KR20210143645A - Method and Apparatus for Managing Building Energy Based on Deep Learning - Google Patents

Abstract

Provided are a deep learning-based building energy management method and device. An embodiment of the present invention provides a building energy management method and device optimizing energy consumption in terms of cost by selecting one of a plurality of energy production methods, which can satisfy the predicted demand, after predicting the energy demand based on deep learning for reducing energy consumption and costs by integratively predicting air conditioning and heating energy demand and production in a large building.

Description

Method and Apparatus for Managing Building Energy Based on Deep Learning

The present disclosure relates to a deep learning-based building energy management method and apparatus. More specifically, in order to reduce consumption and cost by comprehensively predicting heating and cooling energy demand and production in large buildings, after predicting energy demand based on deep learning, multiple It relates to a building energy management method and apparatus for performing optimization of energy consumption in terms of cost by selecting one of the energy production methods of

The content described below merely provides background information related to the present invention and does not constitute the prior art.

Predicting the demand for building energy and efficient operation based on it has a very large impact on energy costs and the global environment. In the case of large office buildings, it is a recent trend to control power, various facilities, and heating and cooling based on BEMS (Building Energy Management System). Such BEMS not only enables automatic operation of buildings according to its utilization plan, but also minimizes energy consumption and cost. For this, methods to improve the mechanical and electrical efficiency of each facility and improvement of operation methods for them should be supported.

BEMS is capable of accumulating big data by recording output data of sensors installed in each facility and external environmental variables (external temperature, humidity, etc.) in a database in real time for a long period of time. Based on the massive big data accumulated in the BEMS database, various AI (Artificial Intelligence) technologies such as machine learning and deep neural network are used to predict the heating and cooling energy demand and reduce the consumption and cost. Various methods have been proposed to suggest possible driving conditions.

Most of these methods are based on a binary method of extracting an energy demand forecasting model from big data and determining the operation type based on the appropriate capacity and number of facilities to meet this demand. As a method applied to extracting a demand prediction model, a statistical technique based on traditional linear regression, an AI technique based on a deep neural network, etc. are used. In addition, machine learning techniques such as random forest regression are sometimes used.

Recently, there is also an example of reducing the amount of cooling energy used in a data center by applying a reinforcement learning technique (see Non-Reference 1). Data centers have clear energy demand and temperature control targets, and temperature response to energy input works well, so advanced AI techniques such as reinforcement learning are easy to apply. On the other hand, a large office building has a problem in that it is difficult to apply the AI technique collectively because it contains many environmental and operational constraints that make it difficult for the advanced AI technique to work well. For example, the existing method confirmed the possibility of energy saving by applying reinforcement learning to a data center where there is little change by external climate environment variables and building users. In contrast, conventional techniques based on reinforcement learning cannot be applied to large buildings because energy input and internal temperature fluctuate frequently according to various external disturbance factors.

Therefore, despite the various external factors of large buildings, it is necessary to consider a method for predicting the energy demand of large buildings based on deep learning technology and efficiently operating the buildings using them.

“Transforming Cooling Optimization for Green Data Center via Deep Reinforcement Learning”, Yuanlong Li, Yonggang Wen, KyleGuan, and Dacheng Tao, 2018.

In order to reduce consumption and cost by integrally predicting heating and cooling energy demand and production in a large building, the present disclosure predicts energy demand based on deep learning, and then provides a plurality of A main object is to provide a building energy management method and apparatus that optimizes energy consumption in terms of cost by selecting one of the energy production methods.

According to an embodiment of the present disclosure, in a building energy management method performed by a computing device, the process of obtaining first input data and second input data; predicting an energy demand by using a deep learning-based prediction model based on the first input data; selecting a plurality of possible actions for operating a plurality of air conditioners/heaters to satisfy the energy demand; calculating energy production for each of the plurality of possible actions; and calculating a cost for the energy production and comparing the cost of each of the plurality of possible actions, thereby determining a possible action that minimizes the cost as an optimal action. do.

According to another embodiment of the present disclosure, the energy management method further comprising: operating all or a part of the plurality of air conditioners/heaters based on the optimal action, and measuring energy consumption according to the operation provides

According to another embodiment of the present disclosure, an input unit for obtaining first input data and second input data; Predicting energy demand using a deep learning-based prediction model trained in advance based on the first input data, and a deep learning-based environmental risk prediction model trained in advance based on the second input data a demand forecasting unit that generates environmental pollution risk using Selecting a plurality of feasible actions for operating a plurality of air conditioners/heaters to satisfy the energy demand, obtaining energy production for each of the plurality of possible actions, and based on a pre-stored cost table, an optimization unit that calculates a cost for energy production and determines an optimal action based on a cost for each of the plurality of possible actions and the environmental stigma risk; And it provides a building energy management device comprising a production estimator for calculating the energy production for each of the plurality of possible actions.

According to another embodiment of the present disclosure, there is provided a computer program stored in a computer-readable recording medium to execute each step included in the building energy management method.

As described above, according to this embodiment, after predicting the energy demand for a large building based on deep learning, one of a plurality of energy production methods that can satisfy the predicted demand is selected to reduce energy consumption in terms of cost. By providing a building energy management method and apparatus for performing optimization, there is an effect that it becomes possible to reduce energy consumption and cost for a large building.

1 is a schematic block diagram of a building energy management apparatus according to an embodiment of the present disclosure.
2 is an exemplary diagram of a deep learning-based prediction model according to an embodiment of the present disclosure.
3 is a schematic illustration of a method for estimating energy production according to an embodiment of the present disclosure.
4 is a schematic illustration of a method for determining an optimal action according to an embodiment of the present disclosure.
5 is a flowchart for a building energy management method according to an embodiment of the present disclosure.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, embodiments of the present invention will be described in detail with reference to exemplary drawings. In adding reference numerals to the components of each drawing, it should be noted that the same components are given the same reference numerals as much as possible even though they are indicated on different drawings. In addition, in describing the present embodiments, if it is determined that a detailed description of a related well-known configuration or function may obscure the gist of the present embodiments, the detailed description thereof will be omitted.

Also, in describing the components of the present embodiments, terms such as first, second, A, B, (a), (b), etc. may be used. These terms are only for distinguishing the elements from other elements, and the essence, order, or order of the elements are not limited by the terms. Throughout the specification, when a part 'includes' or 'includes' a certain component, it means that other components may be further included, rather than excluding other components, unless otherwise stated. . In addition, the '... Terms such as 'unit' and 'module' mean a unit that processes at least one function or operation, which may be implemented as hardware or software or a combination of hardware and software.

DETAILED DESCRIPTION The detailed description set forth below in conjunction with the appended drawings is intended to describe exemplary embodiments of the present invention and is not intended to represent the only embodiments in which the present invention may be practiced.

This embodiment discloses a deep learning-based building energy building energy management method and apparatus. More specifically, in order to reduce consumption and cost by comprehensively predicting heating and cooling energy demand and production in a large building, after predicting energy demand based on deep learning, a plurality of To provide a building energy management method and apparatus for optimizing energy consumption in terms of cost by selecting one of the energy production methods.

Hereinafter, the cooling/heating unit is collectively referred to as a cooling unit, a heating unit, and a cooling/heating unit having all of the cooling/heating functions.

In addition, for convenience, it is assumed that cooling is required when the amount of energy required is positive, and heating is required when the amount of energy is negative.

1 is a schematic block diagram of a building energy management apparatus according to an embodiment of the present disclosure.

The building energy management apparatus 100 (hereinafter 'energy management apparatus') according to this embodiment predicts the energy demand by using a deep learning-based prediction model, and estimates the energy production according to a possible action. In addition, the energy management apparatus 100 determines an optimal action for optimizing energy production based on the cost. The energy management device 100 includes all or part of an input unit 102 , a demand predictor 104 , a production estimator 106 , and an optimizer 108 . Here, components included in the energy management apparatus 100 according to the present embodiment are not necessarily limited thereto. For example, the energy management apparatus 100 may additionally include a training unit (not shown) for training the predictive model, or may be implemented in a form that interworks with an external training unit.

The input unit 102 obtains input data and provides it to the demand prediction unit 104 side. As shown in Table 1, the input data may include all or part of the previous energy demand (which has the same meaning as the previous energy production or consumption), the outside temperature, the outside temperature in the morning, the time and the inlet temperature of the air conditioner. .

Here, the external temperature is a temperature at the time when the energy demand is predicted, and the outside temperature in the morning indicates the temperature at the start of the operation of the air conditioner/heater (eg, 9:00 am). To measure this temperature, the input 102 may include a sensor. The external temperature at times (t+1) and (t+2) after the prediction time t may be input based on, for example, a weather forecast for each time period. In addition, times (t-2), (t-1), and t indicate previous and current prediction times, and intervals between each prediction time may be set at a preset interval (eg, 30 minutes). The reason for using this time as input data is that, in the case of a large building, the amount of energy demand may vary according to time (eg, the operation of the air conditioner/heater may be limited at night). As shown in Table 1, 17 pieces of input data according to the present embodiment are not necessarily limited thereto.

The demand forecasting unit 104 predicts the energy demand at (t+1), (t+2), and (t+3) times by using a prediction model based on the input data. Time points (t+1), (t+2), and (t+3) at which energy demand is predicted are examples, and the frequency and interval of prediction points may vary depending on the size, season, cost, etc. of a building.

2 is an exemplary diagram of a deep learning-based prediction model according to an embodiment of the present disclosure.

As illustrated in FIG. 2 , the predictive model may be implemented as a deep neural network. The predictive model includes an input layer that receives input data, at least one hidden layer, and an output layer. Here, the input layer receives 17 pieces of input data as described above, but performs min-max normalization. In addition, the output layer outputs the amount of energy demand at (t+1), (t+2), and (t+3) times. The example of FIG. 2 assumes that the prediction model is a deep neural network, but is not necessarily limited thereto, and other types of deep learning-based techniques may be used.

On the other hand, the energy management apparatus 100 may learn the prediction of the energy demand amount with respect to the prediction model. To this end, the energy management apparatus 100 may allow the training unit to use the input data for learning and the target energy demand collected in advance. That is, the training unit may perform training on the prediction model based on the input data for learning and the target energy demand amount.

On the other hand, the input data for training is input to the input layer of the predictive model after maximum and minimum normalization is applied, and the amount of energy demand is inferred by the predictive model based on this.

The training unit according to this embodiment defines a loss function based on the difference between the energy demand amount inferred by the predictive model based on the input data for learning and the target energy demand amount, and reduces the loss function of the predictive model. You can update parameters.

The production estimator 106 calculates the energy production at (t+1), (t+2), and (t+3) times based on the analysis result of the energy produced by the cooling/heating unit. To this end, the production estimation unit 106 obtains a possible action from the optimization unit 108 . Here, the possible action represents a scenario for cooling or heating based on a combination of various air conditioners/heaters in a building, operating conditions of each air conditioner/heater, and a driving method of each air conditioner/heater. For example, if the weather suddenly becomes very cold and you have to turn on the heating very strongly in the morning, decide whether to use an electric system air conditioner, an oil stove, or a fireplace for firewood, and how many units to use. Should be. The selection of such an air conditioner/heater may be based on the amount of heating demand, fuel cost/electricity cost consumed, how quickly it is heated after operation, and the like. Therefore, the operation scenario for cooling or heating based on the determination of the cooling/heating machine and the combination of factors affecting it may be an action.

3 is a schematic illustration of a method for estimating energy production according to an embodiment of the present disclosure.

As described above, the production estimator 106 obtains a possible action from the optimizer 108 and calculates the energy production at (t+1), (t+2), (t+3) times. Calculate. In the example of FIG. 3, as air conditioners that can be used by the production estimator 106, AC (Absorption Chiller) and TC (Turbo Chiller, turbo chiller, STC is Small TC, LTC is Large TC) exist and , AC is present as a heater. In general, it is known that AC takes about 1 hour to output the rated capacity, and TC requires about 10 minutes.

Each air conditioner/heater has an analysis table including information on the energy produced representing possible energy production according to the operating state at (t-1), time t. In addition, a plurality of such analysis tables may exist based on weather, time, operating state (initial operation state, rated capacity output state, etc.). The production estimator 106 may calculate the energy production for possible actions based on this analysis table.

In addition, in the case of cooling, the production estimator 106 may calculate the energy production in consideration of the delay until the rated capacity is output after operation (eg, 1 hour in the case of AC, 10 minutes in the case of TC). For example, in the case where AC was not operated in the previous time, the production estimator 106 may calculate the energy production for cooling in consideration of TC first. In the case of heating, the production estimator 106 calculates the delay until the rated capacity is output after operation (eg, 1 hour in the case of AC), and the residual energy (eg, 30 minutes in the case of AC) generated after stopping the operation. Energy production can be calculated by taking this into account.

After the production estimation unit 106 calculates the energy production for one possible action, it is provided to the optimization unit 108 .

4 is a schematic illustration of a method for determining an optimal action according to an embodiment of the present disclosure.

The optimization unit 108 obtains the energy demand at (t+1), (t+2), and (t+3) from the demand prediction unit 104, and selects a possible action that can be applied to the obtained demand. It is provided to the production estimator 106 and obtains energy production at (t+1), (t+2), (t+3) for possible actions. When the energy demand is a value near 0 (for example, a value included in a preset range between -α and α (α is a positive real number)), the optimizer 108 may not operate the air conditioner/heater. , do not select a possible action. For example, spring or autumn may correspond to this case.

Based on the energy production obtained from the production estimator 106, as illustrated in FIG. 4, the optimizer 108 calculates the cost for one possible action using a cost table according to whether cooling or heating. can By repeating the process as described above to calculate and compare costs for a plurality of possible actions, the optimizer 108 may determine as an optimal action a possible action that minimizes costs according to energy production while satisfying energy demand. .

When operating the AC according to the optimal action, the optimizer 108 causes the AC to be activated for at least three time steps (eg, (t+1), (t+2), (t+3)). . This is because, while the time interval according to the present embodiment is set to 30 minutes, it is assumed that the AC is delayed by 1 hour until the rated output is reached.

On the other hand, when the signs of the energy demand at times (t+1), (t+2), and (t+3) obtained from the demand forecasting unit 104 are mixed, the optimizer 108 regards this as a temporary error. It can determine, stop the operation of the air conditioner/heater, or maintain the previous operating state.

When the energy demand at time (t+1), (t+2), and (t+3) acquired from the demand forecasting unit 104 is all 0, the optimization unit 108 performs the previous operation for the cooling/heating unit. state can be maintained.

The optimization unit 108 may operate the air conditioner/heater based on the optimal action, measure the energy consumption actually consumed by the air conditioner/heater, and then cause the production estimation unit 106 to store the measurement result.

Meanwhile, the energy management apparatus 100 may use the production estimation unit 106 as a database for the table and existing legacy data as described above.

As legacy data, a legacy capable action (eg, a possible action at time (t+1), (t+2), (t+3), hereinafter 'legacy action') may be stored. The legacy action can be later compared with a possible action (hereinafter, 'deep learning action') by the deep learning-based energy management method according to the present embodiment in terms of air conditioner operation, energy consumption, cost, and the like.

As another legacy data, energy production (eg, energy production at time (t+1), (t+2), (t+3)) for cooling or heating according to the legacy action may be stored. This can also be compared to deep learning actions in terms of energy consumption and cost.

An analysis table used for estimation of energy production (eg, energy production at time (t+1), (t+2), (t+3)) for cooling or heating according to the deep learning action may be stored. .

A table of costs for energy production (eg, costs at time (t+1), (t+2), (t+3)) may be stored. As such an expense table, a table for each season, month, day, and time period may be stored. On the other hand, deep learning actions and legacy actions can use the same cost table.

The production estimator 106 may store measured data on the amount of energy consumption actually consumed by the air conditioner/heater, and use it as input data of the prediction model in the next time step.

The device (not shown) on which the energy management apparatus 100 according to the present embodiment is mounted may be a programmable computer, and it is assumed that it is mounted on a server or a programmable computer having a computing capability equivalent to that of the server.

5 is a flowchart for a building energy management method according to an embodiment of the present disclosure.

The energy management device 100 acquires input data (S500). As shown in Table 1, the input data may include all or part of the previous energy demand (which has the same meaning as the previous energy production or consumption accordingly), the outside temperature, the outside temperature in the morning, time, and the inlet temperature of the air conditioner. .

The energy management device 100 predicts the energy demand by using a deep learning-based prediction model based on the input data (S502). The predictive model is implemented as a deep learning-based deep neural network, and can be trained in advance based on the input data for learning and the target energy demand.

The energy management device 100 selects a plurality of possible actions for operating a plurality of air conditioners/heaters in order to satisfy the energy demand (S504). Here, the possible action represents a scenario for cooling or heating based on a combination of selection of various air conditioners/heaters in a building, operating conditions of each air conditioner/heater, and an operation method of each air conditioner/heater.

The energy management apparatus 100 calculates the energy production for each of a plurality of possible actions (S506). Each air conditioner/heater has an analysis table containing information about the energy produced representing the possible energy production according to the operating conditions at times t-1, t. Based on this analysis table, the energy management apparatus 100 may calculate the energy production for the possible action.

The energy management apparatus 100 calculates the cost for energy production and compares the cost for each of a plurality of possible actions, and determines a possible action that minimizes the cost as an optimal action ( S508 ). A cost for one possible action may be calculated using a seasonal cost table according to whether cooling or heating is performed. By calculating and comparing costs for a plurality of possible actions, the energy management apparatus 100 may determine a possible action that minimizes costs according to energy production while satisfying energy demand as an optimal action.

The energy management apparatus 100 operates all or part of a plurality of air conditioners/heaters based on an optimal action, and measures energy consumption according to the operation ( S510 ). After the measured energy consumption is stored in the database, it can be used as input data of the predictive model at the next time step.

As described above, according to this embodiment, after predicting the energy demand for a large building based on deep learning, one of a plurality of energy production methods that can satisfy the predicted demand is selected to reduce energy consumption in terms of cost. By providing a building energy management method and apparatus for performing optimization, there is an effect that it becomes possible to reduce energy consumption and cost for a large building.

In another embodiment of the present disclosure, the energy management device 100 predicts energy demand and ancillary indicators using a predictive model and auxiliary deep learning models (hereinafter, 'auxiliary models'), Estimate energy production according to possible actions based on these energy demand and ancillary indicators. It also determines the optimal action to optimize energy production based on cost and ancillary metrics. Here, ancillary indicators include (i) facility indicators such as expected lifespan of air conditioners/heaters, failure probability, etc., (ii) environmental indicators such as environmental pollution risk, and (iii) number of occupants in the building, prediction of specific areas in the building It may include all or part of other indicators such as sales. Accordingly, auxiliary models may be used to estimate all or some of the ancillary indicators.

The input unit 102 may additionally obtain input data for auxiliary models.

Input data for estimating facility indicators include total operating time for each cooling/heating unit, operating time during operation, time elapsed since the last operation was completed, vibration data (when a noise/vibration sensor is installed in the rotating part of the cooling/heating unit), It may include lubricating oil temperature, motor input current, and the like.

Among other indicators, the input data for estimating the number of occupants in the building is the number of people extracted from each gate/escalator/elevator using CCTV/IR (Infrared) cameras, the number of people connected to the mobile phone base station at the site where the building is located, (office It can include the number of people entering the security gate (in the case of a building), the number of card transactions (in the case of a commercial building), and the number of vehicles entering and exiting the parking lot.

The estimated sales of a specific area within a building among other indicators may be obtained after being calculated from previous sales and sales plans of stores within the specific area.

Input data for estimating environmental indicators include carbon dioxide concentration, dust sensor data, noxious gas sensor data, outside air fine dust measurement value (or fine dust related weather forecast), airborne transmission risk alarm situation (eg, flu or Covid-19). etc.) and the like.

The input data for estimating the ancillary indicator is not necessarily limited to the above, and any potential data correlating with the ancillary indicator may be added as input data.

The demand forecasting unit 104 may include all or part of a predictive maintenance model, a personnel estimation model, and an environmental risk prediction model as auxiliary models in addition to the predictive model. Here, the predictive maintenance model uses the input data for estimating the facility index, predict indicators. The occupancy model predicts the number of occupants in the building at (t+1), (t+2), and (t+3) times using input data for estimating the number of occupants. In this case, the number of occupants may be predicted for each area in the building. The environmental risk prediction model predicts environmental indicators such as environmental pollution risk at (t+1), (t+2), and (t+3) times using input data for estimating environmental indicators. Accordingly, the demand forecasting unit 104 may generate an energy demand amount and ancillary indicators by using the predictive model and auxiliary models.

Auxiliary models including a predictive maintenance model, a personnel estimation model, and an environmental risk prediction model are implemented with a deep neural network, and each auxiliary model may include an input layer that accepts input data, at least one hidden layer, and an output layer. . In addition, although it is assumed that each auxiliary model is a deep neural network, it is not necessarily limited thereto, and other types of deep learning-based techniques may be used.

Meanwhile, the energy management apparatus 100 may train each of the auxiliary models. To this end, the energy management apparatus 100 may allow the training unit to use input data and target data for learning for each auxiliary model, collected in advance. That is, the training unit may perform training on each of the auxiliary models based on the input data for learning and the target data.

Meanwhile, in predicting the amount of energy demand, the demand forecasting unit 104 may include all or part of ancillary indicators generated by the auxiliary models in the input data for the predictive model. Accordingly, the predictive model may provide an amount of energy demand based on all or part of the input data illustrated in Table 1, and ancillary indicators.

The production estimator 106 calculates the energy production at (t+1), (t+2), (t+3) time for one possible action based on the analysis result of the energy produced by the cooling/heating unit do. As mentioned above, the possible action is a driving scenario for cooling or heating. As in the present embodiment, when an environmental indicator is used as one of the ancillary indicators, the possible action may reflect the damper open ratio of the air conditioner as one of the driving conditions. Here, the damper opening ratio is a factor controlling the mixing ratio of the outside air and the indoor air.

The optimization unit 108 obtains the energy demand amount and ancillary indices at (t+1), (t+2), and (t+3) from the demand forecasting unit 104, and uses the obtained demand quantity and the ancillary indices. A possible action that can be satisfied is selected and provided to the production estimator 106 side.

The optimization unit 108 may select possible actions in consideration of the damper opening ratio. Since the inflow ratio of indoor air to the air conditioner can be adjusted by adjusting the damper opening ratio according to the temperature difference between indoors and outdoors, the optimizer 108 can provide a possible action according to the damper opening ratio.

The optimizer 108 may select a possible action in consideration of the facility index. Based on the expected lifespan and failure probability of the air conditioner/heater, a possible action may be provided so that a plurality of air conditioners/heaters can be used evenly so that a specific air conditioner/heater is not intensively used in terms of predictive maintenance of facilities.

The optimization unit 108 may select a possible action in consideration of the number of occupants in the building among other indicators. When the number of occupants is large, a possible action of strongly operating cooling/heating may be provided. In addition, when there are many occupants of a specific area in the building, a possible action for strongly operating cooling/heating for this specific area may be provided.

The optimizer 108 may select a possible action in consideration of the expected sales of a specific area within the building among other indicators. When the expected sales of a specific area are expected to increase (eg, when a store in a specific area conducts a sale), a possible action for strongly operating air conditioning/heating for the specific area may be provided.

The optimization unit 108 obtains the energy production at (t+1), (t+2), (t+3) for the possible actions from the production estimation unit 106, and based on this, whether cooling or heating The cost for one possible action can be calculated using the cost table according to . In this case, the optimizer 108 may calculate the cost in consideration of the ancillary indices obtained from the demand predictor 104 . By repeating the process as described above to calculate and compare costs for a plurality of possible actions, the optimizer 108 selects the possible actions that minimize the cost according to the energy production while satisfying the energy demand and ancillary indicators. can be determined as

For example, in the process of determining the optimal action, the optimizer 108 may consider the energy production cost according to the damper opening ratio. When the temperature difference between indoor and outdoor is large, energy production cost can be reduced by adjusting the damper opening ratio to increase the indoor air inflow ratio to the air conditioner. On the other hand, since environmental indicators may be affected, the optimizer 108 may perform an appropriate compromise between reducing production costs and complying with environmental indicators in consideration of regulatory matters.

The optimizer 108 may select an optimal action in consideration of the facility index. Based on the expected lifespan and failure probability of the air conditioner/heater, the optimal action can be selected by compromising the predictive maintenance aspect of the facility and the production cost reduction.

The optimization unit 108 may determine the optimal action in consideration of the estimated sales of a specific area within the building among other indicators. When the expected sales of a specific area are expected to increase, even if the production cost is higher, the possible action of strongly operating the cooling/heating for the specific area may be determined as the optimal action.

Although it is described that each process is sequentially executed in each flowchart according to the present embodiment, the present invention is not limited thereto. In other words, since it may be applicable to change and execute the processes described in the flowchart or to execute one or more processes in parallel, the flowchart is not limited to a time-series order.

Various implementations of the systems and techniques described herein include digital electronic circuitry, integrated circuits, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), computer hardware, firmware, software, and/or combination can be realized. These various implementations may include being implemented in one or more computer programs executable on a programmable system. The programmable system includes at least one programmable processor (which may be a special purpose processor) coupled to receive data and instructions from, and transmit data and instructions to, a storage system, at least one input device, and at least one output device. or may be a general-purpose processor). Computer programs (also known as programs, software, software applications or code) contain instructions for a programmable processor and are stored on a "computer-readable recording medium".

The computer-readable recording medium includes all types of recording devices in which data readable by a computer system is stored. These computer-readable recording media are non-volatile or non-transitory, such as ROM, CD-ROM, magnetic tape, floppy disk, memory card, hard disk, magneto-optical disk, and storage device. media, and may further include transitory media such as carrier waves (eg, transmission over the Internet) and data transmission media. In addition, the computer-readable recording medium is distributed in network-connected computer systems, and computer-readable codes may be stored and executed in a distributed manner.

The above description is merely illustrative of the technical idea of this embodiment, and various modifications and variations will be possible by those skilled in the art to which this embodiment belongs without departing from the essential characteristics of the present embodiment. Accordingly, the present embodiments are intended to explain rather than limit the technical spirit of the present embodiment, and the scope of the technical spirit of the present embodiment is not limited by these embodiments. The protection scope of this embodiment should be interpreted by the following claims, and all technical ideas within the equivalent range should be interpreted as being included in the scope of the present embodiment.

100: building energy management device
102: input unit 104: demand forecasting unit
106: production estimation unit 108: optimization unit

Claims (14)

In the building energy management method performed by the computing device,
a process of obtaining first input data and second input data, wherein the first input data is a factor affecting the energy demand, and the second input data is a factor affecting the risk of environmental pollution;
predicting the energy demand by using a deep learning-based prediction model based on the first input data;
selecting a plurality of possible actions for operating a plurality of air conditioners/heaters to satisfy the energy demand;
calculating energy production for each of the plurality of possible actions; and
The process of calculating a cost for the energy production and determining an optimal action using the cost for each of the plurality of possible actions
Energy management method comprising a. According to claim 1,
The method of claim 1, further comprising: operating all or part of the plurality of air conditioners/heaters based on the optimal action, and measuring energy consumption according to the operation. According to claim 1,
The first input data is
Energy management method, characterized in that it includes all or a part of previous energy consumption, outside temperature, morning outside temperature, predicted time, and air conditioner inlet temperature. 4. The method of claim 3,
The external temperature is an external temperature at the predicted time, and the morning external temperature is an external temperature when the air conditioner/heater starts to operate. According to claim 1,
The predictive model is
An energy management method implemented as a deep neural network, characterized in that it is trained in advance based on input data for learning and a target energy demand amount. According to claim 1,
The possible actions are
and a scenario for cooling or heating based on a decision to use the plurality of air conditioners/heaters and a combination of operation methods for the plurality of air conditioners/heaters. According to claim 1,
The process of calculating the energy production is,
The energy management method, characterized in that the energy production is calculated based on an analysis table including information on the energy produced according to the operating states of the plurality of air conditioners/heaters at the previous time and the predicted time. According to claim 1,
The process of determining the optimal action is
The energy management method, characterized in that calculating the cost for the energy production according to each of the plurality of actions based on a cost table according to cooling or heating. According to claim 1,
The second input data is
Energy management method, characterized in that it includes all or part of carbon dioxide concentration, dust sensor data, harmful gas sensor data, external air fine dust measurement value or fine dust-related weather forecast, and airborne transmission risk alarm situation. According to claim 1,
The process of estimating the energy demand is,
Energy management method, characterized in that generating the environmental pollution risk using a deep learning-based environmental risk prediction model trained in advance based on the second input data. 11. The method of claim 10,
The process of determining the optimal action is
The energy management method, characterized in that determining the optimal action based on the cost of each of the plurality of possible actions and the environmental stigma risk. an input unit for acquiring first input data and second input data, wherein the first input data is a factor influencing the amount of energy demand, and the second input data is a factor influencing the risk of environmental pollution;
Predicting the energy demand using a deep learning-based prediction model trained in advance based on the first input data, and predicting environmental risk based on deep learning trained in advance based on the second input data a demand forecasting unit for generating the environmental pollution risk using a model;
Selecting a plurality of possible actions for operating a plurality of air conditioners/heaters to satisfy the energy demand, obtaining energy production for each of the plurality of possible actions, and based on a pre-stored cost table, an optimization unit that calculates a cost for energy production, and determines an optimal action based on a cost for each of the plurality of possible actions and the environmental stigma risk; and
A production estimator for calculating the energy production for each of the plurality of possible actions
Building energy management device comprising a. 13. The method of claim 12,
The optimization unit,
Based on the optimal action, all or part of the plurality of air conditioners/heaters are operated, and energy consumption according to the operation is measured. A computer program stored in a computer-readable recording medium to execute each step included in the building energy management method according to any one of claims 1 to 11.

Images