Charging pile power distribution cabinet load balance control method and system based on LSTM neural network
Through the load prediction and regulation method based on LSTM neural network, the impact of new energy vehicle charging on the load of community distribution cabinets is solved, load balancing is achieved, and the safety and reliability of power supply are improved.
Patent Information
- Application Number
- CN202411152248.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-21
- Publication Date
- 2024-12-03
AI Technical Summary
AC charging of new energy vehicles affects the load of community distribution cabinets, especially during peak electricity consumption, centralized charging leads to an increase in load, which poses safety hazards and affects the stability of power supply.
The load balancing control method of charging pile distribution cabinet based on LSTM neural network is adopted to establish a model by obtaining historical electricity consumption data, predict the load trend at the next moment, and regulate the working conditions of charging piles based on the prediction results to avoid overload.
Accurate prediction and adjustment of the distribution cabinet load is achieved, the safety and reliability of the distribution cabinet is improved, overloading is avoided, and power supply stability is ensured.
Smart Images

Figure CN119070282A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of charging pile power distribution, and specifically to a load balancing control method and system for a charging pile power distribution cabinet based on an LSTM neural network. Background Art
[0002] With the continuous development of new energy technology, new energy vehicles are attracting more and more attention from consumers with their extremely low energy consumption, and the popularity of new energy electric vehicles is rising rapidly. At the same time, the demand for electric vehicle charging piles has also increased significantly. New energy vehicle AC charging piles provide AC power to electric vehicles through conduction. When AC power is input into the charging pile, the control system inside the charging pile will convert it into AC power suitable for the electric vehicle on-board charger to receive, and then output it to the electric vehicle through the charging gun for charging. The on-board charger inside the electric vehicle will convert AC power into DC power, and finally charge the on-board power battery.
[0003] At present, most new energy vehicles use AC charging piles. However, since the total load that the community distribution cabinet can withstand is fixed, the impact of electric vehicle charging on the community power grid, especially the concentrated charging of electric vehicles during peak hours of electricity consumption, will lead to an increase in load, which will affect the power supply stability and load safety of the community, and there are certain safety hazards.
[0004] The information disclosed in this background technology section is only used to deepen the understanding of the background technology of the present disclosure, and should not be regarded as acknowledging or suggesting in any form that the information constitutes the prior art known to those skilled in the art. Summary of the invention
[0005] In view of at least one of the above technical problems, the present disclosure provides a load balancing control method and system for a charging pile distribution cabinet based on an LSTM neural network, aiming to solve the problem that car charging affects other electricity usage scenarios in the community due to the limited total load of the community distribution cabinet.
[0006] According to one aspect of the present disclosure, a method for load balancing control of a charging pile distribution cabinet based on an LSTM neural network is provided, which comprises the following steps: (1) Obtain the historical power consumption data of the distribution cabinet, establish the distribution cabinet load LSTM neural network model, and train the distribution cabinet load LSTM neural network model based on the historical power consumption data of the distribution cabinet; (2) Obtain the total input power information of the distribution cabinet and the working requirements of each charging pile; (3) Based on the trained LSTM neural network model of the distribution cabinet load, the load trend of the distribution cabinet at the next moment is predicted according to the total input power information of the distribution cabinet; (4) According to the load trend of the distribution cabinet and the working requirements of the charging pile, the working state of the charging pile is adjusted accordingly under the rated load of the distribution cabinet.
[0007] In some embodiments of the present disclosure, in step (1), historical power consumption data of the power distribution cabinet is collected regularly within a certain time period.
[0008] In some embodiments of the present disclosure, the distribution cabinet load LSTM neural network model performs single-feature load prediction, and the training data of the single-feature load includes three dimensions: the number of input samples, the time step, and the number of input features.
[0009] In some embodiments of the present disclosure, the distribution cabinet load LSTM neural network model includes three LSTM hidden layers and two fully connected layers.
[0010] According to another aspect of the present disclosure, a charging pile distribution cabinet load balancing control system is provided, which includes an electric energy collector electrically connected to the total input end of the distribution cabinet, a server connected to the electric energy collector in real time communication and used to run the distribution cabinet load LSTM neural network model and issue control instructions according to the distribution cabinet load trend, a charging pile connected to the server in communication and used to respond to the server control instructions, and a client terminal connected to the charging pile in communication.
[0011] In some embodiments of the present disclosure, the power harvester is a three-phase power harvester or a single-phase power harvester.
[0012] In some embodiments of the present disclosure, the server includes a basic data unit, a charging data unit, a collection data unit, and a switch control unit; the basic data unit includes a load setting module, a charging pile parameter setting module, and a load prediction module; the charging data unit includes a charging setting module, a charging power module, and a charging time module; the collection data unit includes a charging status monitoring module, a charging data collection module, and a charging information query module; the switch control unit includes a charging pile switch module and a charging pile emergency stop module.
[0013] One or more technical solutions provided in the embodiments of the present application have at least any of the following technical effects or advantages: After the LSTM neural network model of the distribution cabinet is trained with the historical power load data of the distribution cabinet, it can accurately predict the power load of the next period through the data collected by the power collector as input, so that the power quota allocation of the distribution cabinet can be adjusted in advance according to the model prediction trend to avoid overload problems caused by exceeding the maximum load of the distribution cabinet, which can effectively improve the safety and reliability of power distribution. BRIEF DESCRIPTION OF THE DRAWINGS
[0014] Figure 1This is a schematic diagram of the LSTM network architecture in one embodiment of the present application.
[0015] Figure 2 This is a comparison chart of the LSTM neural network model prediction curve and the actual load curve of the distribution cabinet in one embodiment of the present application.
[0016] Figure 3 This is a principle block diagram of a load balancing control system for a charging pile distribution cabinet in one embodiment of the present application.
[0017] In the above figures, 1 is the LSTM hidden layer and 2 is the fully connected layer. DETAILED DESCRIPTION
[0018] The programs involved or relied upon in the following embodiments are all conventional programs or simple programs in the technical field, and those skilled in the art can make conventional selections or adaptive adjustments according to specific application scenarios.
[0019] In order to better understand the technical solution of the present application, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.
[0020] In order to avoid the impact of AC charging piles on the load of the community distribution cabinet, which leads to conflicts between AC charging piles and the electricity consumption of community residents, resulting in the problem of overload operation of the distribution cabinet, this example discloses a load balancing control method for charging pile distribution cabinets based on LSTM neural network, which adjusts the load of the distribution cabinet in advance by predicting the load trend at the next moment. It specifically includes the following steps: (1) Obtain the historical power consumption data of the distribution cabinet, establish the distribution cabinet load LSTM neural network model, and train the distribution cabinet load LSTM neural network model based on the historical power consumption data of the distribution cabinet.
[0021] Considering that the load that the community distribution cabinet can bear is fixed, after too many AC charging piles are connected to the community distribution cabinet, it will inevitably affect the electricity consumption of residents, elevators, community lighting, etc. Therefore, it is necessary to use the idle capacity of the distribution cabinet for AC charging pile power supply and distribution under the premise of ensuring the necessary electricity consumption of the community. However, since the power consumption of necessary electricity is not constant, it shows different power consumption patterns in different time periods. Therefore, it is necessary to predict the power consumption of necessary electricity so that the working status of the AC charging pile can be adjusted in advance to prevent the distribution cabinet from overloading.
[0022] Specifically, the core of orderly electricity consumption is to link known historical data with factors that affect power load, characterize the mapping relationship through mathematical models, and achieve high-precision power load forecasting. In this embodiment, the load trend of the distribution cabinet is predicted through the LSTM neural network model. Among them, the LSTM neural network is a special recurrent neural network, which includes N LSTM modules. The model data is trained in chronological order, and the module at each moment records the weight information of the previous moment. It includes three input quantities: the input sample at the current moment, the short-term memory information at the previous moment, and the long-term memory at the previous moment; the forgetting and updating of information are controlled by the forget gate, update gate, and output gate inside the LSTM. These gates are controlled by sigmoid The activation function generates a value between 0 and 1 to control the forgetting and updating of training data.
[0023] In this embodiment, the power consumption data of the power distribution cabinet for one year is counted as the model training set, and the LSTM neural network model is trained with the training set to obtain a data model of power consumption for one year. The data model can predict the power consumption at the next moment based on the real-time data, so that through real-time monitoring data and predicted values, measures can be taken before the power distribution cabinet exceeds the load to ensure the load balance of the power distribution cabinet. Specifically, in this example, the power distribution cabinet power consumption data is recorded every 15 minutes with a time interval of 15 minutes. Therefore, the amount of data recorded in one day is 96, and the amount of data recorded in one year is 35040. Therefore, the load data of the power distribution cabinet for one year can be expressed as: ; For X(m, n), m is the data corresponding to the mth day of the year, and n is the data corresponding to the nth moment of the day. Then, the LSTM neural network model of the distribution cabinet load is trained based on the historical power consumption data of the distribution cabinet.
[0024] In addition, in this embodiment, the network architecture of the LSTM neural network model is shown in Figure 1 , which includes three LSTM hidden layers and two fully connected layers. The LSTM hidden layer is used to process and store information in the sequence data, and avoids the problems of gradient disappearance and gradient explosion through the forget gate, input gate, and output gate. Figure 1 In this embodiment, links between different hidden layers are established; thus, the short-term memory information H of the previous hidden state is t-1 It is not only transmitted to the output layer, but also transmitted to the next time step, and together with the current input sample, it constructs short-term memory information. In addition, in this example, the output gate of the LSTM hidden layer is connected to the fully connected layer, which is used to perform weighted summation on the output of the previous layer of the network and increase the expression ability through the activation function. After three LSTM hidden layers, the final X n+96The neurons of the time step output gate are connected to the fully connected layer, and the Y value is output through two layers of fully connected layers, thereby establishing the connection between each input unit and each output unit.
[0025] Since the LSTM hidden layer already has nonlinear characteristics, the introduction of nonlinear activation functions through the fully connected layer can further enhance the nonlinear transformation ability of the model, so that the LSTM neural network model may be better adapted to complex time series data than the traditional model. In addition, the fully connected layer can integrate and extract useful information in the hidden state output by the LSTM hidden layer to form a higher-level feature representation. At the same time, the fully connected layer can also map the features extracted by the LSTM hidden layer to the sample label space, so as to output the probability distribution or specific numerical prediction of each category, thereby improving the accuracy of the model prediction classification; in addition, considering that in this embodiment, the model input data sequence is relatively long, in order to avoid the overfitting problem of the LSTM neural network model due to the long input sequence length, this example introduces regularization technology through the fully connected layer to avoid model overfitting.
[0026] (2) Obtain the power information of the total input end of the distribution cabinet and the working requirements of each charging pile.
[0027] In order to predict the load trend of the power distribution cabinet at the next moment, it is necessary to obtain the load data of the power distribution cabinet at the current moment and the previous moment as a prediction benchmark. In this embodiment, by obtaining 96 data volumes of the power distribution cabinet in one day, the power consumption at the next moment is predicted through the LSTM neural network model. The data is expressed as follows: [[[X 1 ],[X 2 ]…[X 96 ]],[[X 2 ],[X 3 ]…[X 97 ]]…[[X n+1 ],[X n+2 ]…[X n+96 ]]].
[0028] (3) Based on the trained LSTM neural network model of the distribution cabinet load, the load trend of the distribution cabinet at the next moment is predicted according to the total input power information of the distribution cabinet.
[0029] In this embodiment, a single-feature load forecast is performed based on the trained distribution cabinet load LSTM neural network model, that is, future power consumption data is predicted based on historical power consumption data, and the input load data is in the form of a three-dimensional data structure, and the three dimensions include the number of input samples each time batch_size, the time step time_step, and the number of input features input_dimensions; in this embodiment, the data volume of the three dimensions corresponds to (34944, 96, 1), and the distribution cabinet load LSTM neural network model outputs a two-dimensional data structure with a value of (34944, 1). Therefore, the distribution cabinet load LSTM neural network model predicts the load data of the nth period based on the n-1th to n-96th periods.
[0030] Specifically, when implementing the above steps, the historical power data is first stored in a table in .csv format, and then the data is trained, verified and tested in keras. Before data training, the data in the .csv table is sorted by numpy, and then input into keras for data processing, the training model is output, and the data is predicted through the training model.
[0031] See also Figure 2 By comparing the predicted data of the distribution cabinet load LSTM neural network model with the actual distribution cabinet load data, it can be seen that the predicted value of the distribution cabinet load LSTM neural network model is consistent with the load change trend of the actual output of the distribution cabinet, and the predicted load curve basically coincides with the actual load curve. This shows that the distribution cabinet load LSTM neural network model based on the single feature of this method can accurately predict the load power.
[0032] (4) According to the load trend of the distribution cabinet and the working requirements of the charging pile, the working state of the charging pile is adjusted accordingly under the rated load of the distribution cabinet.
[0033] After the load prediction trend of the distribution cabinet is obtained through the LSTM neural network model of the distribution cabinet load, combined with the total capacity of the distribution cabinet and the current power supply requirements of each AC charging pile, such as the number of working AC charging piles and the charging power of each charging pile, the working state of each charging pile is adjusted according to the idle capacity of the distribution cabinet space to achieve orderly charging of the charging piles and avoid the total load exceeding the rated value of the distribution cabinet. In addition, peak shifting and valley filling can be carried out according to the charging demand of the charging piles, thereby reducing the charging cost.
[0034] This example also discloses a charging pile distribution cabinet load balancing control system, based on which the above-mentioned distribution cabinet load balancing control method is implemented. The system includes an electric energy collector, a server, a charging pile and a client terminal.
[0035] The power collector is electrically connected to the main input terminal of the power distribution cabinet, and is used to monitor the total power consumption information of all devices under the power distribution cabinet, including voltage, current, power, and power information. In this embodiment, the power collector is a three-phase power collector, and in some other embodiments, the power collector is a single-phase power collector.
[0036] The server is the control center of the control system. It is connected to the power collector to obtain the power information collected by the power collector, and performs algorithmic processing on the collected data based on the collected power information. According to the LSTM neural network model of the distribution cabinet load, the power load trend of the distribution cabinet is predicted.
[0037] In this embodiment, see Figure 3 The server includes a basic data unit, a charging data unit, a data collection unit, and a switch control unit. Among them, the basic data unit includes a load setting module, a charging pile parameter setting module, and a load prediction module; the load setting module is used to set the maximum load of the distribution cabinet, the charging pile parameter setting includes the setting of the charging pile for three-phase or one-way, the maximum charging power, the number of charging channels, and the load prediction module is used to set the predicted load value predicted according to the historical load. The charging data unit includes a charging setting module, a charging power module, and a charging time module; the charging setting module is used to set and store information such as charging time, charging power, and maximum protection power, the charging power module is used to store the actual power when the car is charging, and the charging time module is used to record the start time of the AC charging pile, so as to optimize the real charging time according to this time; the data collection unit includes a charging status monitoring module, a charging data collection module, and a charging information query module; the charging status monitoring module is used to implement the charging status of the monitoring equipment such as overvoltage, overcurrent, overpower, leakage, etc., the charging data collection module is used to collect the voltage, current, power and electric energy information of the charging pile, and the charging information query module is used to query the real-time charging information, historical reports, etc. of the charging pile. The switch control unit includes a charging pile switch module and a charging pile emergency stop module. When adjusting the load, the charging pile switch module controls the output of the relay electrically connected to the charging pile to reduce the load of the distribution cabinet, and disconnects the power supply of the charging pile when charging is completed. The charging pile emergency stop module is used to disconnect the power supply of the AC charging pile in an emergency.
[0038] The charging pile is connected to the server for transmitting its charging status information to the server and responding to the control instructions of the server. In this embodiment, the charging pile and the server use 4G network wireless communication, thereby arbitrarily grouping and networking the charging pile devices, and counting the power consumption information of all charging pile devices.
[0039] The client terminal mainly configures, controls, views, and analyzes digital information of each charging device through communication with the server, and can send alarm information to the user through pop-up windows or notifications, so that the user can understand the operating status of the control system in real time. In this embodiment, the client terminal is an APP based on a mobile phone terminal. In some other embodiments, the client terminal is managed through a computer, WeChat applet, etc.
[0040] Although some preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present application.
[0041] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the inventive concept. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.
Claims
1. A method for load balancing control of charging pile distribution cabinet based on LSTM neural network, characterized in that: The steps include: (1) Obtain the historical power consumption data of the distribution cabinet, establish the distribution cabinet load LSTM neural network model, and train the distribution cabinet load LSTM neural network model based on the historical power consumption data of the distribution cabinet; (2) Obtain the total input power information of the distribution cabinet and the working requirements of each charging pile; (3) Based on the trained LSTM neural network model of the distribution cabinet load, the load trend of the distribution cabinet at the next moment is predicted according to the total input power information of the distribution cabinet; (4) According to the load trend of the distribution cabinet and the working requirements of the charging pile, the working state of the charging pile is adjusted accordingly under the rated load of the distribution cabinet.
2. The method for controlling load balancing of a charging pile distribution cabinet according to claim 1, characterized in that: In the step (1), the historical power consumption data of the power distribution cabinet is collected regularly within a certain time period.
3. The method for controlling load balancing of a charging pile distribution cabinet according to claim 1, characterized in that: The LSTM neural network model of the distribution cabinet load is used to predict the single-feature load. The training data of the single-feature load includes three dimensions: the number of input samples, the time step, and the number of input features.
4. The method for controlling load balancing of a charging pile distribution cabinet according to claim 1, characterized in that: The distribution cabinet load LSTM neural network model includes three LSTM hidden layers and two fully connected layers.
5. A load balancing control system for a charging pile distribution cabinet, characterized in that: It includes an electric energy collector electrically connected to the total input terminal of the distribution cabinet, a server which is connected to the electric energy collector in real time communication and is used to run the LSTM neural network model of the distribution cabinet load and issue control instructions according to the load trend of the distribution cabinet, a charging pile which is connected to the server in communication and is used to respond to the server control instructions, and a client terminal which is connected to the charging pile in communication.
6. The charging pile distribution cabinet load balancing control system according to claim 6, characterized in that: The electric energy collector is a three-phase electric energy collector or a single-phase electric energy collector.
7. The charging pile distribution cabinet load balancing control system according to claim 6, characterized in that: The server includes a basic data unit, a charging data unit, a collection data unit, and a switch control unit; the basic data unit includes a load setting module, a charging pile parameter setting module, and a load prediction module; the charging data unit includes a charging setting module, a charging power module, and a charging time module; the collection data unit includes a charging status monitoring module, a charging data collection module, and a charging information query module; the switch control unit includes a charging pile switch module and a charging pile emergency stop module.
Citation Information
Cited By
Charging pile load supervision method and system based on AI technology
CN119858477A
Current magnitude intelligent control method and system based on large model
CN121367225A
A Current Quantity Intelligent Control Method and System Based on a Large Model
CN121367225B
