KR102427294B1 - Apparatus for predicting demand for solar power based on weather forecasting and method therefor - Google Patents

Abstract

A device for forecasting demand for solar power generation based on weather forecasting is provided. The device includes an energy storage device for storing power generated by the photovoltaic generator, and a local load receiving power from at least one of the photovoltaic generator and the energy storage device.

Description

Apparatus for predicting demand for solar power based on weather forecasting and method therefor

The present invention relates to a demand forecasting technology for photovoltaic power generation, and more particularly, to an apparatus for predicting a demand for photovoltaic power generation based on weather forecasting and a method therefor.

Energy storage refers to the storage of energy using a device or a physical medium. The device used for this is called an accumulator, and the entire system in a wider range is called an Energy Storage System (ESS). Even small batteries used in households or small batteries used in electronic products can convert electrical energy into other energy types and store them, but such a small power storage device is not called an ESS. is called ESS. ESS can be installed and operated in the power system for power generation, transmission and distribution, and at consumers. used as a function. ESS can contribute to improving energy use efficiency, enhancing the utilization of new and renewable energy, and stabilizing the power supply system by storing electric energy when it is used less and supplying it when needed. An EMS may be provided to manage such an ESS. EMS refers to a company-wide energy management system that sets energy efficiency improvement goals and systematically and continuously promotes the management system according to certain procedures and techniques to achieve them.

Registered Korea Patent Publication No. 1779516 on September 12, 2017 (Name: Solar power demand management device in rural areas)

SUMMARY OF THE INVENTION It is an object of the present invention to provide an apparatus for forecasting demand for solar power generation based on weather forecast and a method therefor.

A device for forecasting demand for solar power generation based on weather prediction according to a preferred embodiment of the present invention for achieving the above object includes a communication unit for communication with a plurality of sensor devices belonging to a plurality of microgrids, and the When a report message including sensor data is received from the sensor device through the communication unit, a report frame and a sub report frame that have received the report message, a report frame assigned to the microgrid to which the sensor device belongs, and a report frame assigned to the sensor device It is checked whether the sub-report frames are the same, and as a result of the check, the report frame and the sub-report frame that have received the report message, the report frame assigned to the microgrid to which the sensor device belongs, and the sub-report frame assigned to the sensor device If the report frame is the same, the verification unit includes a verification unit for verifying the integrity of the sensor data by comparing the sensor data of the sensor device and other sensor data received in the same report frame through the verification model.

The device allocates a report frame that is repeated at a predetermined period to each of the plurality of microgrids, and receives a registration request message requesting registration from any one sensor device belonging to any one of the plurality of microgrids through the communication unit. , allocates any one sub report frame within the report frame assigned to the corresponding microgrid to the sensor device, and transmits a registration response message together with an indicator indicating the assigned sub report frame to the sensor device through the communication unit It further includes a register that

When an input vector including a plurality of sensor data vectors generated from sensor data of the verification target sensor device and sensor data of another sensor device received in the same report frame as the sensor data of the verification target sensor device is input, the verification model is , It is characterized in that output values representing the probability that the sensor data to be verified will succeed in integrity authentication and the probability that it will fail are calculated through a plurality of operations in which the weights learned between a plurality of layers are applied to the input vector.

The sensor data vector is a frame vector indicating a report frame allocated to the microgrid to which the sensor device belongs, a subframe vector indicating a sub report frame allocated to the sensor device, and a time vector indicating the time the sensor device measures power, current and temperature. and a measurement vector representing the measurement value measured by the sensor device for power, current, and temperature.

에서 상기

을 0으로 설정한 후, 학습용 입력벡터에 대한 상기 검증모델의 출력값과 분류 레이블의 차이인 분류 손실이 최소가 되도록 상기 검증모델의 가중치를 수정하는 분류 손실 최적화를 수행하고, 상기 손실함수에서 상기

을 0.5로 설정한 후, 학습용 입력벡터에 대한 상기 검증모델의 출력값과 분류 레이블의 차이인 분류 손실과 상기 검증모델의 완전연결층의 연산노드값인 은닉벡터와 은닉 레이블과의 차이인 은닉 손실을 포함하는 복합 손실이 최소가 되도록 상기 검증모델의 가중치를 수정하는 복합 손실 최적화를 수행하는 학습부를 더 포함한다. The device prepares an input vector for learning, and a loss function

above in

is set to 0, and classification loss optimization is performed by correcting the weight of the verification model so that the classification loss, which is the difference between the output value of the verification model and the classification label with respect to the input vector for training, is minimized, and in the loss function,

After setting to 0.5, the classification loss, which is the difference between the output value of the validation model and the classification label for the training input vector, and the hidden loss, which is the difference between the hidden vector and the hidden label, which are the computational node values of the fully connected layer of the validation model, The method further includes a learning unit that performs complex loss optimization that corrects the weight of the verification model so that the included complex loss is minimized.

는 하이퍼파라미터인 것을 특징으로 한다. Here, the Eintergrity is a loss function, the oi is the output value of the output layer of the verification model, the zi is the classification label corresponding to the output value, and the eij is the operation node value of the fully connected layer of the verification model. a vector, wherein hij is a hidden label corresponding to the hidden vector, i is an index corresponding to an output node of an output layer of the verification model, and j is an index corresponding to an operation node of a fully connected layer of the verification model. and said

is a hyperparameter.

A method for forecasting demand for solar power generation based on weather forecasting according to a preferred embodiment of the present invention for achieving the above object comprises the steps of: a communication unit receiving a report message including sensor data from a sensor device; The verification unit determines the integrity of the sensor data according to whether the report frame and sub report frame in which the report message is received and the report frame assigned to the microgrid to which the sensor device belongs and the sub report frame assigned to the sensor device are the same. In the step of authenticating by car, and when the verification unit succeeds in the first authentication, the sensor data of the sensor device is compared with the sensor data of another sensor device received in the assigned report frame through a verification model to ensure the integrity of the sensor data Including the step of performing secondary authentication for the.

The method includes the steps of, before receiving a report message including sensor data from the sensor device, a registration unit allocating a report frame that is repeated at a predetermined period for each of a plurality of microgrids; Allocating any one sub report frame within a report frame allocated to the corresponding microgrid for the sensor device when receiving a registration request message requesting registration from any one sensor device belonging to any one; The method further includes transmitting a registration response message to the sensor device together with an indicator indicating the allocated sub report frame.

In the step of performing the secondary authentication, the verification model includes a plurality of sensor data generated from sensor data of the verification target sensor device and sensor data of another sensor device received in the same report frame as the sensor data of the verification target sensor device, respectively. A step of receiving an input vector including a vector, and a probability that the sensor data to be verified succeeds and fails in integrity authentication through a plurality of operations in which the verification model applies a weight learned between a plurality of layers with respect to the input vector and calculating an output value representing

The sensor data vector is a frame vector indicating a report frame allocated to the microgrid to which the sensor device belongs, a subframe vector indicating a sub report frame allocated to the sensor device, and a time vector indicating the time the sensor device measures power, current and temperature. and a measurement vector indicating the measurement values measured by the sensor device for power, current, and temperature.

에서 상기

을 0으로 설정한 후, 학습용 입력벡터에 대한 상기 검증모델의 출력값과 분류 레이블의 차이인 분류 손실이 최소가 되도록 상기 검증모델의 가중치를 수정하는 분류 손실 최적화를 수행하는 단계와, 상기 학습부가 상기 손실함수에서 상기

을 0.5로 설정한 후, 학습용 입력벡터에 대한 상기 검증모델의 출력값과 분류 레이블의 차이인 분류 손실과 상기 검증모델의 완전연결층의 연산노드값인 은닉벡터와 은닉 레이블과의 차이인 은닉 손실을 포함하는 복합 손실이 최소가 되도록 상기 검증모델의 가중치를 수정하는 복합 손실 최적화를 수행하는 단계를 더 포함한다. In the method, before receiving the report message, the learning unit provides an input vector for learning, and the learning unit provides a loss function.

above in

is set to 0, and then performing classification loss optimization of correcting the weight of the verification model so that the classification loss that is the difference between the output value of the verification model and the classification label with respect to the input vector for training is minimized; In the loss function,

After setting to 0.5, the classification loss, which is the difference between the output value of the validation model and the classification label for the training input vector, and the hidden loss, which is the difference between the hidden vector and the hidden label, which are the computational node values of the fully connected layer of the validation model, The method further includes performing complex loss optimization by modifying the weight of the verification model so that the included complex loss is minimized.

는 하이퍼파라미터인 것을 특징으로 한다. Here, the Eintergrity is a loss function, the oi is the output value of the output layer of the verification model, the zi is the classification label corresponding to the output value, and the eij is the operation node value of the fully connected layer of the verification model. a vector, wherein hij is a hidden label corresponding to the hidden vector, i is an index corresponding to an output node of an output layer of the verification model, and j is an index corresponding to an operation node of a fully connected layer of the verification model. and said

is a hyperparameter.

An energy storage device for storing power produced by a photovoltaic generator according to a preferred embodiment of the present invention for achieving the above object, and a local power supply from at least one of the photovoltaic generator and the energy storage device A device for forecasting the demand for solar power generation based on weather forecasting in a microgrid including a load measures power, current and temperature for an energy storage device, a photovoltaic generator, and a local load through LoRa communication from a plurality of sensor devices A plurality of sensor data including one measurement value and measurement time are continuously received, and a plurality of sensor data continuously received from the weather information server, diary information about the diary of each measurement time, and a plurality of sensor data continuously input A communication unit for receiving weather prediction information, which is the predicted weather after a preset time from each measurement time, and power representing the power generated by the photovoltaic generator, each of which is generated based on the measurement value and the measurement time, and the weather information It includes a production vector, a power storage vector representing power stored by the energy storage device, a power consumption vector representing power consumed by the local load, a time vector representing the measurement time, and a weather vector representing diary information of the measurement time. while generating a plurality of state vectors arranged in chronological order, and generating a predicted diary vector indicating a predicted diary after a preset time from the time indicated by the time vector of the first state vector in time among the plurality of state vectors, , from the time indicated by the time vector of the first state vector in time among the plurality of state vectors through a plurality of operations in which the learned weights of the plurality of layers of the predictive model are applied to the plurality of state vectors and the prediction diary vector and a prediction unit for calculating predicted values of the amount of power generation of the photovoltaic generator after a preset time, the storage amount of the energy storage device, and the demand amount of the local load.

In the device, when the prediction unit calculates the predicted values of the generation amount, the storage amount, and the demand amount of each of the plurality of microgrids, each of the plurality of microgrids according to the generation amount, the storage amount, and the demand amount of each of the plurality of microgrids A control message for deriving an excess or shortfall of the amount of power, and supplying the excess or shortfall of the amount of power to the energy storage device of the corresponding microgrid through the communication unit, or supplying the excess of the amount of power according to the derived excess or deficiency of the amount of power through the communication unit. It further includes a management unit for transmitting.

The predictive model is a microgrid photovoltaic generator according to the passage of time and weather changes through a plurality of operations in which the learned weights of a plurality of layers of a plurality of stages are applied to a plurality of state vectors arranged in time order. A state transmission network that calculates a state feature vector representing a correlation characteristic with respect to the amount of power generation, the storage amount of the energy storage device, and the demand amount of a local load, and a state input vector including the predicted diary vector and the state feature vector calculated by the state transfer network are input Sunlight after a preset time from the time indicated by the time vector of the first state vector in time among the plurality of state vectors through a plurality of operations in which the received and learned weights of a plurality of layers are applied to the input state input vector It includes a prediction network that calculates the predicted value of the power generation amount of the generator, the storage amount of the energy storage device, and the demand amount of the local load.

The apparatus prepares learning data including a plurality of state vectors and a prediction diary vector arranged in time order, and a predetermined time after the time indicated by the time vector of the first state vector among the plurality of state vectors of the learning data. The generation amount of the solar generator, the storage amount of the energy storage device, and the demand amount of the local load are set as the prediction labels, the learning data is input to the prediction model, and the state transmission network of the prediction model is a plurality of stages of a plurality of stages for a plurality of state vectors. The state feature vector is calculated through a plurality of operations to which the unlearned weight of the layer is applied, the prediction network of the predictive model receives the state input vector including the state feature vector and the predicted diary vector, and the input state input The amount of power generated by the photovoltaic generator after a preset time from the time indicated by the time vector of the first state vector among the plurality of state vectors of the learning data through a plurality of operations in which the unlearned weights of a plurality of layers are applied to the vector , when the output value that is the predicted value for the storage amount of the energy storage device and the demand amount of the local load is calculated, the prediction loss optimization that corrects the weight of the prediction model so that the prediction loss of the loss function representing the difference between the prediction label and the output value of the output layer is minimized It further includes a learning unit to perform.

를 이용하여 예측 레이블과 상기 출력값의 차이인 예측 손실이 최소가 되도록 예측모델의 가중치를 수정하며, 상기 Eforecast는 손실함수이고, 상기 R은 상기 출력값이고, 상기 L은 상기 예측 레이블이고, 상기 i는 출력노드의 인덱스인 것을 특징으로 한다. The learning unit

The weight of the prediction model is modified so that the prediction loss, which is the difference between the prediction label and the output value, is minimized using It is characterized in that it is the index of the output node.

An energy storage device for storing power produced by a photovoltaic generator according to a preferred embodiment of the present invention for achieving the above object, and a local power supply from at least one of the photovoltaic generator and the energy storage device A method for forecasting the demand for solar power generation based on weather prediction in a microgrid including a load is a method for a communication unit to transmit power, current and temperature to an energy storage device, a photovoltaic generator and a local load through LoRa communication from a plurality of sensor devices. Continuously receiving a plurality of sensor data including a measurement value and measurement time measured by Receiving the weather forecast information, which is the predicted weather after a preset time from each measurement time of a plurality of sensor data input as , and the weather generated based on the measured value and the measurement time, and the weather information, respectively, by a prediction unit A power generation vector representing power produced by the photogenerator, a power storage vector representing power stored by the energy storage device, a power consumption vector representing power consumed by the local load, a time vector representing the measurement time, and the measurement time generating a plurality of state vectors arranged in chronological order while including a diary vector indicating diary information of Generating a predictive diary vector representing the predicted diary after a set time, and the predictor through a plurality of operations in which the learned weights of a plurality of layers of the predictive model are applied to the plurality of state vectors and the predictive diary vectors Calculating the predicted values of the power generation amount of the photovoltaic generator, the storage amount of the energy storage device, and the demand amount of the local load after a preset time from the time indicated by the time vector of the first state vector in time among the state vectors of do.

In the method, after the step of calculating the predicted value, when the prediction unit calculates the predicted values of the generation amount, the storage amount, and the demand amount of each of the plurality of microgrids, the management unit the generation amount of each of the plurality of microgrids, the storage amount and the The step of deriving an extra or insufficient amount of power of each of the plurality of microgrids according to the amount of demand; The method further includes transmitting a control message for controlling to supply or supply an insufficient amount of power.

The step of calculating the predicted value is performed through a plurality of operations in which the learned weights of a plurality of layers of a plurality of stages are applied to a plurality of state vectors arranged in chronological order by the state transmission network of the predictive model. Calculating a state feature vector representing a correlation characteristic with respect to the amount of generation of the photovoltaic generator of the microgrid according to the change, the storage amount of the energy storage device, and the demand amount of a local load; The state input vector including the state feature vector calculated by the state transmission network is received, and the first sequence number in time among the plurality of state vectors through a plurality of operations in which the learned weights of a plurality of layers are applied to the input state input vector. and calculating predicted values of the amount of power generation of the photovoltaic generator, the storage amount of the energy storage device, and the demand amount of the local load after a preset time from the time indicated by the time vector of the state vector of .

The method includes the steps of: before the continuously receiving the plurality of sensor data, the learning unit provides learning data including a plurality of state vectors and prediction diary vectors arranged in chronological order; Setting the generation amount of the photovoltaic generator after a preset time from the time indicated by the time vector of the first state vector among the state vectors, the storage amount of the energy storage device, and the demand amount of the local load as a prediction label, and the learning unit learning data inputting into the predictive model, and the state transmission network of the predictive model calculating the state feature vector through a plurality of operations to which unlearned weights of a plurality of layers of a plurality of stages are applied to a plurality of state vectors; A prediction network of a predictive model receives a state input vector including the state feature vector and the predictive diary vector, and learns through a plurality of operations in which unlearned weights of a plurality of layers are applied to the input state input vector. Calculating an output value that is a predicted value for the amount of power generation of the photovoltaic generator, the storage amount of the energy storage device, and the demand amount of the local load after a preset time from the time indicated by the time vector of the first state vector among the plurality of state vectors of the data; The method further includes performing, by the learning unit, prediction loss optimization of correcting the weight of the prediction model so that the prediction loss of the loss function representing the difference between the prediction label and the output value is minimized.

를 이용하여 예측 레이블과 상기 출력값의 차이인 예측 손실이 최소가 되도록 예측모델의 가중치를 수정하며, 상기 Eforecast는 손실함수이고, 상기 R은 상기 출력값이고, 상기 L은 상기 예측 레이블이고, 상기 i는 출력노드의 인덱스인 것을 특징으로 한다. The step of performing the prediction loss optimization is the learning unit Equation

The weight of the prediction model is modified so that the prediction loss, which is the difference between the prediction label and the output value, is minimized using It is characterized in that it is the index of the output node.

According to the present invention, demand can be predicted based on sensor data for photovoltaic power transmitted by the sensor device through LoRa communication, and energy according to photovoltaic power can be efficiently distributed and managed.

1 is a view for explaining the configuration of a power system including a plurality of photovoltaic-based microgrids according to an embodiment of the present invention.
2 is a view for explaining the configuration of a photovoltaic-based microgrid according to an embodiment of the present invention.
3 is a view for explaining the configuration of an energy storage device according to an embodiment of the present invention.
4 is a view for explaining the configuration of a battery container according to an embodiment of the present invention.
5 is a block diagram for explaining the configuration of a sensor device according to an embodiment of the present invention.
6 is a block diagram for explaining the configuration of an energy management device according to an embodiment of the present invention.
7 is a block diagram illustrating a detailed configuration of an energy management server control unit according to an embodiment of the present invention.
8 is a diagram for explaining a report frame and a sub report frame allocated to a microgrid and a sensor device according to an embodiment of the present invention.
9 is a view for explaining a verification model for verifying the integrity of sensor data according to an embodiment of the present invention.
10 is a diagram for explaining a fully connected layer and an output layer of a verification model for verifying the integrity of sensor data according to an embodiment of the present invention.
11 is a diagram for explaining a prediction model for predicting the demand of a local load according to an embodiment of the present invention.
12 is a diagram for explaining a state transmission network of a prediction model for demand prediction of a local load according to an embodiment of the present invention.
13 is a diagram for explaining the operation of a state transmission network of a prediction model for demand prediction of a local load according to an embodiment of the present invention.
14 is a diagram for explaining a prediction network of a prediction model for predicting the demand of a local load according to an embodiment of the present invention.
15 is a flowchart illustrating a method of training a verification model (VM) according to an embodiment of the present invention.
16 is an example of a vector space for explaining a method of training a verification model (VM) according to an embodiment of the present invention.
17 is a flowchart illustrating a method for learning a predictive model (FCM) according to an embodiment of the present invention.
18 is a flowchart illustrating a method of managing sensor data for photovoltaic power received through LoRa communication according to an embodiment of the present invention.
19 is a flowchart illustrating a method for predicting a demand for solar power generation using a sensor device performing LoRa communication.
20 is a flowchart illustrating a method for managing demand for solar power generation using a sensor device performing LoRa communication.

Prior to the detailed description of the present invention, the terms or words used in the present specification and claims described below should not be construed as being limited to their ordinary or dictionary meanings, and the inventors should develop their own inventions in the best way. It should be interpreted as meaning and concept consistent with the technical idea of the present invention based on the principle that it can be appropriately defined as a concept of a term for explanation. Accordingly, the embodiments described in this specification and the configurations shown in the drawings are only the most preferred embodiments of the present invention, and do not represent all the technical ideas of the present invention, so various equivalents that can replace them at the time of the present application It should be understood that there may be water and variations.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. In this case, it should be noted that the same components in the accompanying drawings are denoted by the same reference numerals as much as possible. In addition, detailed descriptions of well-known functions and configurations that may obscure the gist of the present invention will be omitted. For the same reason, some components are exaggerated, omitted, or schematically illustrated in the accompanying drawings, and the size of each component does not fully reflect the actual size.

First, a power system including a plurality of microgrids based on photovoltaic power according to an embodiment of the present invention will be described. 1 is a view for explaining the configuration of a power system including a plurality of photovoltaic-based microgrids according to an embodiment of the present invention. 2 is a view for explaining the configuration of a photovoltaic-based microgrid according to an embodiment of the present invention.

1, the power system according to an embodiment of the present invention includes a plurality of micro-grids (MG: micro-grid, MG1, MG2, ..., MGn) and a plurality of micro-grids (MG: MG1, MG2, . .., MGn), including an energy management server (EMS: Energy Management System, 100). Moreover, the power system of the present invention further includes a weather information server (500). The meteorological information server 500 is a server that provides information on the weather including the weather and temperature predicted through the Meteorological Agency's Forecaster's Analysis System and the like and the weather including the weather and temperature measured by the Meteorological Agency.

2, each of the plurality of microgrids (MG: MG1, MG2, ..., MGn) is a photovoltaic generator 200, an energy storage system (ESS: Energy Storage System, 300) and a local load 400 includes In particular, each of the plurality of microgrids (MG: MG1, MG2, ..., MGn) includes a plurality of measuring instruments (10: 10a, 10b, 10c, 10d, 10e) and a plurality of power converters (PCS: Power Conditioning System or DC /DC converter, 20: 20a, 20b, 20c).

The photovoltaic generator 200 uses a photovoltaic effect to convert light coming from the sun into electrical energy to generate electric power. The photovoltaic generator 200 basically provides the generated power, that is, the generated power to the load 400 . However, the power output of the photovoltaic generator 200 varies greatly depending on the weather and cannot be adjusted. Therefore, the photovoltaic generator 200 may provide the energy storage device 300 with excess power compared to the capacity of the local load 400 and store it.

The energy storage device 300 is for storing power generated by the photovoltaic generator 200 . In particular, the energy storage device 300 may store the excess generated power when the generator 10 over-produces power compared to the capacity of the local load 30 . In addition, the energy storage device 300 provides the stored power to the load 400 when the power of the local load 400 is required.

The power converter 20 generates and outputs an output voltage by step-down or step-up of an input voltage according to an operable voltage range of the output-side device. For example, when the power generated by the photovoltaic generator 200 is input, the voltage of the power is stepped down and output according to the operable voltage range of the energy storage device 300 or the local load 400 .

The plurality of sensor devices 10 are connected to the plurality of power converters 20 , the photovoltaic generator 200 and the energy storage device 300 , and the voltage, current, and energy storage device 300 output from the photovoltaic generator 200 . ) input or output voltage, current, temperature of the energy storage device 300 , voltage input to the load 400 , current, and voltage and current converted by the power converter 20 are measured. In addition, the sensor device 10 may transmit the measured voltage, current, and temperature to the energy management server 100 through RoLA communication.

The energy management server 100 is for managing the power system according to an embodiment of the present invention, that is, a plurality of microgrids (MG) based on photovoltaic power generation. In particular, the energy management server 100 receives information related to energy measured by the sensor device 10 through LoRa communication, receives weather related information from the weather information server 500, and the information received through the deep learning model. After analyzing , the demand among a plurality of microgrids (MG) can be adjusted according to the analysis result.

Next, a configuration of an energy storage system (ESS) 300 according to an embodiment of the present invention will be described. 3 is a view for explaining the configuration of an energy storage device according to an embodiment of the present invention. 4 is a view for explaining the configuration of a battery container according to an embodiment of the present invention.

Referring to FIG. 3 , the energy storage device 300 according to an embodiment of the present invention includes a battery container 310 , an air conditioner 320 , a communicator 330 , and a controller 340 . The battery container 310 , the air conditioner 320 , the communicator 330 , and the controller 340 may be installed in the enclosure 301 to be physically protected by the enclosure 301 .

Referring to FIG. 4 , the battery container 310 includes a plurality of unit storage devices for actually storing energy. In an embodiment of the present invention, the smallest unit storage device is a battery cell. The battery cell is a secondary battery, and may be, for example, a nickel-cadmium battery (NiCd), a nickel hydride battery (NiMH), a lithium ion battery (Li-ion), or a lithium ion polymer battery (Li-ion polymer). As shown, a plurality of battery cells form one battery module. In addition, a plurality of battery modules form one battery rack. And a plurality of battery racks form one battery container. A battery rack is generally a tower type in which a plurality of battery modules are stacked. However, the present invention is not limited thereto, and the form is irrelevant as long as it includes a plurality of battery modules. The battery container 310 is installed in the first space A1 separated from the air conditioner 320 and the controller 340 in the enclosure 10 through the partition wall PW.

Referring back to FIG. 3 , the air conditioner 320 is installed in the second space A2 separated from the battery container 310 in the enclosure 10 through the partition wall PW. The air conditioner 320 performs air conditioning of the first space A1 in which the battery container 310 is installed. That is, the air conditioner 30 supplies air for air conditioning through a pipe connected to the first space A1 in which the battery container 310 is installed, that is, air having at least one attribute of cold air, warm air, moisture and dry air. can

The communicator 330 is for communication with the energy management server 100 through LoRa communication. The communicator 330 may include a radio frequency (RF) transmitter (Tx) for up-converting and amplifying a frequency of a transmitted signal, and an RF receiver (Rx) for low-noise amplifying and down-converting a received signal. In addition, the communicator 330 may include a modem that modulates a transmitted signal and demodulates a received signal.

The controller 340 is for controlling the overall operation and function of the energy storage device 300. In particular, the controller 340 controls the charging and discharging of the energy storage device 300, that is, the battery container 310, An air conditioning function of the air conditioner 320 may be controlled. When the controller 340 receives a command such as charging, discharging, cell balancing and air conditioning (cooling) from the energy management server 100 through the communicator 330, charging, discharging and cell balancing of the energy storage device 300 is performed. or by controlling the air conditioner 320 to perform air conditioning (cooling). In particular, when the controller 340 receives a control message for controlling to supply the excess amount of electric power from the energy management server 100 through the communicator 330 according to the excess or shortfall of the amount of electric power and to supply the insufficient amount of electric power, the control message The battery container 310 is controlled to be charged or discharged to supply an excess of the amount of power or to supply a shortage of the amount of power according to the operation. The controller 340 may be a central processing unit (CPU), a micro controller unit (MCU), a digital signal processor (DSP), or the like.

Next, the configuration of the sensor device 10 according to the embodiment of the present invention will be described. 5 is a block diagram for explaining the configuration of a sensor device according to an embodiment of the present invention. 5, the sensor device 10 according to an embodiment of the present invention includes a voltage sensor 11, a current sensor 12, a plurality of temperature sensors 13, a converter (Analog-to Digital Converter, 14), It includes a signal processing unit 15 and a LoRa communication unit 16 .

The voltage sensor 11 is a sensor for measuring the output or input voltage of the connected equipment or devices 20 , 200 , 300 , and the current sensor 12 is an input or output of the connected equipment or devices 20 , 200 , 300 . It is a sensor for measuring current. The temperature sensor 13 is a sensor for measuring the temperature of the equipment or devices 20, 200, 300 connected in an optional configuration. In particular, the plurality of temperature sensors 13 of the fourth measuring instrument 10d may be dispersedly installed at a plurality of locations of the energy storage device 300 to individually measure the temperature of each of the plurality of installed locations. In addition, the plurality of temperature sensors 13 of the first measuring instrument 10a may be dispersedly installed at a plurality of positions of the photovoltaic generator 200 and separately measure the temperature of each of the plurality of installed positions.

The converter 14 converts the voltage, current, and temperature values measured by the voltage sensor 11, the current sensor 12, and the temperature sensor 13, which are analog signals, into digital signals, and then sends it to the signal processing unit 15. to provide.

The LoRa communication unit 16 is for communication with the energy management server 100 through the LoLa communication (RoLa). The LoRa communication unit 16 may include an RF (Radio Frequency) transmitter (Tx) for up-converting and amplifying the frequency of a transmitted signal, and an RF receiver (Rx) for low-noise amplifying a received signal and down-converting the frequency. . In addition, the LoRa communication unit 16 may include a modem for modulating a transmitted signal and demodulating a received signal.

The signal processing unit 15 is an identifier that can distinguish and identify the measurement target from other devices in the measured values of voltage, current, and temperature converted into digital signals, that is, the power converter 20, the photovoltaic generator 200, the energy An identifier for a measurement target such as the storage device 300 and a measurement time for measuring voltage, current, and temperature are given. In particular, in the case of the power converter 20, an identifier that can know the position to be distinguished from other power converters 20 is given. The signal processing unit 15 may include a memory of a predetermined capacity capable of temporarily storing measurement values of voltage, current, and temperature including a measurement target and measurement time in consideration of retransmission due to a transmission error or the like. The signal processing unit 15 transmits sensor data including measurement values of voltage, current, and temperature including the measurement target and measurement time to the energy management server 100 through the LoRa communication unit 16 .

On the other hand, the energy management server 100 allocates a different report frame to each of the plurality of microgrids, and to each of the plurality of sensor devices 10 belonging to the microgrid, different sub-frames within the report frame allocated to the microgrid. Allocate a report frame. Accordingly, when the signal processing unit 15 transmits the above-described sensor data through the LoRa communication unit 16 , it is transmitted to the energy management server 100 through the assigned report frame and the sub report frame.

Next, the configuration of the energy management server 100 according to an embodiment of the present invention will be described. 6 is a block diagram for explaining the configuration of an energy management device according to an embodiment of the present invention. Referring to FIG. 6 , the energy management server 100 according to an embodiment of the present invention includes a communication unit 110 , a storage unit 120 , and a control unit 130 .

The communication unit 110 is for communication with the sensor device 10 , the energy storage device 300 , and the weather information server 500 . The communication unit 110 performs communication for exchanging data including necessary information with the sensor device 10 and the energy storage device 300 through LoRa communication. Also, the communication unit 110 may communicate with the weather information server 500 according to an IP communication protocol. The communication unit 110 may receive, from the sensor device 10 , measured values of voltage, current, and temperature measured by the sensor device 10 and a measurement time. Also, the communication unit 110 may receive weather information from the weather information server 500 . The communication unit 110 transmits the measured values of voltage, current, and temperature, measurement time, and weather information such as weather and weather forecast to the controller 130 . In addition, when the communication unit 110 receives data from the control unit 130 , the communication unit 110 configures the received data into packets and transmits the received data. For example, the communication unit 110 may transmit a control command to the energy storage device 300 . These control commands are for the energy storage device 300 to execute air conditioning, air conditioning, charging, discharging, balancing, alarm, emergency stop, and the like. In particular, the control command may be to discharge or charge excess power or insufficient power.

The storage unit 120 serves to store programs and data necessary for the operation of the energy management server 100 . In particular, the storage unit 120 is data generated according to the operation of the energy management server 100, that is, for example, the measured weather and forecasted weather, power, voltage, temperature, measured values and measurement time, time vector, diary Vectors, power consumption vectors, power generation vectors, temperature vectors, and predicted voltages, currents, and temperatures may be stored. Various types of data stored in the storage unit 120 may be deleted, changed, or added according to a user's manipulation.

The control unit 130 may control the overall operation of the energy management server 100 and the signal flow between internal blocks of the energy management server 100 and perform a data processing function of processing data. In addition, the control unit 130 basically performs a role of controlling various functions of the energy management server (100). The controller 130 may include a central processing unit (CPU), a digital signal processor (DSP), and the like.

Then, the detailed configuration of the control unit 130 of the energy management server 100 will be described in more detail. 7 is a block diagram illustrating a detailed configuration of an energy management server control unit according to an embodiment of the present invention. 8 is a diagram for explaining a report frame and a sub report frame allocated to a microgrid and a sensor device according to an embodiment of the present invention. 9 is a view for explaining a verification model for verifying the integrity of sensor data according to an embodiment of the present invention. 10 is a diagram for explaining a fully connected layer and an output layer of a verification model for verifying the integrity of sensor data according to an embodiment of the present invention. 11 is a diagram for explaining a prediction model for predicting the demand of a local load according to an embodiment of the present invention. 12 is a diagram for explaining a state transmission network of a prediction model for demand prediction of a local load according to an embodiment of the present invention. 13 is a diagram for explaining the operation of a state transmission network of a prediction model for demand prediction of a local load according to an embodiment of the present invention. And FIG. 14 is a diagram for explaining a prediction network of a prediction model for demand prediction of a local load according to an embodiment of the present invention.

First, referring to FIG. 7 , the control unit 130 includes a registration unit 131 , a verification unit 132 , a prediction unit 133 , a learning unit 134 , and a management unit 135 .

The registration unit 131 allocates a report frame P to each of the plurality of microgrids, and allocates a sub report frame S to each of the plurality of sensor devices 10 belonging to each of the plurality of microgrids MG. it is for More specifically, the registration unit 131 first, a plurality of microgrids (MG: MG1, MG2, ..., MGn) for each of the report frame (P1, P2, ..., Pn) repeated at a predetermined cycle allocate In this state, the registration unit 131 requests registration from any one sensor device 10 belonging to any one of a plurality of microgrids (MG: MG1, MG2, ..., MGn) through the communication unit 110 . You can receive a registration request message. Then, as shown in FIG. 8 , the registration unit 131 reports a report allocated to the corresponding microgrid MG among the plurality of microgrids MG (MG1, MG2, ..., MGn) for the sensor device 10 . Any one sub report frame (one of S1, S2, S3, S4, S5) is allocated within the frame (eg, P3). Then, the registration unit 131 may transmit a registration response message to the corresponding sensor device 10 together with an indicator indicating the sub report frame allocated through the communication unit 110 .

The verification unit 132 is for performing an inspection on the integrity of the sensor data of the sensor device 10 . More specifically, when the verification unit 132 receives a report message including sensor data from the sensor device 10 through the communication unit 110 , the report frame P and the sub report frame ( It is checked whether S) and the report frame P assigned to the microgrid MG to which the sensor device 10 belongs and the sub report frame S assigned to the sensor device 10 are the same. According to this confirmation, the report frame (P) and the sub report frame (S) that received the report message, the report frame (P) and the sensor device (10) allocated to the microgrid (MG) to which the sensor device 10 belongs If the assigned sub report frame S is the same, the verification unit 132 verifies the integrity of the received sensor data by comparing the sensor data of the sensor device 10 with the sensor data of other sensor devices received in the same report frame. do. At this time, the verification unit 132 verifies the integrity of the sensor data through the verification model (VM). This verification model (VM) is a deep neural network (DNN) model in which learning is performed by deep learning. A deep neural network (DNN) model may typically be exemplified by a convolutional neural network (CNN).

First, the verification unit 132 receives a predetermined value from each of the sensor data of the verification target sensor device 10 and the sensor data of the other sensor device 10 received in the same report frame as the sensor data of the verification target sensor device 10 . A sensor data vector is generated by detecting the feature. Here, the sensor data vector is a frame vector indicating a report frame allocated to the microgrid to which the sensor apparatus 10 belongs, a subframe vector indicating a sub report frame allocated to the sensor apparatus 10, power, It includes a time vector indicating a time at which current and temperature are measured, and a measurement vector indicating a measurement value obtained by measuring power, current, and temperature by the sensor device 10 .

As an example, referring to FIGS. 2 and 8 , the third microgrid MG3 includes the first to fifth sensor devices 10a, 10b, 10c, 10d, and 10e, and the verification target is the fourth sensor device ( 10d) is assumed. Also, it is assumed that the third report frame P3 is allocated to the third microgrid MG3. Also, it is assumed that the fourth sensor device 10d to be verified has allocated the third sub report frame S3. The verification unit 132 includes the first sensor data of the fourth sensor device 10d, which is the verification target, and the third report frame P3, which is the same report frame as the sensor data of the fourth sensor device 10d, which is the verification target. A sensor data vector is generated by detecting a predetermined characteristic from each of the sensor data of the second, third, and fifth sensor devices 10a, 10b, 10c, and 10e. As a result, the verification unit 132 receives the sensor data vector from each of the sensor data of the first to fifth sensor devices 10a, 10b, 10c, 10d, and 10e received through the third report frame P3, that is, the first to the fifth sensor data vectors SV1, SV2, SV3, SV4, and SV5 are derived. Then, the input vector IV is generated by connecting the first to fifth sensor data vectors.

The measurement time and measurement value of sensor data transmitted by a plurality of sensor devices 10a, 10b, 10c, 10d, and 10e within the same microgrid within the same report frame are the photovoltaic generator 200, the energy storage device ( 300) and the local load 400, so it has a very high correlation. Therefore, according to the present invention, the learning unit 134 learns the correlation between the sensor data to the verification model (VM), and the verification unit 132 verifies the integrity of the sensor data through the learned verification model (VM). .

9 and 10 , the verification model (VM) sequentially includes a verification input layer (VIL), a convolutional layer (VCL), a verification pooling layer (VPL), and a fully connected layer. It includes a (verification fully-connected layer: VFL) and an output layer (verification output layer: VOL). Here, each of the convolution layer VCL, the pooling layer VPL, and the fully connected layer VFL may be two or more. The convolution layer (VCL) and the pooling layer (VPL) include at least one feature map (FM). The feature map FM is derived as a result of receiving a value applied with a weight and a threshold to the operation result of the previous layer, and performing an operation on the input value. These weights are applied through a filter or kernel W, which is a weight matrix of a predetermined size. In the embodiment of the present invention, the first filter W1 is used for the convolution operation of the convolutional layer VCL, and the second filter W2 is used for the pooling operation of the pooling layer VPL.

When an input vector (IV=[SV1, SV2, SV3, SV4, SV5]) is input to the input layer VIL, the convolution layer VCL performs a first filter on the input vector IV of the input layer VIL. At least one first feature map FM1 is derived by performing a convolution operation using (W1) and an operation by an activation function. Next, the pooling layer VPL performs a pooling or sub-sampling operation using the second filter W2 on at least one first feature map FM1 of the convolution layer VCL to obtain at least one A second feature map FM2 is derived.

The final connection layer VFL includes a plurality of operation nodes E1 to En. The plurality of operation nodes E1 to En of the final connection layer VFL are calculated using an activation function with respect to at least one second feature map FM2 of the pooling layer VPL. , e2, e3, ..., en]. Such a plurality of operation node values E[e1, e2, e3, ..., en] is called a hidden vector.

The output layer VOL includes two output nodes O1 and O2. Referring to FIG. 10 , each of the plurality of operation nodes E1 to En of the fully connected layer VFL is a channel (indicated by a dotted line) having a weight W, and the output nodes O1 and O2 of the output layer OL. ) is associated with In other words, the plurality of operation node values e1, e2, e3, ..., en of the plurality of operation nodes E1 to En are inputted to the output nodes O1 and O2, respectively. Accordingly, the output nodes O1 and O2 of the output layer VOL are calculated by the activation function for the plurality of operation node values e1, e2, e3, ..., en to which the weight is applied to the input layer VIL. ) to calculate the probability of success and failure of the integrity verification of the sensor data that is the verification target corresponding to the input vector.

Each of the two output nodes O1 and O2 of the output layer VOL corresponds to an integrity verification result, success and failure. That is, the first output node O1 corresponds to success, and the second output node O2 corresponds to failure. Accordingly, the output value of the first output node O1 is the probability that the verification target sensor data of the input vector input to the input layer VIL succeeds in integrity verification, and the output value of the second output node O2 is the input layer ( VIL) indicates the probability that the sensor data to be verified of the input vector will fail the integrity verification. For example, each of the output nodes O1, O1 has a weight w = [w1, w2, ... , wn] is applied, and the output value is calculated by taking the activation function F on the result. For example, if the output values of the first and second output nodes O1 and O2 are 0.086 and 0.914, respectively, the probability that the sensor data to be verified succeeds in integrity verification is 9%, and the probability that the sensor data to be verified fails the integrity verification This indicates that it is 91%. Output values of the first and second output nodes O1 and O2 of the output layer VOL become output values of the verification model VM. When the verification model VM outputs probabilities (0.086, 0.914), the verification unit 132 may determine that the verification target sensor data has failed integrity verification according to these probabilities (0.086, 0.914).

Activation functions used in the above-described convolutional layer (CL), final connection layer (FL) and output layer (OL) are Sigmoid, Hyperbolic tangent (tanh), Exponential Linear Unit (ELU), and ReLU. (Rectified Linear Unit), Leakly ReLU, Maxout, Minout, Softmax, etc. may be exemplified. Any one of these activation functions may be selected and applied to the convolutional layer CL, the final connection layer FL, and the output layer OL.

Again, referring to FIG. 7 , the prediction unit 133 calculates the amount of power generated by the photovoltaic generator 200 of the microgrid (MG) after a preset time from the present time, the amount of storage of the energy storage device 300 and the local load 400 . to predict the demand for For this, the prediction unit 133 uses a prediction model (FCM). The prediction method of the prediction unit 133 using such a prediction model (FCM) will be described in more detail as follows.

When the verification unit 132 successfully verifies the integrity of the sensor data, the prediction unit 133 receives the sensor data from the verification unit 132 . That is, the prediction unit 133 measures power, current, and temperature of the energy storage device 300 , the photovoltaic generator 200 , and the local load 400 from a plurality of sensor devices through the LoRa communication, and the measured values and measurements. It is possible to receive sensor data including time. In addition, the prediction unit 133 predicts the diary information of the diary of the measurement time of the sensor data from the weather information server 500 through the communication unit 110 and the diary of the time to be predicted after a preset time from the current time. Receive prediction information.

Then, the prediction unit 133 generates a state vector including a power generation vector, a power storage vector, a power consumption vector, a time vector, and a weather vector from the sensor data and weather information, and generates a predicted diary vector from the weather prediction information. do. The power production vector represents the measured power, current and temperature of the photovoltaic generator 200 , the power storage vector represents the measured power, current and temperature of the energy storage device 300 , and the power consumption vector represents the local load 400 . ) represents the measured power, current and temperature. The time vector represents a measurement time during which the sensor device 10 measures power, current, and temperature. The weather vector represents the weather observed at the measurement time when the sensor device 10 measures power, current, and temperature. In addition, the predicted diary vector represents a predicted diary of a time to be predicted, which is a preset time after the current time point.

Meanwhile, referring to FIG. 11 , the prediction model (FCM) includes two sub-networks (STN and EN), which include a state transmission network (STN) and a prediction network (FCN).

The state transmission network (STN) receives a state vector, and from the state vector, the amount of power generated by the photovoltaic generator 200 of the corresponding microgrid according to the passage of time and the change of weather, the amount of storage of the energy storage device 300, and the amount of demand for local load A state feature vector representing the correlation feature for (400) is derived.

The prediction network (FCN) receives the predicted weather vector and the state feature vector, and calculates the prediction result vector predicting the amount of power generation of the microgrid photovoltaic generator, the storage amount of the energy storage device, and the demand amount of the local load after a preset time. .

The state transfer network (STN) may be a recurrent neural network such as a recurrent neural network (RNN) or a long short-term memory (LTSM). The state propagation network (STN) includes an input layer (RIL: Recurrent Input Layer), a hidden layer (RHL: Recurrent Hidden Layer), and an output layer (ROL: Recurrent Output Layer). The input layer RIL, the hidden layer RHL, and the output layer ROL of the state transmission network STN may be divided into a plurality of stages St. As shown in FIG. 12 , in the embodiment of the present invention, it is assumed that the state transmission network (STN) consists of four stages S1 to S4.

The input layer IL includes a plurality of input cells IC1 , IC2 , IC3 , IC4 corresponding to the number of stages (eg, 4 ). Each of the plurality of input cells IC1 , IC2 , IC3 , and IC4 receives a plurality of state vectors X1 , X2 , X3 , and X4 generated in time order by the prediction unit 133 for each stage.

The hidden layer HL includes a plurality of hidden cells HC1, HC2, HC3, and HC4 corresponding to the number of stages. Each of the plurality of hidden cells HC1, HC2, HC3, and HC4 calculates a state value corresponding to the current stage by performing an operation on the state value to which the weight of the previous stage is applied and the input value to which the weight of the current stage is applied. .

The output layer ROL may include a plurality of output cells OC1, OC2, OC3, and OC4 corresponding to the number of stages, but may include only one output cell OC4 of the last stage S4. have. The output layer ROL receives a state value to which a weight is applied from the hidden layer RHL, and calculates an output value, ie, a state feature vector.

13, the hidden cell HCt of the t-th stage St receives the input value Xt to which the input weight Wx is applied from the input cell ICt of the t-th stage St, and The state value Ht-1 of the previous stage to which the state weight Wht-1 is applied is received from the hidden cell HCt-1 of the stage, that is, the t-1 th stage St-1, and the input weight Wx is applied by applying the threshold value b. The state value Ht of the current stage is calculated by performing an operation on the input value Xt to which is applied and the state value Ht-1 of the previous stage to which the state weight value Wht-1 is applied. The hidden cell HCt of the current stage may transmit the state value Ht of the current stage to the next stage St+1 by applying the state weight Wht. Also, the hidden cell HCt of the current stage may transmit the state value Ht of the current stage to the output cell OCt of the current stage by applying the output weight value Wyt. Then, the output cell OCt of the current stage calculates the output value Yt of the current stage by performing an operation on the state value Ht of the current stage to which the output weight value Wyt is applied.

In the state transfer network (STN), an operation means an operation to which an activation function is applied. The activation function may be exemplified by Sigmoid, Hyperbolic tangent (tanh), Exponential Linear Unit (ELU), Rectified Linear Unit (ReLU), Leakly ReLU, Maxout, Minout, Softmax, and the like.

14 , the prediction network (FCN) sequentially includes an input layer (forecast input layer: FIL), a convolution layer (forecast convolution layer: FCL), a pooling layer (forecast pooling layer: FPL), and a fully connected layer (forecast fully). -connected layer: FFL) and output layer (forecast output layer: FOL). Here, each of the convolution layer FCL, the pooling layer FPL, and the fully connected layer FFL may be two or more. The convolution layer FCL and the pooling layer FPL include at least one feature map (FM). The feature map FM is derived as a result of receiving a value applied with a weight and a threshold to the operation result of the previous layer, and performing an operation on the input value. These weights are applied through a filter or kernel W, which is a weight matrix of a predetermined size. In the embodiment of the present invention, the first filter W1 is used for the convolution operation of the convolutional layer FCL, and the second filter W2 is used for the pooling operation of the pooling layer FPL.

When the state input vector SIV including the predicted diary vector and the state feature vector is input to the input layer FIL, the convolution layer FCL performs a first filter on the state input vector SIV of the input layer FIL. At least one first feature map FM1 is derived by performing a convolution operation using (W1) and an operation by an activation function. Next, the pooling layer FPL performs a pooling or sub-sampling operation using the second filter W2 on at least one first feature map FM1 of the convolution layer FCL to obtain at least one A second feature map FM2 is derived.

The final connection layer FFL includes a plurality of operation nodes D1 to Dn. The plurality of operation nodes D1 to Dn of the final connection layer FFL calculates a plurality of operation values through an operation using an activation function with respect to at least one second feature map FM2 of the pooling layer FPL.

The output layer FOL includes three output nodes O1, O2, and O3. Each of the plurality of operation nodes D1 to Dn of the fully connected layer FFL is connected to the output nodes O1, O2, and O3 of the output layer OL through a channel (indicated by a dotted line) having a weight (W). . In other words, the plurality of operation node values d1, d2, d3, ..., dn of the plurality of operation nodes D1 to Dn are weighted and input to the output nodes O1, O2, O3, respectively. Accordingly, the output nodes O1, O2, and O3 of the output layer FOL are calculated by the activation function for a plurality of operation node values d1, d2, d3, ..., dn to which weights are applied. In response to the state input vector input as (FIL), the power generation amount of the photovoltaic generator 200 of the corresponding microgrid (MG) after a preset time, the storage amount of the energy storage device 300, and the demand amount of the local load 400 are predicted. to calculate the output value. Each of the three output nodes O1, O2, O3 of the output layer FOL is the amount of power generated by the photovoltaic generator 200 of the corresponding microgrid MG after a preset time, the amount of storage of the energy storage device 300 and the local load ( 400) to meet the demand. That is, the output value of the first output node O1 is the predicted amount of power generation of the photovoltaic generator 200, the output value of the second output node O2 is the predicted storage amount of the energy storage device 300, and the third output node The output value of (O3) is the predicted demand amount of the local load 400 .

For example, each of the output nodes O1, O2, O3 has a weight w=[w1, w2, ... , wn] is applied, and the output value is calculated by taking the activation function F on the result. Activation functions used in the above-described convolutional layer (CL), final connection layer (FL) and output layer (OL) are Sigmoid, Hyperbolic tangent (tanh), Exponential Linear Unit (ELU), and ReLU. (Rectified Linear Unit), Leakly ReLU, Maxout, Minout, Softmax, etc. may be exemplified. Any one of these activation functions may be selected and applied to the convolutional layer CL, the final connection layer FL, and the output layer OL.

Again, referring to FIG. 7 , the management unit 135 determines the amount of power generated by the photovoltaic generator 200 of the corresponding microgrid MG predicted by the prediction unit 133 through the prediction model FCM, and the energy storage device 300 . According to the amount of storage and the amount of demand of the local load 400, the excess or deficiency of the amount of power after a preset time of each of the plurality of microgrids is derived. Accordingly, the management unit 135 supplies the excess amount of electric power to the corresponding energy storage device 300 according to the derived excess or deficiency of the amount of electric power, and transmits a control message for controlling the supply of the insufficient amount of electric energy.

Next, a method for training a verification model (VM) according to an embodiment of the present invention will be described. 15 is a flowchart illustrating a method of training a verification model (VM) according to an embodiment of the present invention. 16 is an example of a vector space for explaining a method of training a verification model (VM) according to an embodiment of the present invention.

15 , the learning unit 134 connects learning data, for example, first to fifth sensor data vectors SV1, SV2, SV3, SV4, and SV5, in step S110 to a learning input vector (IV=[SV1) , SV2, SV3, SV4, SV5]). The input vector for learning includes a plurality of sensor devices 10a, 10b, 10c, 10d, 10e in which all of the sensor data underlying the first to fifth sensor data vectors SV1, SV2, SV3, SV4, SV5 are the same. Whether or not is sensor data transmitted within the same report frame is known data.

When such a learning input vector IV is prepared, the learning unit 134 performs learning using a loss function as shown in Equation 1 below.

는 하이퍼파라미터를 나타낸다. In Equation 1, Eintergrity represents a loss function for determining integrity. oi is an output value of the output layer VOL of the verification model VM, and zi is a classification label corresponding to the output value of the output layer VOL of the verification model VM. eij is the operation node value of the fully connected layer (VFL) of the verification model (VM), and hij is the hidden label corresponding to the operation value of the fully connected layer (VFL) of the verification model (VM). i is an index corresponding to the output nodes O1 and O2 of the output layer (1, 2), and j is an index corresponding to the operation nodes E1 to En of the fully connected layer (VFL) (1 to n). and

represents the hyperparameter.

는 0으로 설정한다. 이에 따라, 분류 손실을 최적화하는 학습에서 수학식 1의 손실함수는 다음의 수학식 2와 같다. The learning unit 134 performs settings for optimizing classification loss in step S120 . At this time, the learning unit 134 sets a hyperparameter and a classification label. In this step S320, the learning unit 134 is a hyperparameter

is set to 0. Accordingly, in the learning to optimize the classification loss, the loss function of Equation 1 is as shown in Equation 2 below.

Also, in step S120 , the learning unit 134 sets a classification label. When all of the sensor data is sensor data transmitted by a plurality of sensor devices 10a, 10b, 10c, 10d, and 10e in the same microgrid within the same report frame, the first output node O1 succeeds and the second A classification label [1, 0] is assigned to the output node O2 failure. On the other hand, when all of the sensor data is not sensor data transmitted by the plurality of sensor devices 10a, 10b, 10c, 10d, and 10e in the same microgrid within the same report frame, the first output node O1 succeeds and a classification label [0, 1] corresponding to the failure of the second output node O2.

When the hyperparameter and the classification label are set as described above, the learning unit 134 performs learning to optimize the classification loss in step S130 . At this time, the learning unit 134 inputs the training input vector to the verification model VM. Then, the verification model (VM) calculates an output value through a plurality of operations in which a plurality of inter-layer weights are applied to the input vector for learning. Then, the learning unit 134 uses the loss function as in Equation 2 in which the hyperparameter is set to 0, the classification loss of the loss function of Equation 2 indicating the difference between the previously set classification label and the output value of the output layer VOL is the minimum Classification loss optimization is performed by modifying the parameters of the verification model (VM) to become

This classification loss optimization is iteratively performed using a plurality of different input vectors for learning, and such repetition may be performed until a desired accuracy is reached through an evaluation index.

When the learning according to the classification loss optimization as described above is completed, the learning unit 134 selects a representative vector from among the hidden vectors for each classification target class of the verification model VM in step S140 . The method of selecting a representative vector is as follows.

The hidden vector, which is the operation node value of the fully connected layer (VFL) of the verification model VM, is derived by re-inputting the training input vector used above (S130) into the verification model VM on which the classification loss optimization is made. That is, the hidden vector is E[e1, e2, e3, ... , en]. The learning unit 134 embeds a plurality of hidden vectors in a predetermined vector space. Fig. 16A shows an example in which a plurality of hidden vectors are embedded in a predetermined vector space. Basically, some degree of classification is made for each class of a plurality of hidden vectors by a classification line (DLINE) in a vector space by classification loss optimization. Here, the class to be classified includes class 1 (CLASS1) in which the integrity verification result is success and class 2 (CLASS2) in which the integrity verification result is failure.

That is, in FIG. 13A , class 1 (CLASS1) indicates that the integrity verification result is success, and class 2 (CLASS2) indicates that the integrity verification result is failure. The learning unit 134 may select a representative vector for each class from among a plurality of hidden vectors embedded in this vector space. As the representative vector, as shown in (A) of 13, it is preferable to select an intermediate value among the hidden vectors in the same class.

는 0.5로 설정한다. 이에 따라, 분류 손실을 최적화하는 학습에서 수학식 1의 손실함수는 다음의 수학식 3과 같다. Next, the learning unit 134 performs settings for optimizing the complex loss in step S150 . At this time, the learning unit 134 sets a hyperparameter, a classification label, and a hidden label. In this step S150, the learning unit 134 is a hyperparameter

is set to 0.5. Accordingly, in the learning to optimize the classification loss, the loss function of Equation 1 is the following Equation 3.

Also, in step S350 , the learning unit 134 sets a classification label and a hidden label. The classification label is set as the first output when all sensor data is sensor data transmitted within the same report frame by a plurality of sensor devices 10a, 10b, 10c, 10d, 10e in the same microgrid, as in step S120. A classification label [1, 0] is given to the node O1 success and failure of the second output node O2. On the other hand, when all of the sensor data is not sensor data transmitted by the plurality of sensor devices 10a, 10b, 10c, 10d, and 10e in the same microgrid within the same report frame, the first output node O1 succeeds and a classification label [0, 1] corresponding to the failure of the second output node O2. The setting of the hidden label uses the representative vector selected previously (S140). Accordingly, a representative vector for each class is set as a hidden label.

When the hyperparameter, classification label, and hidden label are set as described above, when the parameter is set as described above, the learning unit 134 performs learning to optimize the complex loss in step S160. At this time, the learning unit 134 inputs the training input vector to the verification model VM.

Then, the verification model (VM) calculates a hidden vector and an output value through a plurality of operations in which a plurality of inter-layer weights are applied to an input vector for learning. Then, the learning unit 134 performs complex loss optimization by modifying the parameters of the regular model so that the complex loss of the loss function of Equation 3 is minimized by using the loss function as in Equation 3 in which the hyperparameter is set to 0.5. . The complex loss includes a classification loss indicating a difference between a classification label and an output value of an output layer, and a hiding loss indicating a difference between a hidden vector and a representative vector. The complex loss optimization is repeatedly performed using a plurality of different ultrasound images for training, and such repetition may be performed until a desired accuracy is reached through an evaluation index.

When the learning by the complex loss is completed, a plurality of hidden vectors in the vector space may be classified as shown in FIG. 13B. That is, the plurality of hidden vectors are moved toward the representative vector in the vector space. Accordingly, it can be seen that the classification for each class becomes clearer. That is, the classification performance of the verification model (VM) may be improved by learning by complex loss.

Next, a method for learning a predictive model (FCM) according to an embodiment of the present invention will be described. 17 is a flowchart illustrating a method for learning a predictive model (FCM) according to an embodiment of the present invention.

Referring to FIG. 17 , the learning unit 134 prepares a plurality of state vectors and prediction diary vectors arranged in chronological order in step S210 as learning data. Here, the learning data includes the amount of power generated by the photovoltaic generator 200, the amount of storage of the energy storage device 300, and the local load ( 400) is known data. A plurality of state vectors for training are arranged in chronological order, and the number is equal to the length of the stage of the state transmission network (STN). For example, in the case of a state transmission network (STN) as shown in FIG. 12, four state vectors arranged in chronological order are provided for learning. Each of the prepared state vectors includes a power generation vector, a power storage vector, a power consumption vector, a time vector, and a diary vector generated from sensor data and diary information. In addition, the predicted diary vector prepared for learning is generated from the diary after a preset time from the time indicated by the time vector of the first state vector among the plurality of state vectors. Here, in the case of the predicted diary vector prepared for learning, it should be noted that the diary of a time past the learning time is used as it is. That is, if the learning time is T, the time of the prediction diary vector is T-t1, and the time of the first state vector is T-t1-t2 (T, t1, and t2 are all positive numbers).

Next, the learning unit 134 sets a prediction label for the training data in step S220. The set prediction label is the amount of power generated by the photovoltaic generator 200, the amount of storage of the energy storage device 300, and the local load after a preset time from the time indicated by the time vector of the first state vector among the plurality of state vectors of the learning data. (400) is the quantity demanded.

Next, the learning unit 134 inputs the training data, that is, a plurality of state vectors and a prediction diary vector arranged in chronological order, to the prediction model FCM in step S230 . Then, the predictive model (FCM) through a plurality of operations in which the unlearned weights of a plurality of layers of the predictive model (FCM) are applied to the plurality of state vectors and the predictive diary vector arranged in the time order input in step S240. An output value representing the predicted value of the amount of power generation of the photovoltaic generator 200 , the storage amount of the energy storage device 300 , and the demand amount of the local load 400 is calculated. More specifically, the state transmission network (STN) of the predictive model (FCM) is a plurality of layers (RIL, RHL) of a plurality of stages (eg, S1 to S4) for a plurality of state vectors arranged in time order among the training data. , ROL) calculates a state feature vector through a plurality of operations to which unlearned weights are applied, and the predictive network (FCN) of the predictive model (FCM) calculates the state feature vector and the prediction diary for learning by the state transfer network (STN). A state input vector including a vector is received, and an output value is calculated through a plurality of operations in which unlearned weights of a plurality of layers (FIL, FCL, FPL, FFL, FOL) are applied to the input state input vector. Here, the output value is the amount of power generated by the photovoltaic generator 200, the amount of storage of the energy storage device 300, and the local load ( 400) is the predicted value of the quantity demanded.

Next, the learning unit 134 modifies the weight of the prediction model (FCM) so that the prediction loss, which is the difference between the previously set prediction label and the output value of the output layer, is minimized by using the loss function as in Equation 4 in step S250. It performs prediction loss optimization.

In Equation 4, Eforecast represents a loss function for demand forecasting. R is an output value of the output layer FOL, and L is a prediction label corresponding to an output value of the output layer FOL. i is an index corresponding to an output node of the output layer FOL. The first term of the loss function eforecast is the L1-norm loss, and the second term indicates the Structural Similarity Index (SSIM). Since the input and output of the predictive model (FCM) have temporal order and two-dimensional position information, the distortion of order and position can be corrected through SSIM.

That is, the learning unit 134 performs prediction loss optimization by correcting the weight of the prediction model FCM so that the prediction loss of the loss function of Equation 4 indicating the difference between the previously set prediction label and the output value of the output layer is minimized.

The prediction loss optimization of steps S210 to S250 described above is repeatedly performed using a plurality of different state vectors and prediction diary vectors for learning, and this iteration can be made until the desired accuracy is reached through the evaluation index. have.

After the learning is completed as described above, a method of authenticating sensor data for photovoltaic power received through LoRa communication according to an embodiment of the present invention will be described. 18 is a flowchart illustrating a method of managing sensor data for photovoltaic power received through LoRa communication according to an embodiment of the present invention.

Referring to FIG. 18 , the registration unit 131 of the energy management device 100 repeats a report frame (P1, P1, P2, ..., Pn) is assigned. In this state, the registration unit 131 registers from any one sensor device 10 belonging to any one of a plurality of microgrids (MG: MG1, MG2, ..., MGn) through the communication unit 110 in step S320. You can receive a registration request message requesting Then, as shown in FIG. 8 , the registration unit 131 reports a report allocated to the corresponding microgrid MG among the plurality of microgrids MG (MG1, MG2, ..., MGn) for the sensor device 10 . Any one sub report frame (one of S1, S2, S3, S4, S5) is allocated within the frame (eg, P3). Then, the registration unit 131 may transmit a registration response message to the corresponding sensor device 10 together with an indicator indicating the sub report frame allocated through the communication unit 110 .

After the above-described registration is completed, the sensor device 10 generates sensor data by measuring power, current, and temperature for the photovoltaic generator 200 , the energy storage device 300 , and the local load 400 in step S350 . and transmit sensor data through LoRa communication.

Then, when the verification unit 132 receives the report message including the sensor data from the sensor device 10 through the communication unit 110 in step S360, first authenticates the integrity of the sensor data in step S370. That is, the verification unit 132 receives the report message and the report frame P and the sub report frame S, and the report frame P and the sensor device assigned to the microgrid MG to which the sensor device 10 belongs. It is checked whether the sub report frame (S) allocated to (10) is the same. According to this confirmation, the report frame (P) and the sub report frame (S) that received the report message, the report frame (P) and the sensor device (10) allocated to the microgrid (MG) to which the sensor device 10 belongs If the assigned sub report frames (S) are the same, the primary authentication is completed.

If the primary authentication is successful, the verification unit 132 performs secondary authentication in step S380. In the second authentication, the verification unit 132 authenticates the integrity of the received sensor data by comparing the sensor data of the sensor device 10 with the sensor data of other sensor devices received in the same report frame. At this time, the verification unit 132 verifies the integrity of the sensor data through the verification model (VM). The step S380 will be described in more detail as follows.

First, the verification unit 132 receives a predetermined value from each of the sensor data of the verification target sensor device 10 and the sensor data of the other sensor device 10 received in the same report frame as the sensor data of the verification target sensor device 10 . A plurality of sensor data vectors are generated by detecting features. Here, the sensor data vector is a frame vector indicating a report frame allocated to the microgrid to which the sensor apparatus 10 belongs, a subframe vector indicating a sub report frame allocated to the sensor apparatus 10, power, It includes a time vector indicating a time at which current and temperature are measured, and a measurement vector indicating measured values obtained by measuring power, current, and temperature by the sensor device 10 . As an example, referring to FIGS. 2 and 8 , the third microgrid MG3 includes the first to fifth sensor devices 10a, 10b, 10c, 10d, and 10e, and the verification target is the fourth sensor device ( 10d) is assumed. Also, it is assumed that the third report frame P3 is allocated to the third microgrid MG3. In addition, it is assumed that the fourth sensor device 10d to be verified has allocated the third sub report frame S3. The verification unit 132 includes the first sensor data of the fourth sensor device 10d, which is the verification target, and the third report frame P3, which is the same report frame as the sensor data of the fourth sensor device 10d, which is the verification target. A sensor data vector is generated by detecting a predetermined characteristic from each of the sensor data of the second, third, and fifth sensor devices 10a, 10b, 10c, and 10e. As a result, the verification unit 132 receives the sensor data vector from each of the sensor data of the first to fifth sensor devices 10a, 10b, 10c, 10d, and 10e received through the third report frame P3, that is, the first to the fifth sensor data vectors SV1, SV2, SV3, SV4, and SV5 are derived. Then, the input vector IV is generated by connecting the first to fifth sensor data vectors.

The verification unit 132 inputs an input vector (eg, IV=[SV1, SV2, SV3, SV4, SV5]) to the verification model VM. Then, the verification model VM calculates an output value through a plurality of operations to which the weights learned among the plurality of layers (VIL, VCL, VPL, VFL, VOL) are applied with respect to the input vector. For example, if the respective output values of the first and second output nodes O1 and O2 of the verification model (VM) output layer (VOL) are 0.864 and 0.136, respectively, the probability that the sensor data to be verified will succeed in integrity authentication is 86%, It indicates that the probability that the sensor data to be verified will fail integrity verification is 14%. When the verification model VM outputs probabilities (0.086, 0.914), the verification unit 132 may determine that the authentication target sensor data has succeeded in integrity verification according to these probabilities (0.086, 0.914).

If the second authentication is successful, the verification unit 132 stores the corresponding data in step S390 and, if necessary, provides it to the prediction unit 133 . On the other hand, if the secondary authentication fails, the verification unit 132 returns authentication failure.

Meanwhile, according to the present invention, the demand after a preset time can be predicted through the authenticated sensor data. These methods will be described. 19 is a flowchart illustrating a method for predicting a demand for solar power generation using a sensor device performing LoRa communication.

Referring to FIG. 19 , the prediction unit 133 of the energy management server 100 continuously inputs a plurality of sensor data whose integrity is verified by classifying them by a plurality of microgrids (MG) from the verification unit 132 in step S410 . receive In other words, the prediction unit 133 may continuously receive a plurality of sensor data received by the communication unit 110 from the plurality of sensor devices 10 through LoRa communication. The sensor data includes measured values and measurement times of power, current, and temperature for the energy storage device 300 , the photovoltaic generator 200 , and the local load 400 .

In addition, the prediction unit 133 is a plurality of sensor data continuously input from the weather information server 500 through the communication unit 110 in step S420, diary information about the diary of each measurement time and a plurality of sensors continuously input It receives weather prediction information, which is a weather predicted after a preset time from each measurement time of the data.

Then, the prediction unit 133 generates a plurality of state vectors arranged in chronological order from a plurality of sensor data and diary information in step S430, and the time of the first state vector in time among the plurality of generated state vectors A predicted diary vector is generated from weather prediction information that is a predicted diary after a preset time from the time indicated by the vector. Each of the plurality of state vectors includes a power generation vector, a power storage vector, a power consumption vector, a time vector, and a weather vector. Here, the power generation vector represents the measured power, current and temperature of the photovoltaic generator 200 , the power storage vector represents the measured power, current and temperature of the energy storage device 300 , and the power consumption vector represents the local load 400 represents the measured power, current and temperature. The time vector represents a measurement time during which the sensor device 10 measures power, current, and temperature. The weather vector represents the weather observed at the measurement time when the sensor device 10 measures power, current, and temperature. In addition, the predicted diary vector represents the predicted diary after a preset time from the time indicated by the time vector of the first state vector among the generated state vectors.

Next, the prediction unit 133 inputs a plurality of state vectors and prediction diary vectors arranged in chronological order in step S440 to the prediction model (FCM). Then, the predictive model (FCM) has a plurality of subnets (STN, FCN) and a plurality of layers (RIL, RHL) of the predictive model (FCM) for the plurality of state vectors and the predictive diary vector arranged in the time order input in step S450. , ROL and FIL, FCL, FPL, FFL, FOL) through a plurality of operations to which the learned weights are applied, the sun after a preset time from the time indicated by the time vector of the first state vector in time among the plurality of state vectors An output value representing the predicted value of the amount of power generation of the photovoltaic generator 200 , the storage amount of the energy storage device 300 , and the demand amount of the local load 400 is calculated. In more detail with respect to step S450, as follows.

The state transmission network (STN) of the predictive model (FCM) applies the learned weights of a plurality of layers (RIL, RHL, ROL) of a plurality of stages (eg, S1 to S4) for a plurality of state vectors arranged in time order. A state feature vector is calculated through a plurality of operations. The state feature vector represents the correlation characteristics with respect to the amount of power generation of the photovoltaic generator 200 of the corresponding microgrid, the storage amount of the energy storage device 300, and the demand amount 400 of the local load according to the passage of time and the change of weather. Next, the prediction network (FCN) of the prediction model (FCM) receives the state input vector including the state feature vector and the prediction diary vector calculated by the state transmission network (STN), and a plurality of layers ( An output value is calculated through a plurality of operations to which the learned weights (FIL, FCL, FPL, FFL, FOL) are applied. Here, the output value is the amount of power generated by the photovoltaic generator 200, the amount of storage of the energy storage device 300, and the local load 400 after a preset time from the time indicated by the time vector of the first state vector of the plurality of state vectors. ) represents the predicted value of the quantity demanded.

Meanwhile, the present invention may perform demand management based on the predicted value. These methods will be described. 20 is a flowchart illustrating a method for managing demand for solar power generation using a sensor device performing LoRa communication.

Referring to FIG. 20 , when steps S410 to S450 are performed for each of the plurality of microgrids (MG), the amount of power generated by the photovoltaic generator 200 after a preset time of each of the plurality of microgrids (MG), the energy storage device A predicted value of the storage amount of 300 and the demand amount of the local load 400 can be obtained. Then, the management unit 135 determines the amount of power generated by the photovoltaic generator 200 of the plurality of microgrids (MG) predicted by the prediction unit 133 through the prediction model (FCM) in step S510, and the amount of storage of the energy storage device 300 . And, according to the demand amount of the local load 400, the excess or the shortfall of the amount of power after a preset time of each of the plurality of microgrids (MG) is derived. That is, the management unit 135 derives a surplus or a shortage by subtracting the predicted demand from the predicted power generation and storage.

Accordingly, the management unit 135 supplies the excess of the amount of power according to the excess or the shortage of the amount of power derived to the energy storage device 300 of the corresponding microgrid (MG) through the communication unit 110 in step S520, and Transmits a control message to control the supply and demand of the shortage. According to the control message of the management unit 135, the energy storage device 300 of the microgrid (MG), which has a surplus because the predicted power generation and storage amount is larger than the predicted demand amount according to the control message, the predicted power generation amount and the storage amount are smaller than the predicted demand amount. Power is provided to the energy storage device 300 of the microgrid (MG). In particular, according to the control message of the management unit 135, the energy storage device 300 of the microgrid (MG) with the surplus divides the amount of power supplied so that the shortage of the microgrid (MG) with the shortage within the surplus is supplemented in stages. Thus, it is provided to the energy storage device 300 of the microgrid MG at a predetermined cycle from the current time point to the predicted time point.

Meanwhile, the above-described method according to an embodiment of the present invention may be implemented in the form of a program readable by various computer means and recorded in a computer readable recording medium. Here, the recording medium may include a program command, a data file, a data structure, etc. alone or in combination. The program instructions recorded on the recording medium may be specially designed and configured for the present invention, or may be known and available to those skilled in the art of computer software. For example, the recording medium includes magnetic media such as hard disks, floppy disks and magnetic tapes, optical recording media such as CD-ROMs and DVDs, and magneto-optical media such as floppy disks ( magneto-optical media) and hardware devices specially configured to store and execute program instructions such as ROM, RAM, flash memory, and the like. Examples of program instructions may include not only machine language wires such as those generated by a compiler, but also high-level language wires that can be executed by a computer using an interpreter or the like. Such hardware devices may be configured to operate as one or more software modules to perform the operations of the present invention, and vice versa.

Although the present invention has been described above using several preferred embodiments, these examples are illustrative and not restrictive. As such, those of ordinary skill in the art to which the present invention pertains will understand that various changes and modifications can be made in accordance with the doctrine of equivalents without departing from the spirit of the present invention and the scope of rights set forth in the appended claims.

10: instrument
20: power converter
100: energy management device
200: photovoltaic generator
300: energy storage device
400: load
500: weather information server

Claims (10)

Demand for solar power generation based on weather forecast in a microgrid including an energy storage device for storing power produced by a photovoltaic generator, and a local load receiving power from at least one of the photovoltaic generator and the energy storage device A device for prediction, comprising:
Continuously receiving a plurality of sensor data including measurement values and measurement times of power, current and temperature for energy storage devices, photovoltaic generators and local loads through LoRa communication from a plurality of sensor devices,
A plurality of sensor data continuously received from the weather information server to receive diary information for each measurement time and weather forecast information, which is a diary predicted after a preset time from each measurement time of a plurality of sensor data continuously input communication department; and
A power generation vector representing the power generated by the photovoltaic generator, each of which is generated based on the measured value and measurement time, and weather information, a power storage vector representing the power stored by the energy storage device, and the consumption of the local load generating a plurality of state vectors arranged in chronological order while including a power consumption vector indicating the power to
generating a predicted diary vector indicating a predicted diary after a preset time from the time indicated by the time vector of the first state vector of the plurality of state vectors in time;
From the time indicated by the time vector of the first state vector among the plurality of state vectors through a plurality of operations in which the learned weights of the plurality of layers of the predictive model are applied to the plurality of state vectors and the predictive diary vector a prediction unit for calculating predicted values of the amount of power generation of the photovoltaic generator after a set time, the storage amount of the energy storage device, and the demand amount of the local load;
characterized in that it comprises
A device for forecasting demand. According to claim 1,
the device is
When the prediction unit calculates the predicted values of the power generation amount, the storage amount, and the demand amount of each of a plurality of microgrids,
Deriving an extra or insufficient amount of power of each of the plurality of microgrids according to the power generation amount, the storage amount and the demand amount of each of the plurality of microgrids,
a management unit for transmitting a control message for controlling to supply an excess amount of electric power to the energy storage device of a corresponding microgrid through the communication unit or to supply an excess amount of electric power according to the derived excess or shortage of electric energy;
characterized in that it further comprises
A device for forecasting demand. According to claim 1,
The predictive model is
Through a plurality of operations in which the learned weights of a plurality of layers of a plurality of stages are applied to a plurality of state vectors arranged in time order
a state transmission network that calculates a state feature vector representing a correlation characteristic with respect to the amount of power generation of the microgrid photovoltaic generator, the storage amount of the energy storage device, and the demand amount of the local load according to the passage of time and the change of weather; and
A state input vector including the predicted diary vector and the state feature vector calculated by the state transmission network is received as input, and the learned weights of a plurality of layers are applied to the input state input vector through a plurality of operations.
a prediction network for calculating predicted values of the amount of power generation of the photovoltaic generator, the storage amount of the energy storage device, and the demand amount of the local load after a preset time from the time indicated by the time vector of the first state vector of the plurality of state vectors;
characterized by comprising
A device for forecasting demand. According to claim 1,
the device is
Prepared with learning data including a plurality of state vectors and prediction diary vectors arranged in time order,
Set the generation amount of the photovoltaic generator after a preset time from the time indicated by the time vector of the first state vector among the plurality of state vectors of the learning data, the storage amount of the energy storage device, and the demand amount of the local load as a prediction label,
Input the training data to the predictive model,
The state transmission network of the predictive model calculates the state feature vector through a plurality of operations in which unlearned weights of a plurality of layers of a plurality of stages are applied to the plurality of state vectors, and the prediction network of the predictive model uses the state feature vector and A state input vector including the prediction diary vector is received, and a plurality of unlearned weights of a plurality of layers are applied to the input state input vector in a first order of a plurality of operations through a plurality of operations. When calculating the output value that is the predicted value for the amount of power generation of the photovoltaic generator, the amount of storage of the energy storage device, and the amount of demand of the local load after a preset time from the time indicated by the time vector of the state vector,
a learning unit that performs prediction loss optimization by modifying the weight of the prediction model so that the prediction loss of the loss function representing the difference between the prediction label and the output value of the output layer is minimized;
characterized in that it further comprises
A device for forecasting demand. 5. The method of claim 4,
the learning unit
formula

Correct the weight of the prediction model so that the prediction loss, which is the difference between the prediction label and the output value, is minimized using
The forecast is a loss function,
wherein R is the output value,
L is the prediction label,
wherein i is the index of the output node
A device for forecasting demand. Solar power using a weather prediction-based sensor device in a microgrid including an energy storage device for storing power generated by a photovoltaic generator, and a local load receiving power from at least one of the photovoltaic generator and the energy storage device A method for forecasting demand for power generation, the method comprising:
Continuously receiving, by a communication unit, a plurality of sensor data including measurement values and measurement times of power, current and temperature for an energy storage device, a photovoltaic generator, and a local load through LoRa communication from a plurality of sensor devices;
The communication unit is continuously received from the weather information server, weather information on the diary of each measurement time of the plurality of sensor data and the weather prediction information that is the predicted diary after a preset time from the measurement time of each of the plurality of sensor data continuously input receiving;
A power generation vector representing the power generated by the photovoltaic generator generated based on the measured value, the measuring time, and weather information, a power storage vector representing the power stored by the energy storage device, and the local load, respectively, by the prediction unit generating a plurality of state vectors arranged in chronological order while including a power consumption vector indicating the power consumed, a time vector indicating the measurement time, and a diary vector indicating diary information of the measurement time;
generating, by the prediction unit, a predicted diary vector indicating a predicted diary after a preset time from a time indicated by a time vector of a first state vector of a time sequence among the plurality of state vectors;
The time vector of the state vector of the first order in time among the plurality of state vectors through a plurality of operations in which the learned weights of a plurality of layers of a predictive model are applied to the plurality of state vectors and the prediction diary vector by the prediction unit calculating predicted values of the amount of power generation of the photovoltaic generator, the storage amount of the energy storage device, and the demand amount of the local load after a preset time from time;
characterized by comprising
Methods for forecasting demand. 7. The method of claim 6,
After calculating the predicted value,
When the prediction unit calculates the predicted values of the generation amount, the storage amount, and the demand amount of each of the plurality of microgrids, the management unit the amount of power of each of the plurality of microgrids according to the generation amount, the storage amount, and the demand amount of each of the plurality of microgrids deriving an excess or deficiency of ; and
Transmitting, by the management unit, a control message for controlling to supply an excess of the amount of electric power to the energy storage device of the corresponding microgrid through the communication unit or to supply the excess amount of electric power according to the derived excess or deficiency of the amount of electric power;
characterized in that it further comprises
Methods for forecasting demand. 7. The method of claim 6,
The step of calculating the predicted value is
Through a plurality of operations in which the state transmission network of the predictive model applies the learned weights of a plurality of layers of a plurality of stages to a plurality of state vectors arranged in time order.
Calculating a state feature vector representing a correlation characteristic with respect to the amount of power generated by the photovoltaic generator of the microgrid according to the passage of time and the change of weather, the amount of storage of the energy storage device, and the amount of demand of a local load; and
The prediction network of the prediction model receives a state input vector including the prediction diary vector and the state feature vector calculated by the state transmission network, and applies the learned weights of a plurality of layers to the input state input vector. through calculation
Calculating predicted values of the amount of power generation of the photovoltaic generator, the storage amount of the energy storage device, and the demand amount of the local load after a preset time from the time indicated by the time vector of the first state vector of the plurality of state vectors;
characterized in that it comprises
Methods for forecasting demand. 7. The method of claim 6,
Before continuously receiving the plurality of sensor data,
Preparing, by the learning unit, learning data including a plurality of state vectors and prediction diary vectors arranged in chronological order;
The learning unit sets the amount of power generation of the photovoltaic generator, the storage amount of the energy storage device, and the demand amount of the local load after a preset time from the time indicated by the time vector of the first state vector among the plurality of state vectors of the learning data, as a prediction label. step;
inputting the learning data into the predictive model by the learning unit;
calculating, by the state transmission network of the predictive model, a state feature vector through a plurality of operations in which unlearned weights of a plurality of layers of a plurality of stages are applied to a plurality of state vectors;
A prediction network of a predictive model receives a state input vector including the state feature vector and the predictive diary vector, and learns through a plurality of operations in which unlearned weights of a plurality of layers are applied to the input state input vector. Calculating an output value that is a predicted value for the amount of power generation of the photovoltaic generator, the storage amount of the energy storage device, and the demand amount of the local load after a preset time from the time indicated by the time vector of the first state vector among the plurality of state vectors of data;
performing prediction loss optimization in which the learning unit corrects the weight of the prediction model so that the prediction loss of the loss function representing the difference between the prediction label and the output value is minimized;
characterized in that it further comprises
Methods for forecasting demand. 10. The method of claim 9,
The step of performing the prediction loss optimization is
the learning department
formula

Correct the weight of the prediction model so that the prediction loss, which is the difference between the prediction label and the output value, is minimized using
The forecast is a loss function,
wherein R is the output value,
L is the prediction label,
wherein i is the index of the output node
Methods for forecasting demand.

Images