables. However, the present invention is not limited thereto, and any one or other values of the indoor temperature gradient and the outdoor temperature gradient may be calculated as calculation variables.

In step S101, the adaptive neural network model is loaded and the neural network model that is learned and adapted to the actual building to be driven is loaded. The daily precooling time determination mode may be started by performing a step S101 of invoking an adaptive neural network model.

At this time, the adaptive neural network model is backpropagation determined to be trained in the target building to be actually operated in the neural network model determination step (S300) adapted to the target building to the neural network model determined in the training data database construction and neural network model determination step (S200). It can be a neural network model.

Here, the neural network model may be a backpropagation neural network model that is learned by adapting the indoor temperature, the indoor temperature gradient, the outdoor temperature, and the outdoor temperature gradient as input variables and precooling time as output variables. In this case, the precooling time may be a time for starting cooling before the set time. Hereinafter, the present invention will be mainly described with reference to an embodiment in which the air conditioning facility is operated by an optimal starting algorithm using a backpropagation neural network model in order to determine an optimal precooling time.

In this case, the reservation cooling control method of the building using the intelligent control algorithm by the neural network model according to the present invention may be implemented by the optimal starting algorithm. In the case of air conditioning, energy for heat source and power is consumed, so to minimize air conditioning time, general office buildings are air-conditioned during working hours, and when not working or when there are no occupants in the office, Can be. For this purpose, when the air conditioning is performed according to the occupants, air conditioning may be started before the room starting time so that the room temperature reaches the set temperature when the morning room starting time is reached.

For this purpose, it is necessary to calculate when the air conditioning should be started to reach the set temperature at the time of starting the room. At this time, the pre-cooling time may be a pre-cooling time to reach the set temperature at the start of the room. In this case, the air conditioning system may be started according to a schedule set in advance by the optimum start algorithm to determine the optimum precooling time.

Accordingly, the pre-cooling time is calculated using the backpropagation neural network model, and the air conditioning equipment is started before the start time of the pre-cooling time to save the energy used for the air conditioning while maintaining the interior of the building in a pleasant environment. You can do that.

On the other hand, the reservation cooling control method of the building using the intelligent control algorithm by the neural network model may further include a night purge performing step (S121 ~ S124). In the night purge performing steps S121 to S124, night purge inside the building may be performed by introducing external air before the reservation cooling through the neural network model. To this end, outside air cooler than indoor air may be introduced into the room.

In hot weather, such as summer, the air conditioning unit can be operated primarily in cooling mode. However, when the outside air is colder than the indoor air, such as dawn, the indoor air may be cooled by supplying the outside air cooler than the indoor air without using the air conditioner. That is, in the night purge performing steps (S121 to S124) without cooling the operation unit, by cooling the air into the room relatively cooler than the indoor air, it is possible to perform the cooling. Therefore, by performing the night purge, it is possible to keep the room in a comfortable state while minimizing energy consumption.

In particular, when the air conditioning facility is controlled by an intelligent control algorithm using a neural network model, by cooling the air without operating the air conditioner by night purge, it is possible to further reduce the energy used for cooling before the real time. .

At this time, even when the outside temperature is low, if the humidity is high, rather if the night purge using the outside air, the outside air with high humidity is introduced into the room may make the room humidity too high. Therefore, the cold outside air can be introduced into the room only when the indoor humidity is higher than the outside air humidity. In this case, the night purge is performed using cold outside air only when the outside air humidity is lower than the humidity of the indoor air, thereby keeping the room in a comfortable state.

In this case, the concept of enthalpy may be utilized to make comfortable indoor conditions by night purge. That is, by calculating the enthalpy, only when the enthalpy of the indoor air is larger than the enthalpy of the outside air, the external cool air may be introduced into the room to perform the night purge. To this end, the indoor temperature, the outside temperature, the indoor humidity, and the outdoor humidity are measured in the measurement variable measurement step S102, and the indoor variable temperature, the outside air temperature, the indoor humidity, and the outdoor humidity are stored in the calculation variable calculation step S104. In this case, the calculated variable can be calculated and stored.

The night purge performing steps S121 to S124 may include an enthalpy calculation step S121, an enthalpy comparison step S122, and air conditioner operation steps S123 and S124. In the enthalpy calculation step (S121), the indoor air enthalpy and outdoor enthalpy can be calculated using the indoor temperature, the outside air temperature, the indoor humidity, and the outside air humidity. In the enthalpy comparison step (S122), the enthalpy of the indoor air and the enthalpy of the outside air may be compared. In the air conditioner operation steps (S123 and S124), if the enthalpy of the indoor air is larger than the set value of the enthalpy of the outside air, the air supply fan and / or the exhaust fan of the air conditioner are operated to discharge the indoor air to the outside and the cold external air to flow into the room. (S123), if the enthalpy of the indoor air is not greater than the enthalpy of the outside air, the operation of the air supply fan and / or the exhaust fan of the air conditioner can be kept in a stopped state (S124). For example, the set value may be zero.

In the air conditioner, at least a part of the indoor air, that is, a part or all of the air may be discharged, and the outdoor air may be introduced into the room in accordance with the discharged amount of indoor air. At night purge, exhaust air (EA) and outdoor air (OA) may be 100% open. In contrast, when the neural network model is applied and the air conditioner is operated, exhaust air (EA) and outdoor air (OA) may be 100% closed to maximize energy efficiency.

Therefore, in the night purge performing steps (S121 to S124) by using the enthalpy concept, it is possible to create a comfortable indoor conditions by the night purge. However, the present invention is not limited thereto, and other methods may be applied to create a more comfortable indoor environment by comparing indoor air with external air. For example, when the outside temperature is less than the set value than the room temperature, the night purge may be performed by discharging the indoor air to the outside and introducing the outside air into the room.

In the air conditioning facility operation step (S131 ˜ S135), the air conditioning facility of the target building can be operated according to the adapted neural network model. The air conditioning facility operation steps S131 to S135 may include an input variable input step S131, a precooling time calculation step S132, and an operation start step S133 to S135.

In the air conditioning operation step (S131 ~ S135) can be implemented by the daily pre-cooling time determination mode using the adaptive neural network model. That is, by determining the pre-cooling time of the day by the daily pre-cooling time determination mode to start the operation earlier than the operation start time set by the air conditioning equipment, the user prepares to use the building before the time used in earnest by the user, Ease of convenience can be improved. In addition, by calculating the optimal precooling time by an intelligent control algorithm, it is possible to save energy while ensuring user convenience.

In the input variable input step S131, the input variable may be input to the neural network model. In addition, in the precooling time calculation step (S132), the precooling time at the measurement point may be calculated by the neural network model using an input variable. In this case, the precooling time may be calculated as the output variable by inputting the indoor temperature, the outside temperature, the indoor temperature gradient, and the outside temperature gradient as input variables to the backpropagation neural network model.

In the operation start steps (S133 to S135), the difference between the pre-cooling time and the target set time can be calculated and the operation can be performed according to the set operating procedure if it is within the set value. At this time, the difference between the pre-cooling time and the target setting time is calculated (S133), and it is determined whether the difference between the calculated pre-cooling time and the target setting time is within the set value (S134), and the cooling equipment is operated when it is within the set value. It may be (S135).

At this time, when the difference between the calculated pre-cooling time and the target setting time is out of the range of the set value, the measurement variable measurement step (S102) may be performed to measure the measurement variable again, and the subsequent step S103 may be performed. Therefore, it is possible to prevent the waste of energy by operating the cooling facility at the wrong time due to miscalculation.

In the neural network model updating step (S136 ˜ S140), the back propagation neural network model may be updated by re-learning an adapted neural network model for inputting training data collected through the operation of the air conditioning system. That is, by operating the actual air conditioning facilities in the target building, and reflecting the results back to the neural network model, it is possible to reinforce the training data of the adapted neural network model to improve its accuracy. Typically, neural network models can improve the accuracy of their predictions as their training data increases.

The neural network model updating steps S136 to S140 may include a full power operation step S136, an input variable storing step S137, an arrival time storing step S138, and a relearning and storing step S139 and S140. .

In the full power operation step S136, the air conditioning equipment may be started at full load. At this time, input variables can be obtained quickly by operating at full power, and accordingly, the data according to the operation of the air conditioning equipment can be quickly reflected so that the backpropagation neural network model adapted to the target building and the operating conditions can be obtained more quickly and more accurately. do.

In the input variable storage step S137, an operation start time, a measurement variable, and a calculation variable may be stored at an operation start time. In the arrival time storing step S138, the arrival time for reaching the set temperature from the operation start time may be stored. At this time, the arrival time under the given conditions can be reflected in obtaining the next precooling time.

At this time, the learning data including the indoor temperature, the outside temperature, the indoor temperature gradient, the outside temperature gradient, and the arrival time can be measured, calculated, and stored, thereby reflecting the current operating conditions and conditions as the learning data. Relearn neural network models so that they can be updated.

In the re-learning and storage steps S139 and S140, learning data collected through driving may be input to relearn and store the adapted backpropagation neural network model. In the re-learning and storage steps (S139 and S140), retraining the neural network model by inputting learning data collected through driving (S139), and reinforcing the learning information by additionally storing the learning information of the back-propagating neural network model to back-propagating neural network. You can update the model.

In the learning material database construction step (S201 ~ S203) it is possible to build a learning material database. By establishing a reference database for learning the backpropagation neural network model in the training material database construction step (S201 ~ S203), it is possible to determine the neural network model optimized for the neural network model determination step (S204 ~ S210).

Learning data database construction step (S201 ~ S203) may include an influence factor and level determination step (S201), factor setting and combination step (S202), and learning data collection step (S203). Influence factor and level determination step (S201) it is possible to determine the factors and their levels affecting the precooling time. In the setting and combining step (S202), the level of each factor affecting the precooling time may be set and combined. In the learning material collection step (S203), the learning material may be collected to build a learning material database.

In this case, the learning material database may be constructed through simulation and / or experiment. Here, in this embodiment, by building a training material database through simulation, it is possible to easily build a schematic database necessary for learning to build an optimal backpropagation neural network model without an actual experiment. However, the present invention is not limited thereto, and by obtaining learning materials by experiments or other methods, more accurate learning materials can be obtained.

Influence factor and level determination step (S201) it is possible to determine the factors and their levels affecting the precooling time. First, in order to build a learning database, it is possible to determine the factors that affect preheating time and their level.

In an embodiment of the present invention, the factors affecting the precooling time may include at least one of an outdoor condition, a floor area, a target temperature, a window area ratio, a shielding coefficient, a wall structure, an adjacent condition, an indoor heat capacity, and an indoor heating load. Can be. In order to construct an optimal backpropagation neural network model, as shown in FIG. 6, the outdoor condition, floor area, target temperature, window area ratio, shielding coefficient, wall structure, adjacent condition, indoor heat capacity, and indoor heating load are nine factors. It may include.

In this case, as shown in FIG. 6, in order to build an optimal backpropagation neural network model, factors affecting each precooling time may have three levels. Here, the outside conditions may have levels of Seoul, Daejeon, and Busan. Floor area (m 2) can have levels of 100, 250, and 500. The target temperature (° C.) may have levels of 25, 26, and 27. The window area ratios can have levels of 0.40, 0.65, and 0.90. The shielding coefficients can have levels of 0.40, 0.60, and 0.80. The wall structure may have levels of A, B, and C as shown in FIG. 8. Adjacent conditions may have levels of a, b, and c as shown in FIG. 7. The indoor heat capacity (KJ / m 3 K) can have levels of 9, 12, and 15. The indoor heating load (W / m 2) can have levels of 50, 60, and 70.

Adjacent condition is, as shown in FIG. 7, a level surrounds the target space on three sides, b level surrounds the target space on two sides, and c level is surrounded by the outside air. Can be a condition.

On the other hand, the wall structure, as shown in Figure 8, A level may be a wall structure laminated in the order of mortar-cement brick-styrofoam-cement brick-mortar, B level granite-concrete-glass surface-gypsum It can be a wall structure laminated in the order of the board, C level can be a wall structure laminated in the order of glass-air layer-glass-glass surface-gypsum board.

In the setting and combining step (S202), the level of each factor affecting the precooling time may be set and combined. At this time, it can be combined by setting the level of each of the above factors by the experimental design method. Therefore, even a small number of simulations and / or experiments can be used to collect training material for determining the optimal backpropagation neural network model.

In the learning material collection step (S203), the learning material may be collected to build a learning material database. At this time, through the heat transfer analysis using the finite difference method it is possible to build a learning data database by collecting the learning data including at least one of the measured variable, the calculated variable, and the precooling time.

At this time, it is possible to collect the learning data on the indoor temperature, the outside temperature, the indoor temperature gradient, the outdoor temperature gradient and the precooling time by the finite difference method. Therefore, it is possible to construct a learning data database for constructing an optimal backpropagation neural network model conveniently without performing actual experiments using finite difference simulation.

In the neural network model determination step (S204 ~ S210) it is possible to determine the optimized backpropagation neural network model using the training material database. In the neural network model determination step (S204 ~ S210) is determined the input and output variables for optimal backpropagation neural network model training (S204), collected in the training data database construction step (S201 ~ S203) in the backpropagation neural network model program Input the training material (S205), it is possible to determine the optimal back-propagation neural network model from this (S206 ~ S210).

In this case, the input variables for the optimal backpropagation neural network model training may be room temperature, outside temperature, room temperature slope, and outside temperature slope. Also, the output variable for the optimal backpropagation neural network model training may be a precooling time.

Meanwhile, the optimal backpropagation neural network model may be determined by calculating an optimal value of at least one of a learning rate, a moment, a bias, the number of hidden layers, and the number of hidden layer nodes. In the present embodiment, it may be determined by calculating an optimal value of each of the learning rate, moment, bias, number of hidden layers, and number of hidden layer nodes.

To this end, the neural network model determination step (S204 ~ S210) to calculate the optimal value of the learning rate of the backpropagation neural network model for determining the optimal precooling time (S206), the moment of the backpropagation neural network model for determining the optimal precooling time The optimum value is calculated (S207), the optimum value of the bias of the backpropagation neural network model for determining the optimal precooling time is calculated (S208), and the optimum value of the number of hidden layers of the backpropagation neural network model for the optimal precooling time is determined. In operation S209, an optimal value of the number of hidden layer nodes of the backpropagation neural network model for determining an optimal precooling time may be calculated (S206). In this case, each optimum value for determining the optimal precooling time may be calculated through an error minimization technique. Therefore, it is possible to obtain an optimal pre-cooling time to minimize the error by using an intelligent control algorithm according to an embodiment of the present invention.

In the step S300 of determining the neural network model adapted to the target building, the neural network model is trained using the learning data collected through the operation according to the operating procedure set in the air conditioning facility of the target building to obtain the neural network model adapted to the target building. You can decide. Therefore, by adjusting the back propagation neural network model according to the actual building application, it is possible to construct an adaptive back propagation neural network model to obtain a more accurate precooling time.

In the neural network model determination step (S300) adapted to the target building, the indoor temperature and the outside temperature are measured and stored in a set time interval, for example, in 1 minute increments (S301), and the cooling facility is started through schedule control (S302), Neural network model adjustment step (S303 ~ S309) to adjust the back propagation neural network model to the actual building.

At this time, it is possible to determine the neural network model adapted to the target building by collecting the learning data by operating the cooling facility through the schedule control according to the set operation procedure. Therefore, by collecting the learning data reflecting the actual operating conditions of the cooling facility, it is possible to determine a neural network model that can obtain a more accurate pre-cooling time when the cooling facility for the actual building operation.

In the neural network model adjustment step (S303 ~ S309), the back propagation neural network model can be determined by learning the adapted neural network model by inputting the learning data collected through the operation of the air conditioning facility in the actual building. That is, the accuracy of the neural network model can be improved by collecting the training data by operating the actual air conditioning facilities in the target building. This allows for a more accurate precooling time for a given input variable condition.

Neural network model adjustment step (S303 ~ S309) is the full power operation step (S303), input variable storage step (S304), arrival time storage step (S305), neural network model loading step (S307), and neural network model learning and determination step ( S308 and S309) may be included.

In the full power operation step (S303), the air conditioning equipment may be started at full load. At this time, by operating the refrigerator, air conditioner, and FCU (Fan Coil Unit) at full output, the input variables can be obtained quickly, and accordingly, the station adapted to the target building and operating conditions by quickly reflecting the data according to the operation of the air conditioning equipment. The propagation neural network model can be obtained quickly and more accurately.

In the input variable storage step (S304), the operation start time, the room temperature, the outside temperature, the indoor temperature gradient, and the outdoor temperature gradient may be stored at the start point of operation. In the arrival time storage step (S305), it is possible to store the arrival time for reaching the set temperature from the operation start time. At this time, the arrival time under the given conditions can be reflected in obtaining the next precooling time.

At this time, the learning data including the indoor temperature, the outside temperature, the indoor temperature gradient, the outside temperature gradient, and the arrival time can be measured, calculated and / or stored, thereby reflecting the current operating conditions and conditions as the learning data. Train and determine neural network models.

In this case, 20 or more learning data may be collected through 20 days or more of schedule control operation to determine a neural network model adapted to the building to be applied. To this end, the neural network model loading step (S307) may be performed when 20 or more training materials are collected by determining the number of training materials (S306). Therefore, a meaningful neural network model can be constructed with a sufficient number of training materials.

In the neural network model loading step (S307), the neural network model optimized in the neural network model determination steps (S204 to S210) may be called.

In the neural network model training and determination step (S308, S309), it is possible to determine by learning the adapted backpropagation neural network model by inputting the learning data collected through driving. In the neural network model learning and determination step (S308, S309), the neural network model is trained by inputting training data collected through driving (S308), and the neural network model adapted to the target building is stored by storing the training information of the back propagation neural network model. You can decide. Thus, the neural network model can be used to obtain a more accurate precooling time for a given input variable.

4 is a system schematic diagram of an air conditioning facility to which a method for controlling reservation cooling of a building using an intelligent control algorithm by a neural network model according to an embodiment of the present invention is shown. FIG. 5 shows an embodiment of a control and information gathering schematic of the air conditioning facility of FIG. 4.

Referring to the drawings, the air conditioning equipment may include a refrigerator / boiler (R / B), a fan coil unit (FCU), a cooling tower, an air conditioner (AHU). In the drawing, EA (exhasut air) means external exhaust air, OA (outdoor air) means outdoor air, SA (supply air) means supply air, and RA (return air) is used for indoor exhaust. Mean air supplying some of the introduced air back to the room, INV represents the inverter.

In addition, an air handling unit (AHU) is a device that performs heating, cooling, and the like of the supplied air. HD means humidity sensor, FMS1 means flow sensor, FMS2 means air flow sensor, TD means temperature sensor, CV means control valve, and CO2 means carbon dioxide sensor.

In addition, HCWR (Hot Cold Water Return) means the recovery of cold water and hot water, HCWS (Hot Cold Water Supply) means the supply of cold water and hot water, AI (Analog Input) means analog input, AO ( Analog Output) means analog output, SA (Supply Air) means supply air, and SF (Supply Fan) means supply fan.

FIG. 9 schematically illustrates a structure of a back propagation neural network model in a method of pre-cooling control of a building using an intelligent control algorithm based on the neural network model of FIG. 1.

Referring to the drawings, the intelligent control algorithm may be used to optimize the various control methods applied to the automatic building control system energy saving control algorithm to obtain the optimal energy saving effect. In particular, it can be applied to control the air conditioning system of the building automatic control system to save energy while maintaining the comfort of the indoor space in the building, and the corresponding algorithm is installed in the digital direct control (DDC). Can be performed.

Intelligent control algorithm is divided into two parts, vacancy time and re-real time in the building, and independent algorithm is applied according to the time domain to operate the air conditioning system. The control algorithm of the vacancy time may be composed of night driving, night air blowing control, and optimal starting, and the algorithm of re-real time may be composed of power saving operation, enthalpy control, and optimum stopping.

In order to implement this, an artificial neural network as shown in FIG. 1 may be used. The artificial neural network may be implemented by software and may be composed of neurons connected by a network.

The artificial neural network has a function of connecting a plurality of neurons, which are computation processing units, by weighting the outputs of other connected neurons according to the appropriate connection strength and calculating the output by its activation function. One or more neurons make up a layer, which is divided into three layers: the input layer, the hidden layer, and the output layer. The input of the artificial neural network is delivered to neurons in the input layer, and the output of neurons in the input layer is delivered to neurons in the hidden layer.

Finally, the output of the neurons in the hidden layer is delivered to the neurons in the output layer, and the output of the neurons in the output layer is the output of the artificial neural network. There is one input layer and one output layer, but there may be no hidden layer or one or more layers.

In the present invention, as shown in Figure 1 composed of four input layers of the outside temperature, room temperature, outside temperature gradient, room temperature gradient, and the output layer of the air conditioning equipment start time (pre-cooling time in this embodiment) or air conditioning equipment stop time Can be. In other words, the artificial neural network can be used to save energy by adjusting the time to start the air conditioning system before the start of the room, and to save the energy by adjusting the time to stop the air conditioning before the start of the room. An optimal stop can be used to make this possible.

In this case, when the pre-cooling time is a time to start cooling before the set time, the optimal neural network model may have one hidden layer and eight hidden layer nodes, the learning rate is 0.25, the moment is 0.90, and the bias is 1.0. Can be. Accordingly, the pre-cooling time can be obtained more accurately by the optimal neural network model. Therefore, by controlling the air conditioning equipment by an intelligent control algorithm using an optimal neural network model, it is possible to maintain the interior of the building in a comfortable environment and to save energy used for air conditioning.

For optimal start and stop, it is necessary to determine the optimum start time or stop time by repeated trial and error. To this end, the present invention is to adjust the internal coefficients of the artificial neural network by the repetitive back-propagation learning method to derive the optimum result.

Artificial neural network can be applied according to the original intention only if it is properly learned, and prior learning is required to apply it. Sufficient learning data is required for prior learning. The artificial neural network ultimately operates to calculate an appropriate output for any input based on a learning result composed of an input pattern and an output pattern pair, and then uses the calculated result for learning to optimize an internal coefficient.

Looking at the best start and best stop algorithms, the best start is to determine when the air conditioner starts up before the start of the room, and the room temperature will reach the set temperature closest to the room. This is to determine whether the room temperature is outside the lower limit of the set temperature close to the end of the occupancy at the time when the air conditioning equipment is stopped before the time is reached.

In an embodiment of the present invention, an algorithm for determining an optimal starting time and an optimal stopping time using a multilayer artificial neural network may be applied. First of all, learning data is required to train artificial neural networks, so it cannot be applied directly to a building from the beginning.

Therefore, data must be collected for learning artificial neural networks. For this, the value of the neural network input variable can be measured every minute. The input variable values include room temperature, outside temperature, outside temperature slope, and inside temperature slope, and the indoor temperature and outside temperature are measured at regular time, and the outside temperature slope and room temperature slope are measured at regular time. And it means the rate of change over time of the outside temperature.

Mathematical models also require a lot of simulation and experimentation to know the input and output relationships. Neural network models require less specialized knowledge or experiments to determine the relationship between input and output in modeling nonlinear multivariate systems such as air conditioning operating time determination.

In mathematical models, the use of many parameters can cause mathematical solutions. The neural network model does not cause significant performance degradation even when the learning-related variable values of the neural network deviate from the optimal value to some extent or in the case of noise signals in online learning.

When the mathematical model is applied to the real building, it is difficult to input the same as the real building to solve the solution, and it takes a lot of time to simulate, so it is difficult to model the relationship between the various variables. Since neural network model is trained based on prior experience data to model input / output relationship, no separate input data is required for the control target.

Once the neural network is trained in a given network, response can be improved because control input for a given target can be obtained in a very short time. The fact that the characteristics can be learned online while controlling the object is a characteristic of the control by the neural network, and a flexible control method can be implemented.

According to the present invention, the reservation cooling control method of a building using an intelligent control algorithm using the neural network model controls the air conditioning facilities by an intelligent control algorithm using the neural network model, thereby maintaining the interior of the building in a comfortable environment. Can save energy used for air conditioning.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is to be understood that the invention is not limited to the disclosed exemplary embodiments. It will be apparent to those skilled in the art that various modifications and variations can be made therein without departing from the spirit and scope of the invention. It will be understood by those skilled in the art that various changes and modifications may be made without departing from the spirit and scope of the present invention.

Claims (15)

In the pre-cooling control method of a building using an intelligent control algorithm by neural network model,
Importing a neural network model that has been learned and adapted using the room temperature, the room temperature gradient, the outside temperature, and the outside temperature gradient as input variables and precooling time as output variables;
Measuring a measurement variable including the indoor temperature and the outside temperature at a set time interval;
Calculating a calculation variable including the indoor temperature slope and the outside temperature slope from the measured variable;
Calculating the precooling time at a measurement time point by inputting the input variable to the neural network model; And
Calculating a difference between the precooling time and a target setting time and performing an operation according to a set operating procedure if the difference is within a set value;
Before the pre-cooling through the neural network model, the step of night purge the interior of the building by introducing the outside air, the night purging step,
Measuring room temperature, outside air temperature, indoor humidity, and outdoor humidity;
Calculating indoor air enthalpy and outdoor enthalpy using the measured indoor temperature, outside air temperature, indoor humidity, and outdoor humidity; And
And if the indoor air enthalpy is greater than a predetermined value than the outside air enthalpy, discharging the indoor air to the outside and introducing the external air into the room. delete delete delete The method of claim 1,
The method of pre-cooling control of a building using an intelligent control algorithm by a neural network model further comprising the step of removing noise by determining whether the measured variable is noise. The method of claim 1,
Building a learning database;
Determining an optimized backpropagation neural network model using the learning database;
Determining the neural network model adapted to the target building by learning the backpropagation neural network model by using the learning data collected through the operation according to the operating procedure in the air conditioning facility of the building to which the application is applied;
Operating an air conditioning system of the building to be applied according to the adapted neural network model; And
And re-learning the adapted neural network model for inputting the learning data collected through the operation of the air conditioning equipment, and updating the neural network model. The method according to claim 6,
Building the learning material database,
Establishing and combining levels of factors influencing each of said precooling times by design of experiments, and
Intelligent control by a neural network model comprising the step of collecting the learning data including at least one of the measurement variable, the calculation variable, and the precooling time and constructing the learning data database through electrothermal analysis using finite difference method Precooling control method of building using algorithm. The method according to claim 6,
Determining the neural network model adapted to the building to be applied by learning the neural network model,
Measuring the measurement variable at set time intervals;
Starting operation at full power of the air conditioning equipment installed in the application building;
Storing an operation start time, the measurement variable, and the calculation variable at an operation start time;
Storing an arrival time for reaching a set temperature from the operation start time; And
A method of pre-cooling control of a building using an intelligent control algorithm by a neural network model comprising inputting training data collected through driving to train an adaptive backpropagation neural network model. The method according to claim 6,
Reservation of a building using an intelligent control algorithm by a neural network model wherein the factor includes at least one of an outdoor condition, a floor area, a target temperature, a window area ratio, a shielding coefficient, a wall structure, an adjacent condition, an indoor heat capacity, and an indoor heating load. Cooling Control Method. The method according to claim 6,
A method of pre-cooling control of a building using an intelligent control algorithm by a neural network model having each of three factors. The method of claim 1,
The neural network model is determined by calculating an optimal value of learning rate, moment, bias, number of hidden layers, and number of hidden layer nodes,
The reservation cooling control method of a building using an intelligent control algorithm by a neural network model in which the optimal value of the neural network model is calculated through an error minimization technique. The method of claim 1,
The pre-cooling time is a time to start cooling before the set time,
And a neural network model having one hidden layer and eight hidden layer nodes, a learning rate of 0.25, a moment of 0.90, and a bias of 1.0. The method according to claim 6,
A method of pre-cooling control of a building using an intelligent control algorithm based on a neural network model which determines a neural network model adapted to a target building by operating the air conditioning facility through a schedule control according to a set operation procedure. Constructing a learning material database having an indoor temperature, an indoor temperature gradient, an outdoor temperature, and an outdoor temperature gradient as input variables and a precooling time as an output variable;
Determining an optimized neural network model using the learning database;
Determining the neural network model adapted to the target building by learning the neural network model using the learning data collected through the operation according to the operating procedure set in the air conditioning facility of the target building;
Night purging the interior of the building by introducing outside air prior to the reservation cooling through the neural network model;
Operating an air conditioning system of the building to be applied according to the adapted neural network model; And
Re-learning the adapted neural network model for inputting the learning data collected through operation of the air conditioning equipment, and updating the neural network model,
The night purging step,
Measuring room temperature, outside air temperature, indoor humidity, and outdoor humidity;
Calculating indoor air enthalpy and outdoor enthalpy using the measured indoor temperature, outside air temperature, indoor humidity, and outdoor humidity; And
And if the indoor air enthalpy is greater than a predetermined value than the outside air enthalpy, discharging the indoor air to the outside and introducing the external air into the room. delete

Images