KR20200017604A - Energy performance prediction method and energy operation management system by using optimized physical learning model and machine learning methods - Google Patents

Abstract

The present invention relates to an energy performance prediction method using an optimized physical learning model and machine learning, and an energy operation management system including the same. More specifically, by using an optimized method of a physical model imitating the whole energy system for operating and managing a subject operation building, an optimized prediction in a state close to a real system is possible, and long-term artificial intelligence learning can be implemented using short-term input information.

Description

Energy performance prediction method and energy operation management system by using optimized physical learning model and machine learning methods}

The present invention relates to a method for predicting energy performance using physics learning models and machine learning, and a system for performing energy operation management using the same, and more particularly, to a physical model that mimics an entire energy system of an operating object. The present invention relates to an energy performance prediction method for constructing a physics learning model and performing energy performance prediction and operation management through machine learning, and an energy operation management system including the same.

In the operation of building objects such as houses, buildings and apartments, accurate energy consumption prediction in the entire energy system that operates and manages the total energy flow of the building objects is not only stable and economical operation of the energy system, but also at the national level. As it is a prerequisite for stabilizing power supply and supplying high-quality electricity, and closely related to electricity rates in the competitive electricity market, the demand forecast of energy consumption such as electricity is very demanded for the market participants under the competitive electricity market structure. It is important.

Accordingly, Korean Patent Publication No. 10-1779797 (Self-learning HAVC energy management system using big data and its driving method, 2017.09.13.) Shows that BIC defines an object state according to energy consumption factors in a space to be controlled. By building data and providing self-learning HAVC energy management from the optimal object state, the object state reflects the state of the controlled space and surrounding environment and successfully reaches the target object state. A self-learning HAVC energy management system using big data and a method of driving the same are known which increase the efficiency of optimal HAVC energy management by storing and referencing a path.

However, the energy management system utilizing artificial intelligence as described above has to perform a separate learning process for each element and the basic characteristics of machine learning, the time required for learning artificial intelligence for a long time, and actually In case of applying to the operation target building, the problem that abnormal driving result can be applied when learning with sufficient data is not achieved.

Korea Patent Publication No. 10-1779797 (Self-learning HAVC energy management system using big data and its driving method, 2017.09.13.)

The present invention has been made to solve the above problems, it is possible to optimize the prediction of the state close to the actual system, using the short-term input information to learn the optimal physics learning model and machine learning that can learn long-term artificial intelligence An energy performance prediction method and an energy operation management system including the same are provided.

In order to solve the above problems, the energy performance prediction method using the optimal physics learning model and machine learning of the present invention, the past data collection step (S100) of collecting the past operating data of the entire energy system of the operation object, optimization techniques Physical model learning step (S200) of learning the entire system physical model consisting of input information, output information and parameters according to the total energy system, using the complete system physical model completed learning, the long-term learning data Characterized in that it comprises a learning data generation step (S300) to generate and learning the machine learning model built on the basis of artificial intelligence using the long-term learning data (S400).

In this case, the overall system physical model is characterized by consisting of input information, output information and parameters at the overall level to integrate each component model in the total energy system by integrating physical equations.

)와 상기 전체시스템물리모델에서 계산된 출력정보(

)의 차이의 전체합이 최소가 되는 파라메터를 탐색하는 것을 특징으로 한다.In addition, the optimization technique measures the output information measured in the total energy system according to the past input information of a certain period (

) And output information calculated from the overall system physical model (

It is characterized by searching for a parameter in which the total sum of the differences is minimized.

) 및 상기 전체시스템물리모델에서 계산된 에너지소비량(

)을 각각의 상기 출력정보(

,

)로 하는 것을 특징으로 한다.In addition, the optimization technique uses the past standard weather information received in time series as the past input information, and the energy consumption measured in the total energy system according to the past input information (

) And energy consumption calculated from the overall system physical model (

) The respective output information (

,

It is characterized by that).

In this case, the output information may include at least one of energy consumption, room temperature, indoor pollution degree, heat storage temperature, and the state of charge of the power storage device.

In addition, the long-term learning data is generated by calculating the long-term output information using the complete system physical model, the long-term standard weather information, and the long-term input information.

In addition, the energy performance prediction method may further include a performance prediction step (S500) of calculating output information based on weather information and input information corresponding to a period to be predicted using the learned machine learning model.

In addition, the energy performance prediction method is the optimal control variable derivation step of deriving a control variable for operating the entire energy system according to the weather information and input information corresponding to the period to be predicted using the learned machine learning model ( S600) may be further included.

In addition, a local operating server that manages the entire energy system of the operating object, which is built using the method of predicting performance of the entire system model according to the above method, and a remote machine learning server performing wired and wireless communication with the local operating server. In the energy operation management system using the optimal physics learning model and machine learning, the local operation server and the remote machine learning server are constructed as a formal physics model composed of input information, output information and parameters by mimicking the entire energy system. It includes a system learning model constructed based on artificial intelligence by imitating the entire system physical model and the total energy system, the local operation server is the actual data to receive and store past operating data and past weather information from the total energy system The entire system learned through storage modules and optimization techniques And a physical model learning module for searching for and storing optimized parameters in the model, wherein the remote machine learning server receives the optimized parameters from the local operation server for long term learning data for learning the machine learning module. And a machine model learning module for searching for and storing parameters in the machine learning model trained using the learning data generation module for producing and storing the learning data and the long term learning data. The apparatus may further include a performance prediction module that receives the long-term learned parameter from a server and calculates output information according to input information of a period to be predicted using the machine learning model.

The remote machine learning server may further include a learning control module for searching for and storing an optimal control variable for controlling the entire energy system according to input information of a period to be predicted.

At this time, the local operation server is characterized in that for controlling the components of each of the total energy system by transmitting the optimum control variable received from the remote machine learning server to the local control device of the total energy system.

The present invention according to the above configuration, by building a physical model that mimics the entire energy system by utilizing the optimization technique using the measured value in the total energy system and the calculated value in the physical model, the optimized state of the state close to the actual system It has the advantage of being predictable.

In addition, it is possible to train the entire system physics model using short-term input information, and to produce long-term learning data for learning artificial intelligence using the completed system physics model. By producing long-term learning data by calculation without actually taking a long time, there is an advantage that the time for learning artificial intelligence for performance prediction and control of the entire energy system can be significantly reduced.

1 is a flowchart illustrating a performance prediction method according to an embodiment of the present invention.
Figure 2 is a flow chart showing a physical model learning step according to an embodiment of the present invention.
Figure 3 is an exemplary view for explaining the overall system physical model according to an embodiment of the present invention.
Figure 4 is an exemplary view for explaining the machine learning step according to an embodiment of the present invention.
Figure 5 is a block diagram showing an energy operation management system according to an embodiment of the present invention.
6 is a block diagram of an overall system physical model according to an embodiment of the present invention.
Figure 7 is an exemplary view for explaining various embodiments of the overall system physical model of the energy operation management system of the present invention.

As the present invention allows for various changes and numerous embodiments, particular embodiments will be illustrated in the drawings and described in detail. However, this is not intended to limit the present invention to specific embodiments, it should be understood to include all modifications, equivalents, and substitutes included in the spirit and scope of the present invention.

When a component is said to be "connected" or "connected" to another component, it may be directly connected to or connected to that other component, but it may be understood that other components may be present in the middle. Should be.

Unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art.

Terms such as those defined in the commonly used dictionaries should be construed as having meanings consistent with the meanings in the context of the related art, and shall not be construed in ideal or excessively formal meanings unless expressly defined in this application. Do not.

Hereinafter, the technical spirit of the present invention will be described in more detail with reference to the accompanying drawings.

The accompanying drawings are only examples as illustrated in order to describe the technical idea of the present invention in more detail, and thus the technical idea of the present invention is not limited to the accompanying drawings.

5 is a block diagram showing an energy operation management system 1000 according to an embodiment of the present invention, Figure 6 is a block diagram showing an overall system physical model according to an embodiment of the present invention, Figures 5 to Referring to Figure 6, the energy operation management system 1000 of the present invention is a remote operation to perform wired and wireless communication with the local operation server 200 and the local operation server 200 to manage the entire energy system 100 of the operation target object Machine learning server 300 may be configured to include.

The total energy system 100 is a configuration for controlling, observing and managing the input and output of integrated energy including a plurality of load devices, energy storage devices, energy production devices and energy consuming devices provided in the operation target building, respectively It may include a local control device 110 for controlling the devices and a plurality of sensor network 120 is installed in a specific area of the operation target building. In this case, the operation target building may be made of a building having a complex space such as a house, a building, an apartment, a factory, or a specific set or area formed by the aforementioned buildings, and various modifications may be made without departing from the spirit of the present invention. It will be possible. In addition, the entire energy system 100 may be connected to communicate with the local operation server 200 through wired or wireless communication, the local operation server 200 is provided in the interior of the operation object, or the entire target It may be provided in a specific place for the management of the energy system 100.

In addition, the local control device 110 of the total energy system 100 is connected to the heat production control device 111, the power storage control device 112 and the environmental control device 113 of the operation target, the local operation By receiving the control variable from the server 200 may control the heat production control device 111, power storage control device 112 and the environmental control device 113. In this case, the heat production control device 111 is a configuration for controlling a heat production device for producing heat energy, such as a solar collector, a boiler, and the heat storage device for storing the produced heat, the power storage control device 112 ) Is a configuration for controlling a power production device for producing electrical energy and a power storage device for storing the generated power, such as a solar panel, a generator, etc., the environmental control device is a lighting system, an air conditioning system and As a configuration for controlling various environmental devices for controlling the environment such as temperature, humidity, ventilation amount, air pollution degree of the operation target building, the local control device 110 from the local operation server 200 It receives a control variable for controlling each device to operate and manage the operation target building. In addition, the sensor network 120 may be configured in plurality according to the purpose of the object to be measured in a specific area of the operation object, the temperature, humidity, ventilation amount, indoor air pollution measurement sensor or each local of the specific area provided The control device 110 and the distribution unit for measuring the power consumption of the operation target building or the power consumption of the respective devices, the function to transfer the measured values to the local control device 110 or the local operating server 200 To perform.

In addition, the local operating server 200 and the remote machine learning server 300 mimics the entire energy system 100 to form an entire system physical model constructed of a formal physical model consisting of input information, output information and parameters ( 210, 310) and the machine learning model (220, 320) based on artificial intelligence by mimicking the entire energy system, respectively, wherein the local operating server 200 and remote machine learning server 300 It may be configured to include a physical model learning module (240, 340) and machine learning model learning module (250, 350) for learning the entire system physical models (210, 310) and machine learning models (220, 320) have.

At this time, the energy operation management system 1000 of the present invention, the local operation server 200 is a past operation from the sensor network 120 or the local control device 110 including various sensors installed in a specific area of the total energy system It may further include an actual data storage module 230 for receiving and storing data and past weather information, wherein the physical model learning module 210 of the local operation server 200 is the overall system physical model through an optimization technique The optimized parameter may be searched for at 210 and the searched optimized parameter may be stored. In addition, the remote machine learning server 300 receives the optimized parameters from the local operation server 200, the learning data generation module for producing and storing long-term learning data for learning the machine learning module 320 ( The machine learning model learning module 350 of the remote machine learning server 300 further searches for and stores the optimized parameters in the machine learning model by using the long-term learning data. The operating server 200 receives the learned parameters using the long-term generated data from the remote machine learning server 300 to operate, control and manage the entire energy system 100, and data stored at this time These data may be stored in a database by storing time series data having respective characteristic values at predetermined time intervals.

Energy operation management system 1000 of the present invention according to the above-described configuration is provided in the operating object, the local operating server 200 and the remote machine learning server 300 for managing the entire energy system 100 of the operating object ) Performs wired and wireless communication, and the learning and optimization according to the short term input information of the overall system physical model is performed by the local operation server 200, and the local control apparatus Learning control for optimal control 110 is performed by the remote machine learning server 300 provided inside or outside the operation object, thereby reducing the performance and load of the local operation server 200, and relatively In the remote machine learning server 300 having a high performance configuration, by producing a large amount of data of the machine learning model 320 and performing long-term learning, And it has the advantage that you do not collect data to measure long-term study of the existing machine learning model.

The local operation server 200 further includes a performance prediction module 270 that calculates output information according to input information corresponding to a period to be predicted using a learned machine learning model. The output information may be calculated through the machine learning model 220 trained using the parameters of the long term learning received from the 300 to predict the performance of the entire energy system.

In addition, the remote machine learning server may further include a learning control module 380 for searching for and storing optimal control variables for controlling the entire energy system 100 according to input information of a period to be predicted. At this time, the optimum control variable is defined as a variable value for the set value of each control device of the system, and includes the room set temperature, cold and hot water tank set temperature, the power storage device set value, the room environmental control device set value, etc. The optimum control variable means a combination of optimum values for respective set values over time. That is, the indoor set temperature is dynamically changed according to the optimum set temperature value over time, and the entire energy system can be operated in an optimal state in which the objective function is minimized through the optimum change. In addition, as the learning control module 380 learns to search for the optimal control variable through machine learning using the machine learning model 320, the learning control module 380 is preferably performed in the remote machine learning server 300. In this case, the local operation server 200 transfers the optimum control variable received from the remote machine learning server 300 to the local control device 110 of the total energy system 100 to configure each component of the total energy system. Can be controlled.

The overall system physical model 210 is constructed by mimicking the entire energy system of the operation target building, and consists of a mathematical model consisting of input information (X), output information (Y) and parameters (p ＊). And includes a load user schedule model 211, an energy load model 212, an energy storage model 213, an energy exchanger model 214, an energy production model 215, and an energy regulator model 216. It is preferable to establish an integrated model consisting of input information, output information and parameters at the overall level integrating each component model in the total energy system 100, wherein the input of each of the models It is desirable to establish that the flow of energy is integrally connected from the production of energy to the load so that the state values of each model can be calculated by mathematically connecting the output and the output. That is, the overall system physical model 210 consists of physical models for each component, and each physical model consists of a formal connection between mutual input variables and output variables. At this time, each component is composed of input, parameter, and output variable, and the formula constituting the physics model is a mathematical model that applies a physical conservation law according to the law of mass conservation and the law of energy conservation. It is preferably expressed in the form of an equation. In addition, the parameter in the physical model for each component is unknown, and the parameter is searched to minimize the difference between the measured and calculated values for the input and output variables of each component model. After constructing the entire system model, preferably, each component is searched at the same time so as to search all the parameters of the entire system model at the same time to minimize the difference between the measured and calculated values for the input and output variables of the entire system model. There is an advantage that the measured values for input and output variables in between are not needed.

The machine learning model 220 is a learning model established based on artificial intelligence that performs prediction based on attributes learned through training data. The machine learning model 220 may be established using various artificial intelligence learning techniques, preferably in a time series. A cyclic artificial neural network (RNN) model capable of processing data having time-varying characteristics of the input information X may be established based on a database of stored input information X.

That is, the overall system physical models 210 and 310 in the energy operation management system 1000 may be utilized in various ways, as shown in FIG. 7, and preferably (a) of FIG. The entire system physical model 210 is trained using short-term actual data measured in the total energy system 100, and the long-term learning data generated using the trained total system physical model 210 is used. After learning the machine learning module 220, by predicting the optimized performance of the learned machine learning module 220, it is possible to optimize the operation of the entire energy system 100 of the operation target, As shown in b), learning control for deriving an optimal control variable for controlling the local system 110 of the overall energy system 100 using the entire system physical model 210 learned using real data. To The total energy system 100 of the operating object can be controlled by learning, and in addition, as shown in FIG. 7C, optimization of the entire system physical model 210 learned using actual data is performed. Learning control may be performed to derive an optimal control variable for controlling the local control apparatus 110 using the learned data.

The energy operation management system of the present invention according to the above configuration, using the overall system physical model 210 that physically modeled the total energy system 100 of the operation target, and the short-term past weather information of the total energy system and The optimized data can be predicted to be close to the operating object by learning database-based historical data as input information (X) and output information (Y), and receive the input information (X) for a long time. By generating long-term learning data for learning the machine learning model 220 using the output information (Y) and parameters calculated in the overall system model 210, the time for artificial intelligence learning can be significantly reduced. There is an advantage. In this case, the optimization method and the performance prediction method of the present invention will be described in detail with reference to the accompanying drawings.

FIG. 1 is a flowchart illustrating a performance prediction method according to an embodiment of the present invention. Referring to FIG. 1, an energy performance prediction method using an optimal physics learning model and machine learning according to the present invention may include a historical data collection step (S100). , Physical model learning step (S200), learning data generation step (S300), machine learning step (S400), performance prediction step (S500) and the optimal control variable derivation step (S600) is performed, as shown in the figure The order of the detailed configuration and arrangement of the components is not limited, but may be implemented and implemented in various embodiments described below.

In one aspect of the present invention, the energy performance prediction method includes input information (X) according to the entire energy system by using a past data collection step (S100) and an optimization technique of collecting past operation data of the entire energy system of the operation target. Learning data generation step of generating a long-term learning data using the physical model learning step (S200) for learning the entire system physical model consisting of the output information (Y) and parameters, and the complete system physical model completed learning ( S300) and a machine learning step (S400) for learning a machine learning model built on the basis of artificial intelligence using the long-term learning data, wherein the past data collection step (S100) is performed. Actual weather data and energy consumption of a certain period of time in the total energy system 100 of the operation target actual data of the local operation server 200 As a step of storing the data in the storage module 230 and time-series database, the measured values measured from the sensor network 120 or the local control device 110 may be stored, and the measured values are stored in time series. By the database, it can be used in the overall system physical model 210.

2 is a flowchart illustrating a physical model learning step according to an embodiment of the present invention, Figure 3 is an exemplary view for explaining an optimization technique according to an embodiment of the present invention, with reference to Figures 2 to 3 , The physical model learning step (S200) is a step for learning the overall system model 210 by using an optimization technique, more specifically the input information (X), output information (Y) and according to the total energy system Complete system model building step (S210) of establishing a mathematical physical model consisting of the parameter (p), calculating the parameters of the entire system model according to the optimization technique (S220), the calculated parameters of the entire system model storage module ( And a storage step (S230) of storing and storing the data in the database 260, wherein the entire system model 210 includes historical data stored in the measured data storage module 250 and the meteorological tablet. Receive and perform an optimization technique, wherein the calculating step (S220) and the storing step (S230) input and output database input information and output information in time series, calculate parameters according to them, and store the calculated data. do. That is, it is possible to perform learning for the total time n of the input data according to the unit time t inputted in time series, and to search for and determine a parameter for optimizing output information according to the total time of the input data. (S240-S250)

)와 상기 전체시스템모델에서 계산된 단기간의 출력정보(

)의 차이의 전체합이 최소가 되는 파라메터를 탐색하는 것을 특징으로 하며, 더욱 자세하게는 도 3의 (b)에 도시된 바와 같이, 상기 전체시스템모델(210)에 상기 전체에너지시스템(100)의 과거데이터에 따른 입력정보(X)를 입력하여 계산된 출력정보(

)와 상기 전체시스템모델(210)에 입력된 입력정보와 동일한 과거의 입력정보(X)에 따른 상기 전체에너지시스템(100)의 출력정보(

)가 하기의 제1목적함수(J)를 만족하는 파라메터를 탐색하여 상기 전체시스템모델(210)을 학습시키는 것을 특징으로 한다.At this time, the optimization method is a short-term output information measured in the total energy system according to the short-term past input information (X) (

) And short-term output information calculated from the whole system model (

It is characterized in that the search for a parameter that the total sum of the difference is the minimum, and more specifically, as shown in Figure 3 (b), the overall system model 210 of the total energy system 100 Output information calculated by inputting input information (X) according to historical data (

) And the output information of the total energy system 100 according to the past input information (X) which is the same as the input information input to the whole system model 210 (

) To learn the entire system model 210 by searching for a parameter that satisfies the following first objective function (J).

First objective function:

: 단기간의 출력정보(

)의 기간을 의미하며, 상기 전체에너지시스템(100) 또는 로컬제어장치(110)으로부터 측정된 단기간의 출력정보(

)와 단기간의 기상정보 및 입력정보를 이용하여 전체시스템물리모델의 파라메터에 대한 최적의 값을 결정하는 과정을 상기 전체시스템물리모델의 학습과정이라고 하며, 도 3의 (a)에 도시된 바와 같이, 종래의 물리모델 시스템은 상기 전체에너지시스템을 구성하는 복수의 구성요소들 각각에 대해서 학습을 수행함에 따라, 종래의 물리모델 시스템의 주요 학습용 데이터(

)는 각각의 구성요소가 구비된 내부온도(Tz), 목표로하는 내부온도(Tst) 및 각각의 구성요소별 에너지소모량등 구성요소가 증가함에 따라 학습되어야 하는 데이터가 증가함에 따라, 그 학습 효율이 저하되며, 학습을 위해 측정되어야 하는 데이터 또한 증가하여, 학습에 장기간의 시간이 소요된다는 단점이 있다. 그러나 상술한 바와 같이 본 발명은 각각의 구성요소들을 통합하는 전체레벨에서의 물리모델을 구축함으로써, 학습용 데이터를 현저하게 감소시킴으로써, 각각의 구성요소들의 학습을 위한 각각의 구성요소들의 입출입 데이터를 필요로 하기 않기 때문에 상기 전체에너지시스템의 전체레벨에서의 입력 및 출력 데이터만으로 학습이 수행하능하며, 이에 따라 학습에 필요한 시간을 현저하게 감소시킬 수 있는 장점이 있다.From here,

: Short term output information

And a short term output information measured from the total energy system 100 or the local controller 110.

) And the process of determining the optimal value for the parameters of the overall system physical model using short-term weather information and input information is called the learning process of the overall system physical model, as shown in FIG. As the conventional physical model system learns each of a plurality of components constituting the entire energy system, the main training data of the conventional physical model system (

) Is the learning efficiency as the data to be learned increases as the component increases, such as the internal temperature (Tz) provided with each component, the target internal temperature (Tst), and the energy consumption of each component. This is deteriorated, and the data to be measured for learning also increases, so that learning takes a long time. However, as described above, the present invention requires the entry / exit data of each component for learning of each component by building a physical model at the overall level integrating the respective components, thereby significantly reducing the training data. Since the learning can be performed only by the input and output data at the entire level of the total energy system, there is an advantage that can significantly reduce the time required for learning.

) 및 상기 전체시스템모델(210)에서 계산된 에너지소비량(

)으로 하여, 전체에너지시스템(100) 레벨에서의 통합적인 최적화 및 학습을 수행함으로써 하위 레벨에서의 각각의 장치 및 구역에서의 학습에 필요한 시간을 현저하게 감소시킬 수 있으며 이때, 상기 제1목적함수는 하기와 같다.In this case, as shown in (b) of Figure 3, when calculating the output information from the energy production, consumption, storage, load devices according to the measured values of the sensors installed in a specific area in the total energy system of the operation target building, respectively In addition, there is a problem in that the learning performance is lowered and the uncertainty of the prediction in the entire energy system is increased by calculating the number of very complex cases. Therefore, in the optimization technique, data obtained by time-series database of past weather information including outside temperature, solar radiation, etc. according to a predetermined time of the historical data of the operation target building is input data (X) in the optimization technique. The output information (Y) is an energy consumption amount of the total energy system 100 at the total system level (

) And the energy consumption calculated in the overall system model 210 (

), It is possible to significantly reduce the time required for learning in each device and zone at a lower level by performing integrated optimization and learning at the entire energy system 100 level, wherein the first objective function Is as follows.

First objective function:

In addition, the output information (Y) includes at least one or more of energy consumption, room temperature, indoor pollution degree, heat storage temperature, and the state of charge of the power storage device, a plurality of output information (according to the weather information received in time series) The entire system model 210 can be trained using the second objective function J ′, which compares Y).

Second purpose function:

= 각 시간단계에서의 입력변수에 따른 전체에너지시스템에서 측정된 에너지소비량,

= 각 시간단계에서의 입력변수에 따른 전체시스템모델에서 계산된 에너지소비량,

= 각 시간단계에서의 입력변수에 따른 전체에너지시스템에서 측정된 실내온도,

= 각 시간단계에서의 입력변수에 따른 전체시스템모델에서 계산된 실내온도,

= 각 시간단계에서의 입력변수에 따른 전체에너지시스템에서 측정된 냉온수조온도,

= 각 시간단계에서의 입력변수에 따른 전체시스템모델에서 계산된 냉온수조온도,

= 각 시간단계에서의 입력변수에 따른 전체에너지시스템에서 측정된 에너지소비량,

= 각 시간단계에서의 입력변수에 따른 전체시스템모델에서 계산된 에너지상태량,

= 각 시간단계에서의 입력변수에 따른 전체에너지시스템에서 측정된 실내공기오염도,

= 각 시간단계에서의 입력변수에 따른 전체시스템모델에서 계산된 실내공기오염도), 이때 상기 에너지소비량, 실내온도, 냉온수조온도, 실내공기오염도및 에너지상태량은 상기 운영대상건물의 각각의 특정구역에 설치된 센서네트워크 또는 각 요소의 로컬제어장치(110)로부터의 측정값 및 상기 전체시스템모델에서의 각각의 계산값을 각기의 입출력이 연결된 물리적인 모델 수식으로부터 산출하여 전체시스템 레벨로 통합한 값으로써, 상기 전체시스템모델의 구성요소 간의 에너지 입출력은 전체시스템모델의 에너지밸런스와 설정온도를 만족하도록 하는 알고리즘으로 전체시스템모델의 내부변수로서 계산하여 결정된다.Where i = each time step, k = the coefficient of each time step over the total time used for learning,

= Energy consumption measured in the total energy system according to the input variables in each time step,

= Energy consumption calculated from the overall system model according to the input variables in each time step,

= Room temperature measured in the total energy system according to the input variables in each time step,

= Room temperature calculated from the entire system model according to the input variables at each time step,

= Hot and cold bath temperature measured in the entire energy system according to the input variables in each time step,

= Hot and cold bath temperature calculated from the overall system model according to the input variables at each time step,

= Energy consumption measured in the total energy system according to the input variables in each time step,

= Energy state quantity calculated in the overall system model according to the input variables in each time step,

= Indoor air pollution measured in the total energy system according to the input variables in each time step,

= Indoor air pollution degree calculated from the entire system model according to the input variables in each time step), wherein the energy consumption, room temperature, cold / hot water bath temperature, indoor air pollution and energy state amount are stored in each specific zone of the target building. The measured values from the installed sensor network or the local control device 110 of each element and the respective calculated values in the overall system model are calculated from the physical model equations to which the respective inputs and outputs are connected, and are integrated at the total system level. Energy input and output between the components of the overall system model is determined by calculating the internal variables of the overall system model as an algorithm to satisfy the energy balance and the set temperature of the entire system model.

4 is a flowchart illustrating a machine learning step according to an embodiment of the present invention, referring to FIGS. 1 and 4, the method of predicting performance of the entire energy system according to the present invention includes a long-term standard weather information and a long-term input information. After the learning data generation step (S300) of generating a long-term output information by inputting the complete system physical model completed learning, the machine learning model built on the basis of artificial intelligence using the trained whole system model 210 ( And a machine learning step (S400) of learning 220, wherein the long-term learning data generated in the data generation step (S300) is the entire system physical model and a long-term standard for completing the learning. Long term output information may be generated by using weather information and long term input information.

)를 이용하여 상기 기계학습모델을 학습하며, 이때 시계열적으로 입력된 단위시간(t)에 따라 입력받은 상기 장기간의 입력정보의 총 시간동안의 학습을 수행하고 입력받은 데이터의 총 시간에 따른 출력정보가 최적화되는 파라메터를 탐색하여 상기 기계학습모델학습모듈에 저장하여 데이터베이스화 할 수 있다.In addition, the machine learning step (S400) stores the generated long-term standard meteorological information and long-term input information in the learning data generation module 360 in accordance with the time series to make a database, the learning data generation module 360 Output information calculated by inputting the long-term standard weather information and input information stored in the total system model 210 learned in the physical model learning step (S200) (

Learning the machine learning model using;), where the learning is performed for the total time of the long-term input information received according to the unit time (t) inputted in time series and the output according to the total time of the received data. By searching for parameters for which information is optimized, the database can be stored in the machine learning model learning module.

)을 기 학습한 전체시스템물리모델의 입력정보로 하여 계산한 장기간의 출력정보(

)를 생성함으로써, 기존에 기계학습모델을 학습하기 위한 데이터를 장기간에 걸쳐서 측정하여 얻을 수 있었던 학습기간을 획기적으로 단축할 수 있다.At this time, the learning data generation step (S300) is a long-term standard weather information and a long-term input information (

Long-term output information () calculated as input information of the entire system physical model

), It is possible to drastically shorten the learning period obtained by measuring data for learning a machine learning model over a long period of time.

In addition, the energy performance prediction method of the present invention uses a performance prediction step (S500) of calculating output information according to input information corresponding to a period to be predicted using the learned machine learning model and the learned machine learning model. The method may further include an optimal control variable derivation step S600 of deriving a control variable for operating the entire energy system according to input information corresponding to a period to be predicted.

At this time, the optimum control variable is defined as a variable value for the set value of each control device of the system, and includes the room set temperature, cold and hot water tank set temperature, the power storage device set value, the room environmental control device set value, etc. The maximum control variable means a combination of optimum values for respective set values over time. That is, the indoor set temperature is dynamically changed according to the optimum set temperature value over time, and the entire energy system can be operated in an optimal state in which the objective function is minimized through the optimum change.

In addition, the optimal control variable derivation step (S600) may reinforce the learning of the overall system model 210 by collecting operation data in the total energy system 100 of the operation object according to the derived optimal control variable. . That is, the optimal control variable derived in the optimal control variable derivation step (S600) may be transmitted to the local control device 110 of the entire energy system 100 to control and manage the state of the operating object, wherein the optimal If necessary, the entire system model 210 is re-used by using the measured values of the entire energy system 100 or the local control device 110 managed by the control variable and the weather information according to the managed time. By learning, the prediction performance of the overall system model 210 and the machine learning model 220 may be gradually improved.

The present invention is not limited to the above-described embodiments, and the scope of application is not limited, and various modifications can be made without departing from the gist of the present invention as claimed in the claims.

1000: Energy Operation Management System
100: total energy system 110: local control device
120: sensor network
200: local operation server 300: remote machine learning server
210, 310: Physical system physical model 220, 320: Machine learning model
230: measured data storage module 240, 340: physical model learning module
250, 350: machine learning model learning module 270: performance prediction module
360: learning data generation module 380: learning control module

Claims (12)

A historical data collection step (S100) of collecting past operational data of the entire energy system of the operation object;
A physical model learning step of learning an entire system physical model consisting of input information, output information and parameters according to the entire energy system using an optimization technique (S200);
Learning data generation step (S300) for generating a long-term learning data using the complete system physical model completed learning; And
Machine learning step of learning a machine learning model built on the basis of artificial intelligence using the long-term learning data (S400);
Energy performance prediction method comprising a.
The method of claim 1,
The overall system physical model is an energy performance prediction method, characterized in that composed of input information, output information and parameters at the overall level to integrate each component model in the total energy system by integrating physical equations.
The method of claim 2,
The optimization technique measures output information measured in the total energy system according to past input information of a predetermined period.

) And output information calculated from the overall system physical model (

Energy performance prediction method characterized in that the search for a parameter that the total sum of the difference is minimized.
The method of claim 3,
The optimization technique uses the past weather information input in time series as the past input information.
Energy consumption measured in the total energy system according to the past input information (

) And energy consumption calculated from the overall system physical model (

) The respective output information (

,

The energy performance prediction method characterized by the above-mentioned.
The method of claim 4, wherein
The output information may include at least one of energy consumption, indoor temperature, indoor pollution degree, heat storage temperature, and state of charge of the power storage device.
The method of claim 3,
And the long term learning data is generated by calculating long term output information using the entire system physical model, long term standard weather information, and long term input information.
The method of claim 1,
The energy performance prediction method,
A performance prediction step (S500) of calculating output information based on weather information and input information corresponding to a period to be predicted using the learned machine learning model;
Energy performance prediction method of the entire energy system comprising a further.
The method of claim 1,
The energy performance prediction method,
An optimal control variable derivation step (S600) of deriving a control variable for operating the entire energy system according to weather information and input information corresponding to a period to be predicted using the learned machine learning model (S600);
Energy performance prediction method further comprising.
In the energy operation management system using an optimal physics learning model and machine learning, including a local operation server for managing the entire energy system of the operating object and a remote machine learning server performing wired and wireless communication with the local operation server,
The local operation server and remote machine learning server,
An overall system physical model constructed of a formal physical model composed of input information, output information and parameters by mimicking the entire energy system; And
A machine learning model constructed based on artificial intelligence by imitating the entire energy system;
Including,
The local operation server,
An actual data storage module for receiving and storing past operation data and past weather information from the entire energy system; And
A physical model learning module for searching and storing optimized parameters in the overall system physical model learned through an optimization technique;
More,
The remote machine learning server,
A learning data generation module receiving the optimized parameters from the local operation server to produce and store long term learning data for learning the machine learning module; And
A machine model learning module for searching and storing parameters in the machine learning model learned using the long term learning data;
Including more,
And said local operating server receives said long-term learned parameter from said remote machine learning server.
The method of claim 9,
The local operation server,
A performance prediction module for calculating output information according to input information of a period to be predicted using the learned machine learning model;
Energy operation management system, characterized in that it further comprises.
The method according to claim 9 or 10,
The remote machine learning server,
A learning control module for searching for and storing an optimal control variable for controlling the entire energy system according to input information of a period to be predicted;
Energy operation management system, characterized in that it further comprises.
The method of claim 11,
The local operation server,
And transmitting the optimum control variable received from the remote machine learning server to a local control device of the total energy system to control each device of the total energy system.

Images