KR20200084454A - A Device and Method for Predictively Operating An ESS Charging Based On Artificial Intelligence - Google Patents

Abstract

According to an embodiment of the present invention, provided is an artificial intelligence-based energy storage system (ESS) control system which can maximize a profit. The artificial intelligence-based ESS control system comprises: a power information supply device providing power usage information; an analysis server generating a power usage predictive model by using the power usage information and transmitting model data learned by using a neural network; and a predictive charging and discharging controller generating power usage predictive data in accordance with the power usage information based on the learned model data, generating an optimal ESS charging and discharging strategy based on the power usage predictive data and controlling an ESS system based on the optimal ESS charging and discharging strategy.

Description

A device and method for predictively operating an ESS charging based on artificial intelligence}

The present invention relates to an artificial intelligence-based predictive ESS charge/discharge operating device, and more specifically, an artificial intelligence-based ESS control system to predict an optimal driving control time and charge/discharge amount to maximize a user's profit. Intelligent based predictive ESS charge and discharge operation device and method.

Recently, the necessity of dissemination of an ESS (Energy Storage System) is increasing for efficient demand management of industrial customers. Accordingly, the government has prepared and provided policy to supply industrial ESS to increase energy efficiency, expand new energy industry, and reduce greenhouse gas emissions.

The purpose of ESS installation for existing industrial customers is to maximize power cost savings through self-load management. To this end, prior studies have been conducted to calculate the appropriate ESS installation capacity and effectively control the ESS in consideration of the customer load and power usage environment. In addition, the industrial ESS can have additional revenue through market participation in addition to its own load management according to the expansion of the new energy business.

ESS of existing industrial customers performs peak reduction and arbitrage trading to reduce power rates. Currently, the electricity rate of industrial customers is calculated as shown in Equation 1 below by introducing the basic rate system into the hourly differential rate plan.

(Equation 1)

Here, C _month is the electricity fee, C _basic is the basic fee (KRW/kW), P _peak is the peak power (kW), C _usage is the usage fee (KRW/kWh), and E _usage is the electricity consumption (kWh). it means.

At this time, peak reduction is a function that prioritizes to lower the P _peak by managing the ESS not to exceed the set peak. The maximum peak of the year can be managed by charging the ESS in FIG. 9(a) and the light load time period when the usage load of the consumer is low and then discharging the ESS in the morning/afternoon peak period. Therefore, it is possible to manage the total peak power through the method of managing the peak power on the same day, thereby reducing the power charge.

On the other hand, arbitrage trading is a function that saves electricity charges by storing electricity in an inexpensive time in an ESS and discharging it in a relatively expensive time. As shown in Fig. 9(b), the ESS is charged in the light load time period and discharged in full between 14 and 17 hours, which is the maximum load time period, so that the market price of the power price can be gained.

In this case, peak reduction and arbitrage trading functions should be operated organically within a certain battery capacity of the ESS. However, the conventional ESS control uses a charging/discharging strategy that simply discharges the time in the case of a maximum load and charges at a light load point by simply specifying a time as shown in FIGS. 9(a) and 9(b).

Therefore, in the related art, there has been a problem in that charging and discharging lasts unreasonably for a long period of time, adversely affecting battery life, or generating a desired output at a desired time to optimize charging and discharging.

The artificial intelligence-based predictive ESS charge/discharge operating device to automatically minimize the loss of users and maximize profits by automatically predicting the optimal operation time and charge/discharge amount of the ESS when controlling power to be solved by the present invention (Predictive ESS Charging Operator, PECO).

Another problem to be solved by the present invention is to provide an artificial intelligence-based predictive ESS charge/discharge operating device that can maximize energy savings by periodically predicting power demand and performing an optimal operation strategy of the ESS.

In order to solve the problems as described above, the artificial intelligence-based ESS control system according to an embodiment of the present invention includes a power information supply device that provides power usage information; An analysis server generating a power usage prediction model using the power usage information, and transmitting the trained model data using a neural network; And generating power usage prediction data according to the power usage information based on the learned model data, generating an optimal ESS charging and discharging strategy based on the power usage prediction data, and predicting to control the ESS system based on the optimal ESS charging and discharging strategy. It may include a red ESS charging and discharging operation device (Predictive ESS Charging Operator, or predictive charge and discharge controller).

Further, the power information supply device may be connected through an IoT network gateway, and the analysis server may be connected through a broadband communication network router, and further include an Ethernet switch for routing power usage information and power usage prediction model data.

In addition, the optimal ESS charge/discharge strategy may include whether the ESS is charged or discharged, a charge/discharge cycle and a charge/discharge power control amount, and an optimum operation time point.

In addition, the predictive charge/discharge controller includes: a receiving unit receiving a power usage prediction model from an analysis server and receiving power usage information from a power information supply device; An operation unit for calculating a decision variable having a minimum objective function based on the power usage prediction data and the power usage information, and then calculating an optimal ESS charge/discharge strategy based on the decision variable; It may include a control unit for controlling the charge and discharge of the ESS based on the optimal ESS charge and discharge strategy.

Further, the decision variable may be a minimum value for which the objective function is calculated according to Equation 1 below.

(Equation 1)

는 실제 사용된 전력사용량 요금을 나타냄)(Where, Λ is the length between the prediction times (Λ=N/24), ω is the daily base rate, u is the maximum power demand (kWh) between the daily prediction times,

Indicates the actual electricity usage fee)

In addition, the objective function may satisfy a condition calculated according to Equation 2 below.

(Equation 2)

는 계약용량(kW),

는 당월 이전까지 기본요금적용전력량(kWh),

는 예측 시간 사이의 최대충방전 전력량(kWh),

는 일간 예측 시간 사이의 최대전력수요량(kWh)을 나타냄)(here,

Is the contracted capacity (kW),

Is the amount of electricity applied in the basic fee (kWh)

Is the maximum charge/discharge power (kWh) between the forecast times,

Indicates the maximum power demand (kWh) between daily forecast times)

In addition, the objective function may satisfy a condition calculated according to Equation 3 below.

(Equation 3)

는 예측 시간 i의 예측 또는 실측 전력사용량,

는 예측 시간 i의 방전량(kWh),

는 예측 시간 i의 충전량(kWh),

는 일간 예측 시간 사이의 최대전력수요량(kWh)을 나타냄)(here,

Is the prediction or actual power consumption of the prediction time i,

Is the discharge amount of the predicted time i (kWh),

Is the charging amount of the forecast time i (kWh),

Indicates the maximum power demand (kWh) between daily forecast times)

In addition, the objective function may satisfy a condition calculated according to Equation 4 below.

(Equation 4)

는 방전효율,

는 충전효율을 나타냄)(here,

Is the discharge efficiency,

Indicates charging efficiency)

Also, the objective function may satisfy a condition calculated according to Equation 5 below.

(Equation 5)

는 예측 시간 사이의 최대 충방전 전력량(kWh)을 나타냄)(here,

Denotes the maximum amount of charge/discharge power (kWh) between prediction times)

Specific details of other embodiments are included in the detailed description and drawings.

According to the present invention, by periodically predicting the optimal operation time point of the ESS during power control, the peak can be effectively reduced to generate a user's benefit according to the amount of electricity used.

The present invention can maximize energy saving effect by periodically predicting power demand and performing an optimal operation strategy of ESS.

The effects according to the present invention are not limited by the contents exemplified above, and more various effects are included in the present specification.

1 is a schematic diagram of an artificial intelligence-based ESS predictive charge/discharge control system according to an embodiment of the present invention.
2 is a block diagram of a predictive charge/discharge controller according to an embodiment of the present invention.
3 is a block diagram illustrating a predictive ESS charge and discharge operation method according to an embodiment of the present invention.
4 is a diagram for explaining an LSTM network according to an embodiment of the present invention.
5 is a flowchart illustrating a power detection method of an artificial intelligence-based predictive ESS charge/discharge operating apparatus according to an embodiment of the present invention.
6 is a view showing a peak change amount according to ESS control of a conventional ESS control device.
7 is a view showing a peak change amount according to the ESS control of the predictive ESS charge and discharge operation device according to an embodiment of the present invention.
8 is an enlarged view of region A of FIG. 7 and a graph for explaining a change in charging/discharging strategy according to an embodiment of the present invention.
9 is a view for explaining peak reduction and arbitrage trading according to the prior art.

The following is merely illustrative of the principles of the invention. Therefore, those skilled in the art can implement various principles included in the concept and scope of the invention and implement the principles of the invention, although not explicitly described or illustrated in the specification. In addition, all conditional terms and examples listed herein are intended to be understood in principle only for the purpose of understanding the concept of the invention, and should be understood as not limited to the examples and states specifically listed in this way. do.

In addition, in the following description, ordinal expressions such as first and second are intended to describe objects that are equivalent to each other and are independent of each other, and in the order of main/sub or master/slave. It should be understood as meaningless.

The above-described objects, features, and advantages will become more apparent through the following detailed description in connection with the accompanying drawings, and accordingly, those skilled in the art to which the invention pertains can easily implement the technical spirit of the invention. .

Each of the features of the various embodiments of the present invention may be partially or entirely combined or combined with each other, and technically various interlocking and driving may be possible as those skilled in the art can fully understand, and each of the embodiments may be implemented independently of each other. It can also be implemented together in an associative relationship.

Hereinafter, various embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1 is a block diagram of an artificial intelligence-based ESS control system according to an embodiment of the present invention.

Referring to FIG. 1, the artificial intelligence-based predictive ESS charge/discharge operating system includes a power information supply device 200, an analysis server 300, a predictive charge/discharge controller 100, an Ethernet switch 500, and an ESS 400. It includes.

The predictive ESS charge/discharge operating system can predict power demand using a neural network such as ELMAN, RNN or LSTM. In addition, the predictive ESS charge/discharge operating system can establish an optimal operation strategy for battery charge/discharge and power control of the ESS (400, Energy Storage System) and control power. Preferably, the artificial intelligence-based ESS control system may further include a LoRa gateway 600 and an LTE router 700. For example, the optimal driving strategy may include charging/discharging, charging/discharging cycle, and charging/discharging amount.

The power information supply device 200 is a device that periodically measures the amount of power and collects the measured amount of power to supply it to the predictive charge/discharge controller 100. For example, the power information supply device 200 includes a power demand meter 210 that measures the amount of power and a power demand collector 220 that transmits power demand to the power demand collector 220 based on the measured amount of power. . The power demand meter 210 may require a separate power demand collector 220 in a building unit, as there may be a plurality of customers for each customer who is the main subject of each power rate.

The analysis server 300 is a component that generates a power demand learning and prediction model, and trains a power demand pattern using power demand data collected from the power information supply device 200. Here, the power demand pattern is learned through a neural network, for example.

As a result of learning of the analysis server 300 for the predictive charging and discharging controller, the analysis server 300 transmits the corresponding prediction model data through the wireless network when it is necessary to change the existing power demand prediction model. Here, the case where the change is necessary means that the prediction model needs to be updated by evaluating the error between the existing prediction model and the model to which the new power demand is applied. The predictive model data may be, for example, a coefficient of the hidden layer of the neural network where learning has been completed.

The predictive charge/discharge controller 100 generates predicted power usage data based on the predicted model data received from the analysis server 300, and also generates power demand for the predicted power usage data, the remaining charge of the battery, and charges and discharges of the battery. An optimal operation strategy can be established according to the efficiency or power, voltage, and current of the system, and a PCS (410, Power Conditioning System) can be controlled. In addition, the predictive charge/discharge controller 100 may communicate with the power information supply device 200 through the LoRa network and the analysis server 300 through the LTE network. In addition, the predictive charge/discharge controller 100 may be connected to the ESS 400 including the battery 430, the BMS 420, and the PCS 410. Here, the ESS 400 may include a plurality of batteries 430, the BMS 420 manages each battery 430, and the PCS 410 provides DC power to the battery 430 or AC power or It is a configuration for discharging and charging by converting to DC power, and directly controls the ESS 400.

The predictive charge/discharge controller 100 basically reduces the load on the prediction or the data transmission of the analysis server 300, and is a device for controlling the PCS 410 and the BMS 420 more efficiently locally to be. For example, if there is no predictive charge/discharge controller 100, the analysis server 300 and the PCS 410, the BMS 420, and the power demand meter 210 must be directly connected through the Internet. In this case, the amount of communication between each node is absolutely increased, and the load of the analysis server 300 directly connected to a large number of devices is greatly increased, which greatly increases costs and makes maintenance difficult.

In addition, due to such an increase in load, there is a high possibility that proper battery control cannot be performed in a timely manner.

The predictive ESS charge/discharge operating system according to the present invention includes the predictive charge/discharge controller 100, thereby significantly reducing the load on the analysis server 300 and simultaneously controlling the battery timely. In addition, the present invention was implemented to maximize the accuracy and efficiency of analysis by allowing the big data collected by the predictive charge/discharge controller 100 to be analyzed through the resource-rich analysis server 300.

Hereinafter, the predictive charge/discharge controller 100 will be described in detail with reference to FIG. 2.

As shown in FIG. 2, the predictive charge/discharge controller 100 includes a receiver 110, a calculator 120, and a controller 130.

The receiver 110 periodically receives the remaining charge (SoC) of the battery 430 and the power, voltage, and current of the system from the BMS 420 and the PCS 410. In addition, prediction data is periodically received from the analysis server 300.

The calculating unit 120 predicts power demand based on the prediction data received from the receiving unit 110 to reduce the peak. In addition, the calculation unit establishes a charging/discharging strategy by calculating the minimum value of the objective function using linear programming under uncertainty along with power demand prediction. Detailed contents related to this will be described later.

The control unit 130 is a configuration that controls the receiving unit 110 and the calculating unit 120, the remaining charge (SoC) of the battery 430 received from the BMS 420 and the PCS 410 and the power, voltage, and current of the system It can also control the back, and may serve to temporarily store data when the network connection between the predictive charge/discharge controller 100 and the analysis server 300 is disconnected. In addition, when the network connection is unstable and temporarily unavailable (for example, maintenance, system check, etc.), it may serve to transfer data offline.

The Ethernet switch 500 is a configuration that turns on/off the connection between the power information supply device 200 and the predictive charge/discharge controller 100, and periodically monitors states of power, voltage, and current of the system through the PCS 410. Collect and periodically collect the state of charge (SoC) of the battery 430 through the battery management system (BMS). At this time, the Ethernet switch 500 periodically transmits the power demand, SoC, system status, etc. collected through the LTE router 700 to the analysis server 300 when the LTE connection is established, and predictive charging and discharging when the LTE connection is disconnected If data is temporarily stored in the controller 100 and then reconnected, data can be retransmitted in batches. Although the Ethernet switch 500 is illustrated in this specification as being connected to the analysis server 300 and the predictive charge/discharge controller 100 through the LTE network, it is not limited thereto.

Hereinafter, an ESS control method according to an artificial intelligence-based predictive ESS charge/discharge operating apparatus according to an embodiment of the present invention will be described with reference to FIGS. 3 and 4.

3 is a block diagram showing the entire process of the artificial intelligence-based predictive ESS charge and discharge operation method according to an embodiment of the present invention. 4 is a diagram for explaining an AI-based network according to an embodiment of the present invention.

First, according to an embodiment of the present invention, the analysis server 300 generates predictive model data by machine learning, and the predictive charge/discharge controller loads predictive model data (S10).

More specifically, the analysis server 300 generates predictive model data by learning the amount of power received from the power information supply device 200 using a neural network, and the generated predictive model data is predictive charge/discharge controller 100 ). In this specification, it is based on using an LSTM network as a neural network, but is not limited thereto. For example, it may be an RNN network, an ELMAN network, or the like.

According to FIG. 4, power consumption prediction according to an embodiment of the present invention first inputs information (Hidden state) and data input at the current time to the LSTM network.

More specifically, each cell of the LSTM network has a structure as shown in FIG. 4, and when the forget gate, the input gate, and the output gate are expressed by equations, the equations 1 to 8 are as follows.

(Equation 1)

f _t = σ(W _hf h _{t1 +} W _xf x _t +b _f )

Equation 1 is a formula representing the logic at the oblivion gate, where f _t represents a matrix containing a ratio to remove current time information, h _t1 represents a hidden state matrix of the previous time LSTM cell, and x _t represents the current time Represents the input data matrix, Wh _f represents the weight matrix of the sigmoid layer to be matrix multiplied with h _t1 at the oblivion gate, and W _xf is the weight matrix of the sigmoid layer to be matrix multiplied with x _t at the oblivion gate. And b _f represents the bias matrix of the sigmoid layer at the oblivion gate.

In Equation 1, the oblivion gate is responsible for removing information of information that has passed for a long time. In other words, it is possible to calculate how much information that has been elapsed for a long time to be removed is reflected by reflecting the information currently given through Equation (1). At this time, the output value of the oblivion gate is between 0 and 1, and information indicating a real state can be determined by removing information that has been elapsed for a long period according to the output value.

Subsequently, the following equation 2 is used to determine how long to forget the cell state after a long period of time, by combining the hidden state data up to the previous time and the input data of the current time point.

(Equation 2)

Here, C _t denotes a cell state matrix at the current time point, C _t1 denotes a cell state matrix up to the immediately preceding time, and each information in the cell state is removed by multiplication of elements. .

(Equation 3)

는 현재시점 추가될 정보의 비율을 담은 행렬을 나타내고,

은 직전시간 LSTM 셀의 hidden state 행렬을 나타내고,

는 입력 게이트에서

과 행렬곱셈 될 시그모이드 레이어의 가중치 행렬을 나타내고,

는 입력 게이트에서

와 행렬곱셈될 시그모이드 레이어의 가중치 행렬을 나타내고,

는 입력 게이트에서 시그모이드 레이어의 바이어스 행렬을 나타낸다.Equation 3 is a formula representing logic at an input gate,

Denotes a matrix containing the proportion of information to be added at the present time,

Represents the hidden state matrix of the previous time LSTM cell,

At the input gate

And the weight matrix of the sigmoid layer to be multiplied by,

At the input gate

And a weight matrix of the sigmoid layer to be multiplied by,

Denotes the bias matrix of the sigmoid layer at the input gate.

At this time, a temporary cell state is generated by combining the hidden state and the input data using Equations 4 and 5 below.

(Equation 4)

(Equation 5)

는 현재시점에서의 임시 cell state를 나타내고,

은 직전시간 LSTM 셀의 출력을 나타내고, x_t는 현재시점의 입력 데이터를 나타내고,

는 입력 게이트에서

와 행렬곱셈될 탄젠트하이퍼(tanh) 레이어의 가중치 행렬을 나타내고,

는 입력 게이트에서 x_t와 행렬곱셈될 탄젠트하이퍼 레이어의 가중치 행렬을 나타내고,

는 입력 게이트에서 탄젠트하이퍼 레이어의 바이어스 행렬을 나타낸다.here,

Indicates a temporary cell state at the present time,

Denotes the output of the LSTM cell of the immediately preceding time, x _t denotes the input data at the current time,

At the input gate

And the weight matrix of the tangent hyper (tanh) layer to be matrix multiplied,

Denotes the weight matrix of the tangent hyper layer to be multiplied by x _t at the input gate,

Denotes the bias matrix of the tangent hyper layer at the input gate.

)들과 현재 주어진 데이터(

)를 조합하여 생성할 수 있다.At this time, as shown in Equation 5, information newly added to the existing cell state is information organized from the initial to the immediately preceding point (

) And the currently given data (

).

는 입력 게이트에서 업데이트된 cell state 행렬을 나타내고,

는 현재시점에서의 임시 cell state를 나타내고,

는 정보를 제거할 비율을 담은 행렬을 나타내고, i_t는 현재시점 추가될 정보의 비율을 담은 행렬을 나타낸다.In Equation 5

Denotes the updated cell state matrix at the input gate,

Indicates a temporary cell state at the present time,

Denotes a matrix containing a ratio to remove information, and i _t denotes a matrix containing a ratio of information to be added at the present time.

(Equation 6)

Equation (6) is a formula representing logic at an output gate, and it is possible to generate a ratio of hidden information at the current time by combining the information given up to the previous point and the currently given data. .

At this time, as shown in Equation 7 below, the current state can be generated by combining the cell state organized up to the current time and the ratio of information to be added to the current time.

(Equation 7)

는 현재시점까지의 cell state 행렬을 나타내고,

는 현재시점 추가될 정보의 비율을 담은 행렬을 나타낸다.Here, h _t represents the organized information of the current time point,

Denotes the cell state matrix up to the current point in time,

Denotes a matrix containing the proportion of information to be added at the present time.

As described above, the present invention is a process of initializing the cell state at the end of the period by constructing an artificial neural network as shown in FIG. 4 and maintaining and managing the cell state for a long period of sequential data in order to learn the pattern of the sequential data for a long period. By repeating, it is possible to predict the power consumption of the next day.

The hidden state of each unit time is transmitted to the next neural network, and can be used as input information of the corresponding unit time in the delivered neural network. At this time, the deeper the depth of the artificial neural network, the more complicated the pattern can be learned.

Specifically, power usage patterns of each site may be learned by using at least two of four feature data of power usage (Usage), month (Month), day (Day), and time (Hour) as input data. Accordingly, up to four features including power consumption may be generated as output (prediction) unit data of the model.

In addition, the unit time of the input data is preferably 15 minutes, but is not limited thereto, and may be changed to more or less unit time depending on the user's setting. For example, in the case of i-smart, since the power unit is 15 minutes, 96 unit data can be given per day. Accordingly, since power data of one day is used for power bidding the next day, 96 data can be output.

Subsequently, in the case of learning through LSTM, the batch size that bundles the sequential data is determined according to the length of the sequential data, and the batch size may be determined by Equation 8 below.

(Equation 8)

는 순차데이터를 묶어주는 배치크기이며,

는 단위시간의 크기로서 15분 단위로 1식 증가한다. 또한,

은 현재시점 이전까지 입력되는 데이터의 수로서 과거 2일간의 데이터를 통해 바로 다음 시점을 예측할 수 있다. 또한,

는 윈도우의 크기로서, 윈도우 크기만큼의 단위 데이터가 한번의 입력으로 사용될 수 있다.here,

Is the batch size that binds the sequential data,

Is the size of unit time, and is increased by 1 meal in 15 minute increments. Also,

Is the number of data input before the current time point, and the next time point can be predicted directly from the data of the past two days. Also,

Is the size of the window, and the unit data of the window size can be used as one input.

In addition, the data at the time point and the next unit time point can be grouped into input and correct answer pairs for learning. For example, the input data at the time of 00:00 may be used as the correct answer at 00:15. On the other hand, if the unit time is 30 minutes, the input data at 00:00 can be used as the correct answer at 00:30.

In addition, it is possible to periodically collect power consumption sequential data and provide it in a form that the model can learn. That is, only the Unix time and power consumption are provided for the original data, and the model can be trained by changing it to four feature data.

Thereafter, the trained data can be randomly extracted from the power consumption sequential data in a specific time period to obtain a predicted value from the LSTM model, and then the model can be evaluated using MAPE. In some cases, SSE, SDE, RMSE, etc. can also be used. . At this time, models with poor evaluation results are not reflected in the evaluation.

The present invention may predict the power demand (usage) of the next day at a specific time point through the learning method described above, or may further predict power demand of the next day by inputting past data of a specific period. In particular, in order to predict power demand, data as much as the batch size should be used as input.

As illustrated in FIG. 4, performance of a network trained through an LSTM network can be evaluated, for example, mean absolute percentage error (MAPE), sum squared error (SSE), and Standard deviations of error (SDE) can be used.

For example, the average absolute ratio error may be defined by Equation 9 below.

(Equation 9)

는 시간 h에 대한 실제 전기 요금을 의미하고,

는 시간 h에 대한 예측 전기 요금을 의미하고, N은　예측된 시간들의 개수를 의미한다.here,

Is the actual electricity bill for time h,

Is the predicted electricity rate for time h, and N is the number of hours predicted.

In addition, the accuracy of demand forecasting may drop significantly as it becomes farther from the forecasting time. Accordingly, the predictive charge/discharge controller 100 periodically receives the prediction data to increase prediction accuracy and performs demand prediction based on the received prediction data. The predictive charging/discharging controller 100 may periodically make predictions for every period preset by the user among units of seconds (s), minutes (m), and hours (h), for example, the period may be 1 hour. have.

Subsequently, the predictive charge/discharge controller 100 generates predicted power usage data based on the received predictive model data (S20).

The predictive charge/discharge controller 100 may generate predicted power usage data based on the predicted model data received from the analysis server 300. In this case, the received prediction model data may be, for example, coefficients of the LSTM network calculated by the analysis server 300. The predictive charge/discharge controller 100 may implement the same LSTM network as the analysis server 300, and then generate predicted power usage data using the prediction model data downloaded from the analysis server 300.

On the other hand, after the predicted power usage data is generated, the predictive charge/discharge controller 100 may calculate an optimal operating condition based on the predicted power usage data (S30). Here, the optimum operation condition is a condition for moving the load of the maximum load prediction time zone to the light load prediction time zone in order to minimize the power rate charge.

On the other hand, if the period of prediction is too short, the PCS 410 cannot charge and discharge normally, and the battery 430 cannot measure a reliable SoC change in a too short period. In addition, if the period of prediction is too long, prediction accuracy may drop and optimal operation may fail.

Therefore, the predictive charge/discharge controller 100 may use a linear programming method for optimization for efficient operation of the ESS 400. The optimal charging and discharging strategy using a specific linear programming method will be described later.

Finally, the predictive charge/discharge controller 100 controls charge/discharge of the ESS based on the calculated optimal operating condition (S40).

Hereinafter, an optimal operation strategy and power control will be described in detail with reference to FIG. 5.

5 is a flowchart illustrating a power detection method of an artificial intelligence-based predictive ESS charge/discharge operating apparatus according to an embodiment of the present invention. Since the power detection method of the artificial intelligence-based predictive ESS charge/discharge operating apparatus shown in FIG. 5 is the same as the control method of the ESS control apparatus shown in FIG. 3, a duplicate description will be omitted.

First, power demand is predicted from the current time until midnight of the next day (S100). Subsequently, an optimal ESS operation strategy is calculated based on a linear programming method after the current time (S200). Subsequently, the optimal driving command until the next prediction and the optimal driving calculation time is transmitted (S300). Subsequently, a command according to the optimal driving strategy until the next prediction and the optimal driving calculation time is transmitted (S400).

The predictive charge/discharge controller 100 may use an integer linear programming to optimize for efficient operation of the ESS 400. In other words, the optimal charging/discharging strategy using the linear planning method is a method used to adjust the peak according to the prediction cycle. The linear planning method determines when and how much power (kWh) is charged and discharged, and based on this, the optimal operation command You can do

Specifically, the linear programming method finds a determinant with a minimum objective function and updates it in memory.

The objective function (O function) is calculated as shown in Equation 10 below.

(Equation 10)

는 예측 기본 요금을 의미하는 것으로 볼 수 있다. 여기서,

및

도 결정 변수로서

는 예측 시간 i의 계통으로 방전된 전력량을 의미하고,

는 예측 시간 i의 충전에 사용된 전력량을 의미한다.Λ denotes the length between the predicted times (Λ=N/24), ω denotes the unit price per day, u denotes the deciding variable, and the maximum power demand (kWh) between the predicted times of the day. Therefore,

Can be seen as referring to the predicted base rate. here,

And

As a determining variable

Is the amount of power discharged to the system at the predicted time i,

Is the amount of power used to charge the prediction time i.

는 예측 시간 i의 사용량 요금 단가를 의미하며,

는 예측 시간 i의 예측 또는 실측 전력사용량을 의미한다. 이에,

는 실제 사용된 전력사용량 요금을 의미하는 것으로 볼 수 있다.Also,

Means the unit price of the usage fee for the forecast time i,

예측 means the prediction or actual power consumption of the prediction time i. Therefore,

Can be seen as referring to the actual electricity usage fee.

Therefore, the objective function is a variable representing a fee that enters the predicted time, and the objective function is output based on the usage fee, actual power consumption, discharge amount, and charge amount.

Accordingly, in order to minimize the charge, all four of the following conditions must be satisfied while finding the determinant whose objective function is the minimum.

Each of the four conditions is obtained by the following formula.

(Equation 11)

는 계약용량(kW)을 의미하며,

는 당월 이전까지 기본요금적용전력량(kWh)을 의미하며,

는 예측 시간 사이의 최대충방전 전력량(kWh)을 의미하며,

는 일간 예측 시간 사이의 최대전력수요량(kWh)을 의미하는 결정변수이다.here,

Means contract capacity (kW),

Means the amount of electricity applied in basic charge (kWh) until the previous month,

Means the maximum amount of charge/discharge power (kWh) between the forecast times,

Is a determinant variable that means the maximum power demand (kWh) between daily forecast times.

Previously, Equation 11 is a formula for obtaining the first condition for finding a determinant that minimizes the objective function. As shown in Equation 5, peak reduction is not performed if there is no basic rate saving effect.

(Equation 12)

는 예측 시간 i의 예측 또는 실측 전력사용량을 의미하며,

는 예측 시간 i의 방전량(kWh)을 의미하며,

는 예측 시간 i의 충전량(kWh)을 의미하며,

는 일간 예측 시간 사이의 최대전력수요량(kWh)을 의미한다.here,

Is the prediction or actual power consumption of the prediction time i,

Means the discharge amount (kWh) of the prediction time i,

Means the charging amount (kWh) of the predicted time i,

Means the maximum power demand (kWh) between daily forecast times.

(Equation 13)

는 방전효율을 의미하고,

는 충전효율을 의미한다.here,

Means discharge efficiency,

Means charging efficiency.

(Equation 14)

는 예측 시간 사이의 최대 충방전 전력량(kWh)을 의미한다.here,

Means the maximum amount of charge/discharge power (kWh) between the prediction times.

Equation 12 means a condition for peak reduction and prevention of backflow. In other words, it means to prevent the back flow of electric power through over-discharge while performing peak reduction by preventing the pure water from exceeding a specific value. In addition, Equation 7 means a condition for limiting the battery capacity, and Equation 8 means a condition for limiting the capacity of the PCS 410.

Meanwhile, the predictive charge/discharge controller 100 periodically collects the remaining charge (SoC) of the battery through the BMS 420. However, since the information calculated by the BMS 420 may not be accurate in the charging/discharging efficiency of the battery, the predictive charging/discharging controller 100 performs real-time charging residual charge (SoC) in order to normally perform a desired charging/discharging command. While monitoring, the charging/discharging power (kW) of the PCS 410 is continuously adjusted.

In addition, the predictive charge/discharge controller 100 may control the charge/discharge power of the PCS 410 by applying a feedback technique such as a PID (Proportional Integral Derivative) controller until the charge/discharge command is issued.

On the other hand, when the optimal operation (or strategy execution) specified during the charge/discharge command execution is not possible, the operation may be stopped and a new strategy may be established to perform power control. Specifically, if it is determined that it is impossible to perform the optimum operation during the charging/discharging command, it is determined that the next prediction time is set to the current time, and if it is determined that the optimum operation can be performed, it is determined whether an abnormal change in power demand occurs.

Subsequently, the predictive charge/discharge controller 100 continuously monitors the predictive data and, when anomaly occurrence is detected, establishes a new strategy together with a new prediction to perform power control. Here, when measuring the power demand, it is determined that an abnormal change has occurred when the power prediction value does not reach a preset reference range. The detailed operation will be described later with reference to FIG. 8.

Accordingly, when it is determined that an abnormality in power demand has occurred, the next prediction time is set as the current time, and when it is determined that no abnormality in power demand has occurred, the optimal operation is completed. As shown in FIG. 5, the optimal operation is completed. If not, the optimal operation based on the feedback technique may be performed again (S400).

Hereinafter, a comparative example (prior art) and an ESS control device according to an embodiment of the present invention will be described with reference to FIGS. 6 to 8.

6 is a view showing a peak change amount according to ESS control of a general ESS control device. 7 is a view showing a peak change amount according to the ESS control of the artificial intelligence-based predictive ESS charge and discharge operation device according to an embodiment of the present invention. 8 is an enlarged view of area A of FIG. 7 and a graph for explaining a change in charging/discharging strategy according to an embodiment of the present invention.

Referring to FIG. 6, the left bar graph indicates charging and discharging according to unit time, and the right bar graph shows the amount of power over time. Here, the first period (unit time) is indicated as being based on 1 hour, but is not limited thereto and may be in minutes or seconds.

When charging from 1 to 5, and discharging at 10, 11 and 14 to 16, as shown in Figure 6, the amount of power at 10, 11 and 14 to 16 as shown in the right bar graph Notice that it decreases.

That is, the general ESS control device performs a method of continuously charging within a predetermined time range and continuously discharging within a next time range by designating a charging and discharging section based on 24 hours per day as described above.

In addition, since a certain amount of charging and discharging is performed for a certain period of time as shown in FIG. 6, it is easy to adjust the peak to correspond to the occurrence of an event when an event occurs in the amount of power (for example, when there is more or less power than usual). not. In other words, since only a certain amount of output is charged and discharged, there is a problem in that charging and discharging cannot be optimized by generating a desired output at a desired time.

On the contrary, the present invention controls charging and discharging by adjusting/charging/discharging strategy according to evaluation/evaluation every second period (for example, 10 minutes, or 15 minutes) smaller than the first period and controlling the ESS 400 with the modified strategy. , It can be done while flexibly adjusting the peak. In this case, the charging/discharging strategy may include charging or discharging, charging/discharging cycle, and charging/discharging amount.

8(a), the present invention is divided into 10-minute units within one hour, for example, after evaluating the power consumption by the method according to FIG. 4 or other methods at each time point, and changing the charging/discharging strategy based on the evaluation It might be.

For example, as shown in Fig. 8(b), between 14 and 15 o'clock, it is usually controlled by a charging/discharging strategy that discharges at a maximum load section. In this case, when the usage is rapidly reduced due to various reasons such as sudden shutdown of the user or factory shutdown, the efficiency of peak reduction due to ESS discharge may be lowered. Therefore, after modifying the charging/discharging strategy as shown in FIG. 8(c), the ESS 400 can be controlled according to the modified strategy. However, in general, between 14 and 15:00, it is a maximum load section, so charging may not be performed because electricity is expensive.

In addition, similar events can occur at intermediate loads. For example, when the electricity consumption between 11 and 12 o'clock of the consumer suddenly falls due to sudden closure, etc., the efficiency of peak reduction due to ESS discharge may be lowered, and in this case, FIG. 8(c) and The ESS 400 can be controlled according to a similarly modified strategy.

In addition, the present invention does not only charge and discharge as much as a certain amount of output, but has an effect of quickly controlling the amount of power even when an event occurs by performing charging and discharging with each other. For example, in the case of adjusting the discharge amount, the discharge amount may be adjusted by reducing the number of discharges per unit time as shown in part B of FIG. 8(c).

The charging and discharging strategy may be changed to control the PCS 410 and the BMS 420. In FIGS. 8(a) to 8(c), although the second period is evaluated as the evaluation/charge/discharge strategy modification and control in units of 10 minutes, it may be a smaller unit, and the second period may be adjusted to correspond to the optimum change amount. Can be changed.

As described above, the artificial intelligence-based ESS control system according to an embodiment of the present invention can periodically predict the optimal operation time point of the ESS during power control, thereby effectively reducing the peak to generate a user's benefit according to the amount of electricity used.

In addition, the artificial intelligence-based ESS control system according to an embodiment of the present invention can maximize power consumption and increase battery life by periodically predicting power demand and performing an optimal operation strategy of the ESS 400.

The embodiments of the present invention have been described in more detail with reference to the accompanying drawings, but the present invention is not necessarily limited to these embodiments, and may be variously modified without departing from the technical spirit of the present invention. . Therefore, the embodiments disclosed in the present invention are not intended to limit the technical spirit of the present invention, but to explain, and the scope of the technical spirit of the present invention is not limited by these embodiments. Therefore, it should be understood that the above-described embodiments are illustrative in all respects and not restrictive. The scope of protection of the present invention should be interpreted by the claims below, and all technical spirits within the scope equivalent thereto should be interpreted as being included in the scope of the present invention.

100: predictive charge and discharge controller 110: receiver
120: operation unit 130: control unit
200: power information supply device 210: power demand meter
220: power demand collector 300: analysis server
400: ESS 410: PCS
420: BMS 430: battery
500: Ethernet switch 600: LoRa gateway
700: LTE router

Claims (9)

A power information supply device that provides power usage information;
An analysis server generating a power usage prediction model using the power usage information and transmitting the trained model data using a neural network; And
Based on the learned model data, power usage prediction data is generated according to the power usage information, an optimal ESS charge and discharge strategy is generated based on the power usage prediction data, and an ESS system is based on the optimal ESS charge and discharge strategy. Including a predictive charge and discharge controller for controlling,
Artificial intelligence based ESS control system. According to claim 1,
The power information supply device is connected through an IoT network gateway, the analysis server is connected through a broadband communication network router, and further comprising an Ethernet switch for routing the power usage information and the learned model data,
Artificial intelligence based ESS control system. According to claim 1,
The optimal ESS charge/discharge strategy includes whether the ESS is charged or discharged, a charge/discharge cycle and a charge/discharge power control amount, and an optimum operation time point.
Artificial intelligence based ESS control system. According to claim 1,
The predictive charge and discharge controller,
A receiving unit receiving the prediction model data and receiving power usage information from the power information supply device;
A calculating unit calculating a decision variable having a minimum objective function based on the power usage prediction data and the power usage information, and calculating the optimal ESS charging and discharging strategy based on the decision variable; And
Comprising a control unit for controlling the charging and discharging of the ESS based on the optimal ESS charging and discharging strategy,
Artificial intelligence based ESS control system. According to claim 4,
The decision variable is the equation below the objective function 1
(Equation 1)

(Where, Λ is the length between the prediction times (Λ=N/24), ω is the daily base rate, u is the maximum power demand (kWh) between the daily prediction times,

Indicates the actual electricity usage fee)
The minimum value calculated according to,
Artificial intelligence based ESS control system. The method of claim 5,
The objective function is Equation 2 below.
(Equation 2)

(here,

Is the contracted capacity (kW), u bar(

) Is the amount of electricity applied in the basic rate (kWh) until the previous month,

Is the maximum amount of charge/discharge power (kWh) between the forecast times,

Indicates the maximum power demand (kWh) between daily forecast times)
Satisfying the condition calculated according to,
Artificial intelligence based ESS control system. The method of claim 5,
The objective function is Equation 3 below.
(Equation 3)

(here,

Is the predicted or actual power consumption of the predicted time i, d _i is the discharge amount (kWh) of the predicted time i, c _i is the charging amount (kWh) of the predicted time i, and u is the maximum power demand (kWh) between the predicted times per day. Indicated)
Satisfying the condition calculated according to,
Artificial intelligence based ESS control system. The method of claim 5,
The objective function is Equation 4 below.
(Equation 4)

(here,

Is the discharge efficiency,

Indicates charging efficiency)
Satisfying the condition calculated according to,
Artificial intelligence based ESS control system. The method of claim 5,
The objective function is Equation 5 below.
(Equation 5)

(here,

Denotes the maximum amount of charge/discharge power (kWh) between prediction times)
Satisfying the condition calculated according to,
Artificial intelligence based ESS control system.

Images