AC charging pile power distribution method and system

Abstract

The invention provides an AC charging pile power distribution method and system in the technical field of charging piles. The method comprises the following steps: S1, creating a load prediction model; s2, acquiring a large amount of historical power consumption data to construct a data set, and training the load prediction model through the data set; s3, acquiring real-time power consumption data, preprocessing the real-time power consumption data, and inputting the preprocessed real-time power consumption data into the deployed load prediction model to obtain a power consumption load prediction report; s4, acquiring historical charging data of user accounts charged by all AC charging piles in the community, extracting charging behaviors of the user accounts based on the historical charging data, and setting charging priorities of the user accounts based on the charging behaviors; and S5, acquiring a charging rule of the electricity price, and dynamically distributing the power of the AC charging pile based on the electrical load prediction report, the charging priority and the charging rule. The method has the advantages that the reasonability of power distribution of the AC charging pile is greatly improved, and then the user experience of charging is greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of charging piles, and in particular to a power distribution method and system for an AC charging pile. Background Art

[0002] A charging station is a device that provides electrical energy to electric vehicles, allowing them to store enough power to operate. Charging stations can be categorized by charging method: AC (slow charging) and DC (fast charging). AC stations transmit alternating current (AC) from the power grid directly to the electric vehicle's onboard charger, which converts the AC power to DC and charges the battery. DC stations, on the other hand, transmit DC power directly to the electric vehicle's battery, enabling faster charging. AC stations are generally used by individual users within residential communities, while DC stations are typically used at charging stations.

[0003] With the rapid development of electric vehicles, the demand for charging stations is also increasing. To meet this demand, many residential communities have built corresponding AC charging stations. However, the current large-scale deployment of AC charging stations in residential areas has exposed significant power system adaptability issues: during the evening peak electricity consumption period (typically 6:00 PM to 10:00 PM), the combined charging load of the residential area and the residential electricity load create a superposition effect, causing the instantaneous power demand of the distribution network to far exceed the rated capacity of the transformer. According to statistics, when AC charging stations are centrally activated in a typical residential area, the peak load can reach 2.3-3.5 times the base load, which can easily trigger the overload protection mechanism and cause regional power outages, seriously affecting the charging user experience.

[0004] Therefore, how to provide an AC charging pile power distribution method and system to improve the rationality of AC charging pile power distribution and thereby improve the charging user experience has become a technical problem that needs to be solved urgently. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide an AC charging pile power distribution method and system to improve the rationality of AC charging pile power distribution and thereby enhance the user experience of charging.

[0006] In a first aspect, the present invention provides an AC charging pile power distribution method, comprising the following steps:

[0007] Step S1: creating a load forecasting model based on a data preprocessing module, a feature extraction module, a sequence learning module, and an output module, setting the loss function of the load forecasting model to a mean square error function; the data preprocessing module, the feature extraction module, the sequence learning module, and the output module are connected in sequence;

[0008] Step S2: obtaining a large amount of historical electricity usage data, pre-processing each of the historical electricity usage data to construct a data set, training a load forecasting model using the data set, and deploying the trained load forecasting model;

[0009] Step S3: Acquire real-time electricity consumption data, pre-process the real-time electricity consumption data, and input the data into the deployed load forecasting model to obtain an electricity load forecast report;

[0010] Step S4: Obtain historical charging data of user accounts that have charged at all AC charging piles in the community, extract charging behavior of each user account based on the historical charging data, and set a charging priority for each user account based on the charging behavior;

[0011] Step S5: Obtain the electricity price billing rules, and dynamically allocate the power of the AC charging pile based on the electricity load forecast report, charging priority, and billing rules.

[0012] Furthermore, in step S1, the data preprocessing module is used to clean and normalize the electricity consumption data, and input the normalized electricity consumption data into the feature extraction module;

[0013] The feature extraction module is constructed based on a convolutional neural network and is used to extract electricity usage features from electricity usage data and input the electricity usage features into the sequence learning module;

[0014] The sequence learning module is constructed based on several LSTM layers and several Dropout layers, and each of the LSTM layers and Dropout layers is alternately connected; the sequence learning module is used to learn long-term dependencies from each power consumption feature, obtain global features, and input the global features into the output module;

[0015] The output module is used to output a power load forecast report based on global characteristics; the power load forecast report includes an AC charging pile power load curve and a residential power load curve for the next 6 hours.

[0016] Furthermore, the step S2 is specifically as follows:

[0017] Obtain a large amount of historical electricity usage data including electricity usage time, electricity usage date, date type, AC charging station electricity usage data, residential electricity usage data, and environmental data; the date type is weekday, weekend, or holiday; the environmental data includes at least temperature, humidity, wind speed, and weather conditions;

[0018] Performing preprocessing on each of the historical electricity consumption data, including at least outlier removal, missing value filling, duplicate data removal, format unification, and data standardization, and constructing a data set based on each of the preprocessed historical electricity consumption data;

[0019] The data set is divided into a training set, a validation set, and a test set by a ten-fold cross-validation method. The load forecasting model is trained by the training set until the loss value of the loss function is less than a preset loss threshold, and the hyperparameters of the load forecasting model, including at least a learning rate, a random dropout rate, a batch size, and a number of iterations, are continuously optimized during the training process. The trained load forecasting model is then validated by the validation set to determine whether the prediction accuracy is greater than a preset accuracy threshold. If not, the validation fails, and the data set is expanded to continue training. If so, the validation succeeds, and:

[0020] The successfully verified load forecasting model is tested using the test set to determine whether the prediction confidence is greater than a preset confidence threshold. If not, the test fails and the data set is expanded to continue training; if so, the test succeeds, the training ends, and the load forecasting model is deployed.

[0021] Furthermore, the step S3 is specifically as follows:

[0022] Acquire real-time electricity usage data including electricity usage time, electricity usage date, date type, AC charging pile electricity usage data, residential electricity usage data, and environmental data; the AC charging pile electricity usage data and residential electricity usage data are automatically acquired based on the smart meter;

[0023] The real-time electricity consumption data is preprocessed, including at least outlier removal, missing value filling, duplicate data removal, format unification, and data standardization, and then input into the deployed load forecasting model to obtain an electricity load forecast report including the AC charging pile electricity load curve and the residential electricity load curve for the next 6 hours.

[0024] Furthermore, the step S4 is specifically as follows:

[0025] Obtain historical charging data for all user accounts that have charged at AC charging stations in the community. Based on the historical charging data, extract the charging behavior of each user account, including at least charging frequency, average charging power, and charging duration. Set the charging priority of each user account based on the charging frequency, average charging power, and charging duration:

[0026] Charging priority = a / charging frequency + b*average charging power + c / charging time;

[0027] Among them, a, b, and c all represent weight coefficients;

[0028] The step S5 is specifically as follows:

[0029] Obtain the billing rules for electricity prices, set a load threshold, verify the electricity load forecast report based on the load threshold, select a time period in the next 6 hours when the electricity load exceeds the load threshold, and dynamically allocate the power of the AC charging pile within the time period based on the charging priority and billing rules. If the electricity load cannot be met, the photovoltaic energy storage system is used for power supply. If the photovoltaic energy storage system also cannot meet the electricity load, the power output of the AC charging pile corresponding to the user account with the lowest charging priority is shut down in turn until the electricity load is less than the load threshold.

[0030] In a second aspect, the present invention provides an AC charging pile power distribution system, comprising the following modules:

[0031] A load forecasting model creation module, configured to create a load forecasting model based on a data preprocessing module, a feature extraction module, a sequence learning module, and an output module, wherein the loss function of the load forecasting model is set to a mean square error function; the data preprocessing module, the feature extraction module, the sequence learning module, and the output module are connected in sequence;

[0032] A load forecasting model deployment module is used to obtain a large amount of historical electricity usage data, pre-process each of the historical electricity usage data to construct a data set, train the load forecasting model using the data set, and deploy the trained load forecasting model;

[0033] A load forecasting module is used to obtain real-time power consumption data, pre-process the real-time power consumption data, and input the data into the deployed load forecasting model to obtain a power load forecast report;

[0034] A charging priority setting module is used to obtain historical charging data of user accounts that have charged at all AC charging piles in the community, extract the charging behavior of each user account based on the historical charging data, and set the charging priority of each user account based on the charging behavior;

[0035] The power allocation module is used to obtain the billing rules of the electricity price and dynamically allocate the power of the AC charging pile based on the power load forecast report, charging priority and billing rules.

[0036] Furthermore, in the load forecasting model creation module, the data preprocessing module is used to clean and normalize the electricity consumption data, and input the normalized electricity consumption data into the feature extraction module;

[0037] The feature extraction module is constructed based on a convolutional neural network and is used to extract electricity usage features from electricity usage data and input the electricity usage features into the sequence learning module;

[0038] The sequence learning module is constructed based on several LSTM layers and several Dropout layers, and each of the LSTM layers and Dropout layers is alternately connected; the sequence learning module is used to learn long-term dependencies from each power consumption feature, obtain global features, and input the global features into the output module;

[0039] The output module is used to output a power load forecast report based on global characteristics; the power load forecast report includes an AC charging pile power load curve and a residential power load curve for the next 6 hours.

[0040] Furthermore, the load forecasting model deployment module is specifically used to:

[0041] Obtain a large amount of historical electricity usage data including electricity usage time, electricity usage date, date type, AC charging station electricity usage data, residential electricity usage data, and environmental data; the date type is weekday, weekend, or holiday; the environmental data includes at least temperature, humidity, wind speed, and weather conditions;

[0042] Performing preprocessing on each of the historical electricity consumption data, including at least outlier removal, missing value filling, duplicate data removal, format unification, and data standardization, and constructing a data set based on each of the preprocessed historical electricity consumption data;

[0043] The data set is divided into a training set, a validation set, and a test set by a ten-fold cross-validation method. The load forecasting model is trained by the training set until the loss value of the loss function is less than a preset loss threshold, and the hyperparameters of the load forecasting model, including at least a learning rate, a random dropout rate, a batch size, and a number of iterations, are continuously optimized during the training process. The trained load forecasting model is then validated by the validation set to determine whether the prediction accuracy is greater than a preset accuracy threshold. If not, the validation fails, and the data set is expanded to continue training. If so, the validation succeeds, and:

[0044] The successfully verified load forecasting model is tested using the test set to determine whether the prediction confidence is greater than a preset confidence threshold. If not, the test fails and the data set is expanded to continue training; if so, the test succeeds, the training ends, and the load forecasting model is deployed.

[0045] Furthermore, the load forecasting module is specifically used to:

[0046] Acquire real-time electricity usage data including electricity usage time, electricity usage date, date type, AC charging pile electricity usage data, residential electricity usage data, and environmental data; the AC charging pile electricity usage data and residential electricity usage data are automatically acquired based on the smart meter;

[0047] The real-time electricity consumption data is preprocessed, including at least outlier removal, missing value filling, duplicate data removal, format unification, and data standardization, and then input into the deployed load forecasting model to obtain an electricity load forecast report including the AC charging pile electricity load curve and the residential electricity load curve for the next 6 hours.

[0048] Furthermore, the charging priority setting module is specifically used to:

[0049] Obtain historical charging data for all user accounts that have charged at AC charging stations in the community. Based on the historical charging data, extract the charging behavior of each user account, including at least charging frequency, average charging power, and charging duration. Set the charging priority of each user account based on the charging frequency, average charging power, and charging duration:

[0050] Charging priority = a / charging frequency + b*average charging power + c / charging time;

[0051] Among them, a, b, and c all represent weight coefficients;

[0052] The power distribution module is specifically used for:

[0053] Obtain the billing rules for electricity prices, set a load threshold, verify the electricity load forecast report based on the load threshold, select a time period in the next 6 hours when the electricity load exceeds the load threshold, and dynamically allocate the power of the AC charging pile within the time period based on the charging priority and billing rules. If the electricity load cannot be met, the photovoltaic energy storage system is used for power supply. If the photovoltaic energy storage system also cannot meet the electricity load, the power output of the AC charging pile corresponding to the user account with the lowest charging priority is shut down in turn until the electricity load is less than the load threshold.

[0054] The advantages of the present invention are:

[0055] 1. Create a load forecasting model through the data preprocessing module, feature extraction module, sequence learning module and output module, and set the loss function of the load forecasting model; then obtain a large amount of historical electricity consumption data, preprocess each historical electricity consumption data to build a data set, train the load forecasting model through the data set, and deploy the trained load forecasting model; then obtain real-time electricity consumption data, preprocess the real-time electricity consumption data and input it into the deployed load forecasting model to obtain the electricity load forecast report; obtain the historical charging data of all user accounts that have charged at AC charging piles in the community, extract the charging behavior of each user account based on the historical charging data, and set the loss function of each user account based on the charging behavior. charging priority; obtain the billing rules of the electricity price, and dynamically allocate the power of the AC charging pile based on the power load forecast report, charging priority and billing rules; that is, use the pre-trained load forecast model to predict the AC charging pile power load curve and the residential power load curve of the community in the next 6 hours, and comprehensively consider the power consumption of the AC charging pile and the residents' indoor electricity. When the predicted power load has the risk of being too large, the power of the AC charging pile is dynamically allocated within the corresponding time period based on the charging priority and billing rules. That is, while ensuring that the power load is not overloaded, the charging cost is minimized as much as possible, which ultimately greatly improves the rationality of the AC charging pile power allocation, thereby greatly improving the charging user experience.

[0056] 2. By setting up a sequence learning module in the load forecasting model to learn the long-term dependencies of the electricity consumption features output by the feature extraction module, the feature extraction performance is effectively improved, thereby greatly improving the accuracy of the load forecasting model prediction; and the end of each LSTM layer of the sequence learning module is connected to a Dropout layer, which can effectively prevent overfitting and enhance the model's generalization ability.

[0057] 3. The load forecasting model is trained using a data set constructed using electricity consumption time, electricity consumption date, date type, AC charging pile electricity consumption data, residential electricity consumption data, and environmental data. This allows the load forecasting model to learn multi-dimensional features and further improve the accuracy of the load forecasting model's predictions.

[0058] 4. By preprocessing the historical electricity consumption data, including at least outlier removal, missing value filling, duplicate data removal, format unification and data standardization, a data set is constructed based on the preprocessed historical electricity consumption data, which effectively improves the quality of the data set and greatly improves the training effect of the load forecasting model.

[0059] 5. By setting the loss function of the load forecasting model to the mean square error function, the difference between the predicted value and the true value can be effectively measured, thereby greatly improving the training effect of the load forecasting model.

[0060] 6. The data set is divided into training set, validation set and test set through the ten-fold cross-validation method. The load forecasting model is trained through the training set until the loss value of the loss function is less than the preset loss threshold, and the hyperparameters including at least the learning rate, random dropout rate, batch size and number of iterations are continuously optimized during the training process; then the trained load forecasting model is verified through the validation set, and the successfully verified load forecasting model is tested through the test set, and finally the load forecasting model that passes the test is deployed; since the ten-fold cross-validation method divides the data set into ten subsets, a different subset is selected each time as the validation set and test set, and the remaining subset is used as the training set, and rotated ten times, it can avoid overfitting or underfitting problems caused by the randomness of the data set division, and the load forecasting model is continuously verified, tested and optimized during the training process, thereby greatly improving the training speed and training effect of the load forecasting model.

[0061] 7. Set the charging priority of each user account through charging frequency, average charging power and charging time. That is, the higher the charging frequency, the lower the charging priority (a high charging frequency means that the user continues to charge before the battery is used up, and charging is not so urgent at this time); the higher the average charging power, the higher the charging priority (a high average charging power means that the user basically uses up the power of the electric vehicle before charging, and charging is more urgent at this time); the longer the charging time, the lower the charging priority (a long charging time means that there is a large room for power adjustment, and the power can be reduced when the load is heavy and increased when the load is light). This can match the charging needs of different users as much as possible, and give priority to allocating power to users who urgently need charging to improve the charging user experience.

[0062] 8. The load forecasting model adopts a hierarchical structure of data preprocessing module, feature extraction module (CNN), sequence learning module (LSTM+Dropout) and output module, combining the convolutional neural network's ability to extract local features with LSTM's ability to capture long-term dependencies in time series, significantly improving the forecast accuracy.

[0063] 9. Divide the data set through cross-validation and combine it with dynamic optimization of hyperparameters (learning rate, batch size, etc.) to ensure the scientific nature of model training, avoid overfitting or underfitting, and ensure model reliability.

[0064] 10. By combining time-of-use electricity prices and other billing rules, the charging power can be dynamically increased during low-load periods (low electricity prices) to reduce user costs; during peak load periods (high electricity prices or high grid pressure), the allocation strategy can be adjusted to balance grid stability and economy.

[0065] 11. Real-time electricity consumption data is collected through smart meters, and the pre-trained load forecasting model is input to quickly generate electricity load forecast reports, realizing minute-level dynamic power adjustment to adapt to grid load fluctuations. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0067] Figure 1 This is a flow chart of an AC charging pile power distribution method of the present invention.

[0068] Figure 2 It is a structural schematic diagram of an AC charging pile power distribution system of the present invention. DETAILED DESCRIPTION

[0069] The technical solution in the embodiments of the present application has the following overall idea: a pre-trained load prediction model is used to predict the AC charging pile power load curve and the residents' power load curve in the community for the next 6 hours, and the power consumption of the AC charging piles and residents' indoors is comprehensively considered. When the predicted power load has the risk of being too large, the power of the AC charging pile is dynamically allocated within the corresponding time period based on the charging priority and billing rules. That is, while ensuring that the power load is not overloaded, the charging cost is reduced as much as possible to improve the rationality of the AC charging pile power allocation and enhance the charging user experience.

[0070] Please refer to Figures 1 to 2 As shown, a preferred embodiment of an AC charging pile power distribution method of the present invention includes the following steps:

[0071] Step S1: Create a load forecasting model based on the data preprocessing module, feature extraction module, sequence learning module and output module, and set the loss function of the load forecasting model to the mean square error function; the data preprocessing module, feature extraction module, sequence learning module and output module are connected in sequence; by setting the loss function of the load forecasting model to the mean square error function, the difference between the predicted value and the true value can be effectively measured, thereby greatly improving the training effect of the load forecasting model.

[0072] Step S2: obtaining a large amount of historical electricity usage data, pre-processing each of the historical electricity usage data to construct a data set, training a load forecasting model using the data set, and deploying the trained load forecasting model;

[0073] Step S3: Acquire real-time electricity consumption data, pre-process the real-time electricity consumption data, and input the data into the deployed load forecasting model to obtain an electricity load forecast report;

[0074] Step S4: Obtain historical charging data of user accounts that have charged at all AC charging piles in the community, extract the charging behavior of each user account based on the historical charging data, and set the charging priority of each user account based on the charging behavior; extract user charging behavior (such as charging time period, frequency, and duration) based on the historical charging data, set charging priorities in a differentiated manner, give priority to users with high frequency or urgent needs, and improve user satisfaction.

[0075] Step S5: Obtain the electricity price billing rules, and dynamically allocate the power of the AC charging pile based on the electricity load forecast report, charging priority, and billing rules.

[0076] By integrating deep learning, user behavior analysis, and dynamic optimization strategies, this invention achieves high-precision load forecasting and intelligent power distribution. It combines technological advancement with practical value, and is in line with the development trend in the field of smart energy management.

[0077] In step S1, the data preprocessing module is used to clean and normalize the electricity consumption data, and input the normalized electricity consumption data into the feature extraction module; the normalization process eliminates dimensional differences and improves the model convergence efficiency;

[0078] The feature extraction module is constructed based on a convolutional neural network and is used to extract electricity usage features from electricity usage data and input the electricity usage features into the sequence learning module;

[0079] The sequence learning module is constructed based on several LSTM layers and several Dropout layers, and each of the LSTM layers and Dropout layers is alternately connected; the sequence learning module is used to learn long-term dependencies from each power consumption feature, obtain global features, and input the global features into the output module;

[0080] The output module is used to output a power load forecast report based on global characteristics; the power load forecast report includes an AC charging pile power load curve and a residential power load curve for the next 6 hours.

[0081] By setting up a sequence learning module for the load forecasting model to learn the long-term dependencies of the electricity consumption features output by the feature extraction module, the feature extraction performance is effectively improved, thereby greatly improving the accuracy of the load forecasting model prediction; and the end of each LSTM layer of the sequence learning module is connected to a Dropout layer, which can effectively prevent overfitting and enhance the model's generalization ability.

[0082] The load forecasting model adopts a hierarchical structure of data preprocessing module, feature extraction module (CNN), sequence learning module (LSTM+Dropout) and output module, combining the convolutional neural network's ability to extract local features with LSTM's ability to capture long-term dependencies in time series, significantly improving the forecast accuracy.

[0083] The step S2 is specifically as follows:

[0084] Acquire a large amount of historical electricity consumption data, including electricity consumption time, electricity consumption date, date type, AC charging pile electricity consumption data, residential electricity consumption data, and environmental data; the date type is weekday, weekend, or holiday; the environmental data includes at least temperature, humidity, wind speed, and weather conditions. By integrating multi-dimensional features such as electricity consumption time, date type (weekday / holiday), and environmental data (temperature, humidity, weather, etc.), the correlation between load changes and external factors is fully explored to improve the comprehensiveness of the forecast;

[0085] The historical electricity consumption data are subjected to at least preprocessing including outlier removal, missing value filling, duplicate data removal, format unification and data standardization, and a data set is constructed based on the preprocessed historical electricity consumption data; by preprocessing the historical electricity consumption data including at least outlier removal, missing value filling, duplicate data removal, format unification and data standardization, a data set is constructed based on the preprocessed historical electricity consumption data, thereby effectively improving the quality of the data set and greatly improving the training effect of the load forecasting model.

[0086] The data set is divided into a training set, a validation set, and a test set by a ten-fold cross-validation method. The load forecasting model is trained by the training set until the loss value of the loss function is less than a preset loss threshold, and the hyperparameters of the load forecasting model, including at least a learning rate, a random dropout rate, a batch size, and a number of iterations, are continuously optimized during the training process. The trained load forecasting model is then validated by the validation set to determine whether the prediction accuracy is greater than a preset accuracy threshold. If not, the validation fails, and the data set is expanded to continue training. If so, the validation succeeds, and:

[0087] The successfully verified load forecasting model is tested using the test set to determine whether the prediction confidence is greater than a preset confidence threshold. If not, the test fails and the data set is expanded to continue training; if so, the test succeeds, the training ends, and the load forecasting model is deployed.

[0088] The load forecasting model is trained by using a data set constructed from electricity consumption time, electricity consumption date, date type, AC charging pile electricity consumption data, residential electricity consumption data, and environmental data, so that the load forecasting model can learn multi-dimensional features and further improve the accuracy of the load forecasting model's predictions.

[0089] The data set is divided into training set, validation set and test set through the ten-fold cross-validation method. The load forecasting model is trained with the training set until the loss value of the loss function is less than the preset loss threshold, and the hyperparameters including at least the learning rate, random dropout rate, batch size and number of iterations are continuously optimized during the training process; then the trained load forecasting model is verified with the validation set, and the verified load forecasting model is tested with the test set, and finally the load forecasting model that passes the test is deployed; since the ten-fold cross-validation method divides the data set into ten subsets, a different subset is selected each time as the validation set and the test set, and the remaining subset is used as the training set, and rotated ten times, it can avoid overfitting or underfitting problems caused by the randomness of the data set division, and the load forecasting model is continuously verified, tested and optimized during the training process, thereby greatly improving the training speed and training effect of the load forecasting model.

[0090] By dividing the data set through cross-validation and combining it with dynamic optimization of hyperparameters (learning rate, batch size, etc.), we can ensure the scientific nature of model training, avoid overfitting or underfitting, and guarantee model reliability.

[0091] The step S3 is specifically as follows:

[0092] Acquire real-time electricity consumption data including electricity consumption time, electricity consumption date, date type, AC charging pile electricity consumption data, residential electricity consumption data and environmental data; the AC charging pile electricity consumption data and residential electricity consumption data are automatically acquired based on smart meters; collect real-time electricity consumption data through smart meters, input pre-trained load forecasting models to quickly generate electricity load forecast reports, realize minute-level dynamic power adjustment, and adapt to grid load fluctuations.

[0093] The real-time electricity consumption data is preprocessed, including at least outlier removal, missing value filling, duplicate data removal, format unification, and data standardization, and then input into the deployed load forecasting model to obtain an electricity load forecast report including the AC charging pile electricity load curve and the residential electricity load curve for the next 6 hours.

[0094] The step S4 is specifically as follows:

[0095] Obtain historical charging data for all user accounts that have charged at AC charging stations in the community. Based on the historical charging data, extract the charging behavior of each user account, including at least charging frequency, average charging power, and charging duration. Set the charging priority of each user account based on the charging frequency, average charging power, and charging duration:

[0096] Charging priority = a / charging frequency + b*average charging power + c / charging time;

[0097] Among them, a, b, and c all represent weight coefficients;

[0098] The charging priority of each user account is set by charging frequency, average charging power and charging time. That is, the higher the charging frequency, the lower the charging priority (a high charging frequency means that the user continues to charge before the battery is used up, and charging is not so urgent at this time); the higher the average charging power, the higher the charging priority (a high average charging power means that the user basically uses up the power of the electric vehicle before charging, and charging is more urgent at this time); the longer the charging time, the lower the charging priority (a long charging time means that there is a large room for power adjustment, and the power can be reduced when the load is heavy and increased when the load is light). This can match the charging needs of different users as much as possible, and give priority to allocating power to users who urgently need charging, so as to improve the charging user experience.

[0099] The step S5 is specifically as follows:

[0100] Obtain the billing rules for electricity prices, set a load threshold, verify the electricity load forecast report based on the load threshold, select a time period in the next 6 hours when the electricity load exceeds the load threshold, and dynamically allocate the power of the AC charging pile based on the charging priority and billing rules within the time period. If the electricity load cannot be met, the photovoltaic energy storage system is combined to supply power. If the photovoltaic energy storage system cannot meet the electricity load, the power output of the AC charging pile corresponding to the user account with the lowest charging priority is turned off in turn until the electricity load is less than the load threshold. By combining billing rules such as time-of-use electricity prices, the charging power is dynamically increased during low load periods (low electricity prices) to reduce user costs; during peak load periods (high electricity prices or high grid pressure), the allocation strategy is adjusted to balance grid stability and economy.

[0101] A preferred embodiment of an AC charging pile power distribution system of the present invention includes the following modules:

[0102] A load forecasting model creation module is used to create a load forecasting model based on a data preprocessing module, a feature extraction module, a sequence learning module and an output module, and set the loss function of the load forecasting model to a mean square error function; the data preprocessing module, the feature extraction module, the sequence learning module and the output module are connected in sequence; by setting the loss function of the load forecasting model to a mean square error function, the difference between the predicted value and the true value can be effectively measured, thereby greatly improving the training effect of the load forecasting model.

[0103] A load forecasting model deployment module is used to obtain a large amount of historical electricity usage data, pre-process each of the historical electricity usage data to construct a data set, train the load forecasting model using the data set, and deploy the trained load forecasting model;

[0104] A load forecasting module is used to obtain real-time power consumption data, pre-process the real-time power consumption data, and input the data into the deployed load forecasting model to obtain a power load forecast report;

[0105] The charging priority setting module is used to obtain the historical charging data of user accounts that have charged at all AC charging piles in the community, extract the charging behavior of each user account based on the historical charging data, and set the charging priority of each user account based on the charging behavior; extract the user charging behavior (such as charging time period, frequency, and duration) based on the historical charging data, and set the charging priority in a differentiated manner, giving priority to users with high frequency or urgent needs, thereby improving user satisfaction.

[0106] The power allocation module is used to obtain the billing rules of the electricity price and dynamically allocate the power of the AC charging pile based on the power load forecast report, charging priority and billing rules.

[0107] By integrating deep learning, user behavior analysis, and dynamic optimization strategies, this invention achieves high-precision load forecasting and intelligent power distribution. It combines technological advancement with practical value, and is in line with the development trend in the field of smart energy management.

[0108] In the load forecasting model creation module, the data preprocessing module is used to clean and normalize the electricity consumption data and input the normalized electricity consumption data into the feature extraction module; the normalization process eliminates dimensional differences and improves the model convergence efficiency;

[0109] The feature extraction module is constructed based on a convolutional neural network and is used to extract electricity usage features from electricity usage data and input the electricity usage features into the sequence learning module;

[0110] The sequence learning module is constructed based on several LSTM layers and several Dropout layers, and each of the LSTM layers and Dropout layers is alternately connected; the sequence learning module is used to learn long-term dependencies from each power consumption feature, obtain global features, and input the global features into the output module;

[0111] The output module is used to output a power load forecast report based on global characteristics; the power load forecast report includes an AC charging pile power load curve and a residential power load curve for the next 6 hours.

[0112] By setting up a sequence learning module for the load forecasting model to learn the long-term dependencies of the electricity consumption features output by the feature extraction module, the feature extraction performance is effectively improved, thereby greatly improving the accuracy of the load forecasting model prediction; and the end of each LSTM layer of the sequence learning module is connected to a Dropout layer, which can effectively prevent overfitting and enhance the model's generalization ability.

[0113] The load forecasting model adopts a hierarchical structure of data preprocessing module, feature extraction module (CNN), sequence learning module (LSTM+Dropout) and output module, combining the convolutional neural network's ability to extract local features with LSTM's ability to capture long-term dependencies in time series, significantly improving the forecast accuracy.

[0114] The load forecasting model deployment module is specifically used to:

[0115] Acquire a large amount of historical electricity consumption data, including electricity consumption time, electricity consumption date, date type, AC charging pile electricity consumption data, residential electricity consumption data, and environmental data; the date type is weekday, weekend, or holiday; the environmental data includes at least temperature, humidity, wind speed, and weather conditions. By integrating multi-dimensional features such as electricity consumption time, date type (weekday / holiday), and environmental data (temperature, humidity, weather, etc.), the correlation between load changes and external factors is fully explored to improve the comprehensiveness of the forecast;

[0116] The historical electricity consumption data are subjected to at least preprocessing including outlier removal, missing value filling, duplicate data removal, format unification and data standardization, and a data set is constructed based on the preprocessed historical electricity consumption data; by preprocessing the historical electricity consumption data including at least outlier removal, missing value filling, duplicate data removal, format unification and data standardization, a data set is constructed based on the preprocessed historical electricity consumption data, thereby effectively improving the quality of the data set and greatly improving the training effect of the load forecasting model.

[0117] The data set is divided into a training set, a validation set, and a test set by a ten-fold cross-validation method. The load forecasting model is trained by the training set until the loss value of the loss function is less than a preset loss threshold, and the hyperparameters of the load forecasting model, including at least a learning rate, a random dropout rate, a batch size, and a number of iterations, are continuously optimized during the training process. The trained load forecasting model is then validated by the validation set to determine whether the prediction accuracy is greater than a preset accuracy threshold. If not, the validation fails, and the data set is expanded to continue training. If so, the validation succeeds, and:

[0118] The successfully verified load forecasting model is tested using the test set to determine whether the prediction confidence is greater than a preset confidence threshold. If not, the test fails and the data set is expanded to continue training; if so, the test succeeds, the training ends, and the load forecasting model is deployed.

[0119] The load forecasting model is trained by using a data set constructed from electricity consumption time, electricity consumption date, date type, AC charging pile electricity consumption data, residential electricity consumption data, and environmental data, so that the load forecasting model can learn multi-dimensional features and further improve the accuracy of the load forecasting model's predictions.

[0120] The data set is divided into training set, validation set and test set through the ten-fold cross-validation method. The load forecasting model is trained with the training set until the loss value of the loss function is less than the preset loss threshold, and the hyperparameters including at least the learning rate, random dropout rate, batch size and number of iterations are continuously optimized during the training process; then the trained load forecasting model is verified with the validation set, and the verified load forecasting model is tested with the test set, and finally the load forecasting model that passes the test is deployed; since the ten-fold cross-validation method divides the data set into ten subsets, a different subset is selected each time as the validation set and the test set, and the remaining subset is used as the training set, and rotated ten times, it can avoid overfitting or underfitting problems caused by the randomness of the data set division, and the load forecasting model is continuously verified, tested and optimized during the training process, thereby greatly improving the training speed and training effect of the load forecasting model.

[0121] By dividing the data set through cross-validation and combining it with dynamic optimization of hyperparameters (learning rate, batch size, etc.), we can ensure the scientific nature of model training, avoid overfitting or underfitting, and guarantee model reliability.

[0122] The load forecasting module is specifically used for:

[0123] Acquire real-time electricity consumption data including electricity consumption time, electricity consumption date, date type, AC charging pile electricity consumption data, residential electricity consumption data and environmental data; the AC charging pile electricity consumption data and residential electricity consumption data are automatically acquired based on smart meters; collect real-time electricity consumption data through smart meters, input pre-trained load forecasting models to quickly generate electricity load forecast reports, realize minute-level dynamic power adjustment, and adapt to grid load fluctuations.

[0124] The real-time electricity consumption data is preprocessed, including at least outlier removal, missing value filling, duplicate data removal, format unification, and data standardization, and then input into the deployed load forecasting model to obtain an electricity load forecast report including the AC charging pile electricity load curve and the residential electricity load curve for the next 6 hours.

[0125] The charging priority setting module is specifically used to:

[0126] Obtain historical charging data for all user accounts that have charged at AC charging stations in the community. Based on the historical charging data, extract the charging behavior of each user account, including at least charging frequency, average charging power, and charging duration. Set the charging priority of each user account based on the charging frequency, average charging power, and charging duration:

[0127] Charging priority = a / charging frequency + b*average charging power + c / charging time;

[0128] Among them, a, b, and c all represent weight coefficients;

[0129] The charging priority of each user account is set by charging frequency, average charging power and charging time. That is, the higher the charging frequency, the lower the charging priority (a high charging frequency means that the user continues to charge before the battery is used up, and charging is not so urgent at this time); the higher the average charging power, the higher the charging priority (a high average charging power means that the user basically uses up the power of the electric vehicle before charging, and charging is more urgent at this time); the longer the charging time, the lower the charging priority (a long charging time means that there is a large room for power adjustment, and the power can be reduced when the load is heavy and increased when the load is light). This can match the charging needs of different users as much as possible, and give priority to allocating power to users who urgently need charging, so as to improve the charging user experience.

[0130] The power distribution module is specifically used for:

[0131] Obtain the billing rules for electricity prices, set a load threshold, verify the electricity load forecast report based on the load threshold, select a time period in the next 6 hours when the electricity load exceeds the load threshold, and dynamically allocate the power of the AC charging pile based on the charging priority and billing rules within the time period. If the electricity load cannot be met, the photovoltaic energy storage system is combined to supply power. If the photovoltaic energy storage system cannot meet the electricity load, the power output of the AC charging pile corresponding to the user account with the lowest charging priority is turned off in turn until the electricity load is less than the load threshold. By combining billing rules such as time-of-use electricity prices, the charging power is dynamically increased during low load periods (low electricity prices) to reduce user costs; during peak load periods (high electricity prices or high grid pressure), the allocation strategy is adjusted to balance grid stability and economy.

[0132] In summary, the advantages of the present invention are:

[0133] 1. Create a load forecasting model through the data preprocessing module, feature extraction module, sequence learning module and output module, and set the loss function of the load forecasting model; then obtain a large amount of historical electricity consumption data, preprocess each historical electricity consumption data to build a data set, train the load forecasting model through the data set, and deploy the trained load forecasting model; then obtain real-time electricity consumption data, preprocess the real-time electricity consumption data and input it into the deployed load forecasting model to obtain the electricity load forecast report; obtain the historical charging data of all user accounts that have charged at AC charging piles in the community, extract the charging behavior of each user account based on the historical charging data, and set the loss function of each user account based on the charging behavior. charging priority; obtain the billing rules of the electricity price, and dynamically allocate the power of the AC charging pile based on the power load forecast report, charging priority and billing rules; that is, use the pre-trained load forecast model to predict the AC charging pile power load curve and the residential power load curve of the community in the next 6 hours, and comprehensively consider the power consumption of the AC charging pile and the residents' indoor electricity. When the predicted power load has the risk of being too large, the power of the AC charging pile is dynamically allocated within the corresponding time period based on the charging priority and billing rules. That is, while ensuring that the power load is not overloaded, the charging cost is minimized as much as possible, which ultimately greatly improves the rationality of the AC charging pile power allocation, thereby greatly improving the charging user experience.

[0134] 2. By setting up a sequence learning module in the load forecasting model to learn the long-term dependencies of the electricity consumption features output by the feature extraction module, the feature extraction performance is effectively improved, thereby greatly improving the accuracy of the load forecasting model prediction; and the end of each LSTM layer of the sequence learning module is connected to a Dropout layer, which can effectively prevent overfitting and enhance the model's generalization ability.

[0135] 3. The load forecasting model is trained using a data set constructed using electricity consumption time, electricity consumption date, date type, AC charging pile electricity consumption data, residential electricity consumption data, and environmental data. This allows the load forecasting model to learn multi-dimensional features and further improve the accuracy of the load forecasting model's predictions.

[0136] 4. By preprocessing the historical electricity consumption data, including at least outlier removal, missing value filling, duplicate data removal, format unification and data standardization, a data set is constructed based on the preprocessed historical electricity consumption data, which effectively improves the quality of the data set and greatly improves the training effect of the load forecasting model.

[0137] 5. By setting the loss function of the load forecasting model to the mean square error function, the difference between the predicted value and the true value can be effectively measured, thereby greatly improving the training effect of the load forecasting model.

[0138] 6. The data set is divided into training set, validation set and test set through the ten-fold cross-validation method. The load forecasting model is trained through the training set until the loss value of the loss function is less than the preset loss threshold, and the hyperparameters including at least the learning rate, random dropout rate, batch size and number of iterations are continuously optimized during the training process; then the trained load forecasting model is verified through the validation set, and the successfully verified load forecasting model is tested through the test set, and finally the load forecasting model that passes the test is deployed; since the ten-fold cross-validation method divides the data set into ten subsets, a different subset is selected each time as the validation set and test set, and the remaining subset is used as the training set, and rotated ten times, it can avoid overfitting or underfitting problems caused by the randomness of the data set division, and the load forecasting model is continuously verified, tested and optimized during the training process, thereby greatly improving the training speed and training effect of the load forecasting model.

[0139] 7. Set the charging priority of each user account through charging frequency, average charging power and charging time. That is, the higher the charging frequency, the lower the charging priority (a high charging frequency means that the user continues to charge before the battery is used up, and charging is not so urgent at this time); the higher the average charging power, the higher the charging priority (a high average charging power means that the user basically uses up the power of the electric vehicle before charging, and charging is more urgent at this time); the longer the charging time, the lower the charging priority (a long charging time means that there is a large room for power adjustment, and the power can be reduced when the load is heavy and increased when the load is light). This can match the charging needs of different users as much as possible, and give priority to allocating power to users who urgently need charging to improve the charging user experience.

[0140] 8. The load forecasting model adopts a hierarchical structure of data preprocessing module, feature extraction module (CNN), sequence learning module (LSTM+Dropout) and output module, combining the convolutional neural network's ability to extract local features with LSTM's ability to capture long-term dependencies in time series, significantly improving the forecast accuracy.

[0141] 9. Divide the data set through cross-validation and combine it with dynamic optimization of hyperparameters (learning rate, batch size, etc.) to ensure the scientific nature of model training, avoid overfitting or underfitting, and ensure model reliability.

[0142] 10. By combining time-of-use electricity prices and other billing rules, the charging power can be dynamically increased during low-load periods (low electricity prices) to reduce user costs; during peak load periods (high electricity prices or high grid pressure), the allocation strategy can be adjusted to balance grid stability and economy.

[0143] 11. Real-time electricity consumption data is collected through smart meters, and the pre-trained load forecasting model is input to quickly generate electricity load forecast reports, realizing minute-level dynamic power adjustment to adapt to grid load fluctuations.

[0144] Although the specific embodiments of the present invention are described above, those skilled in the art should understand that the specific embodiments described are merely illustrative and are not intended to limit the scope of the present invention. Equivalent modifications and changes made by those skilled in the art in accordance with the spirit of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A method for distributing power to an AC charging pile, characterized by: The steps include: Step S1: creating a load forecasting model based on a data preprocessing module, a feature extraction module, a sequence learning module, and an output module, setting the loss function of the load forecasting model to a mean square error function; the data preprocessing module, the feature extraction module, the sequence learning module, and the output module are connected in sequence; Step S2: obtaining a large amount of historical electricity usage data, pre-processing each of the historical electricity usage data to construct a data set, training a load forecasting model using the data set, and deploying the trained load forecasting model; Step S3: Acquire real-time electricity consumption data, pre-process the real-time electricity consumption data, and input the data into the deployed load forecasting model to obtain an electricity load forecast report; Step S4: Obtain historical charging data of user accounts that have charged at all AC charging piles in the community, extract charging behavior of each user account based on the historical charging data, and set a charging priority for each user account based on the charging behavior; Step S5: Obtain the electricity price billing rules, and dynamically allocate the power of the AC charging pile based on the electricity load forecast report, charging priority, and billing rules.

2. The AC charging pile power distribution method according to claim 1, wherein: In step S1, the data preprocessing module is used to clean and normalize the electricity consumption data, and input the normalized electricity consumption data into the feature extraction module; The feature extraction module is constructed based on a convolutional neural network and is used to extract electricity usage features from electricity usage data and input the electricity usage features into the sequence learning module; The sequence learning module is constructed based on several LSTM layers and several Dropout layers, and each of the LSTM layers and Dropout layers is alternately connected; the sequence learning module is used to learn long-term dependencies from each power consumption feature, obtain global features, and input the global features into the output module; The output module is used to output a power load forecast report based on global characteristics; the power load forecast report includes an AC charging pile power load curve and a residential power load curve for the next 6 hours.

3. The AC charging pile power distribution method according to claim 1, wherein: The step S2 is specifically as follows: Obtain a large amount of historical electricity usage data including electricity usage time, electricity usage date, date type, AC charging station electricity usage data, residential electricity usage data, and environmental data; the date type is weekday, weekend, or holiday; the environmental data includes at least temperature, humidity, wind speed, and weather conditions; Performing preprocessing on each of the historical electricity consumption data, including at least outlier removal, missing value filling, duplicate data removal, format unification, and data standardization, and constructing a data set based on each of the preprocessed historical electricity consumption data; The data set is divided into a training set, a validation set, and a test set by a ten-fold cross-validation method. The load forecasting model is trained by the training set until the loss value of the loss function is less than a preset loss threshold, and the hyperparameters of the load forecasting model, including at least a learning rate, a random dropout rate, a batch size, and a number of iterations, are continuously optimized during the training process. The trained load forecasting model is then validated by the validation set to determine whether the prediction accuracy is greater than a preset accuracy threshold. If not, the validation fails, and the data set is expanded to continue training. If so, the validation succeeds, and: The successfully verified load forecasting model is tested using the test set to determine whether the prediction confidence is greater than a preset confidence threshold. If not, the test fails and the data set is expanded to continue training; if so, the test succeeds, the training ends, and the load forecasting model is deployed.

4. The AC charging pile power distribution method according to claim 1, wherein: The step S3 is specifically as follows: Acquire real-time electricity usage data including electricity usage time, electricity usage date, date type, AC charging pile electricity usage data, residential electricity usage data, and environmental data; the AC charging pile electricity usage data and residential electricity usage data are automatically acquired based on the smart meter; The real-time electricity consumption data is preprocessed, including at least outlier removal, missing value filling, duplicate data removal, format unification, and data standardization, and then input into the deployed load forecasting model to obtain an electricity load forecast report including the AC charging pile electricity load curve and the residential electricity load curve for the next 6 hours.

5. The AC charging pile power distribution method according to claim 1, wherein: The step S4 is specifically as follows: Obtain historical charging data for all user accounts that have charged at AC charging stations in the community. Based on the historical charging data, extract the charging behavior of each user account, including at least charging frequency, average charging power, and charging duration. Set the charging priority of each user account based on the charging frequency, average charging power, and charging duration: Charging priority = a / charging frequency + b*average charging power + c / charging time; Among them, a, b, and c all represent weight coefficients; The step S5 is specifically as follows: Obtain the billing rules for electricity prices, set a load threshold, verify the electricity load forecast report based on the load threshold, select a time period in the next 6 hours when the electricity load exceeds the load threshold, and dynamically allocate the power of the AC charging pile within the time period based on the charging priority and billing rules. If the electricity load cannot be met, the photovoltaic energy storage system is used for power supply. If the photovoltaic energy storage system also cannot meet the electricity load, the power output of the AC charging pile corresponding to the user account with the lowest charging priority is shut down in turn until the electricity load is less than the load threshold.

6. An AC charging pile power distribution system, characterized by: Includes the following modules: A load forecasting model creation module, configured to create a load forecasting model based on a data preprocessing module, a feature extraction module, a sequence learning module, and an output module, wherein the loss function of the load forecasting model is set to a mean square error function; the data preprocessing module, the feature extraction module, the sequence learning module, and the output module are connected in sequence; A load forecasting model deployment module is used to obtain a large amount of historical electricity usage data, pre-process each of the historical electricity usage data to construct a data set, train the load forecasting model using the data set, and deploy the trained load forecasting model; A load forecasting module is used to obtain real-time power consumption data, pre-process the real-time power consumption data, and input the data into the deployed load forecasting model to obtain a power load forecast report; A charging priority setting module is used to obtain historical charging data of user accounts that have charged at all AC charging piles in the community, extract the charging behavior of each user account based on the historical charging data, and set the charging priority of each user account based on the charging behavior; The power allocation module is used to obtain the billing rules of the electricity price and dynamically allocate the power of the AC charging pile based on the power load forecast report, charging priority and billing rules.

7. The AC charging pile power distribution system according to claim 6, characterized in that: In the load forecasting model creation module, the data preprocessing module is used to clean and normalize the electricity consumption data, and input the normalized electricity consumption data into the feature extraction module; The feature extraction module is constructed based on a convolutional neural network and is used to extract electricity usage features from electricity usage data and input the electricity usage features into the sequence learning module; The sequence learning module is constructed based on several LSTM layers and several Dropout layers, and each of the LSTM layers and Dropout layers is alternately connected; the sequence learning module is used to learn long-term dependencies from each power consumption feature, obtain global features, and input the global features into the output module; The output module is used to output a power load forecast report based on global characteristics; the power load forecast report includes an AC charging pile power load curve and a residential power load curve for the next 6 hours.

8. The AC charging pile power distribution system according to claim 6, characterized in that: The load forecasting model deployment module is specifically used to: Obtain a large amount of historical electricity usage data including electricity usage time, electricity usage date, date type, AC charging station electricity usage data, residential electricity usage data, and environmental data; the date type is weekday, weekend, or holiday; the environmental data includes at least temperature, humidity, wind speed, and weather conditions; Performing preprocessing on each of the historical electricity consumption data, including at least outlier removal, missing value filling, duplicate data removal, format unification, and data standardization, and constructing a data set based on each of the preprocessed historical electricity consumption data; The data set is divided into a training set, a validation set, and a test set by a ten-fold cross-validation method. The load forecasting model is trained by the training set until the loss value of the loss function is less than a preset loss threshold, and the hyperparameters of the load forecasting model, including at least a learning rate, a random dropout rate, a batch size, and a number of iterations, are continuously optimized during the training process. The trained load forecasting model is then validated by the validation set to determine whether the prediction accuracy is greater than a preset accuracy threshold. If not, the validation fails, and the data set is expanded to continue training. If so, the validation succeeds, and: The successfully verified load forecasting model is tested using the test set to determine whether the prediction confidence is greater than a preset confidence threshold. If not, the test fails and the data set is expanded to continue training; if so, the test succeeds, the training ends, and the load forecasting model is deployed.

9. The AC charging pile power distribution system according to claim 6, characterized in that: The load forecasting module is specifically used for: Acquire real-time electricity usage data including electricity usage time, electricity usage date, date type, AC charging pile electricity usage data, residential electricity usage data, and environmental data; the AC charging pile electricity usage data and residential electricity usage data are automatically acquired based on the smart meter; The real-time electricity consumption data is preprocessed, including at least outlier removal, missing value filling, duplicate data removal, format unification, and data standardization, and then input into the deployed load forecasting model to obtain an electricity load forecast report including the AC charging pile electricity load curve and the residential electricity load curve for the next 6 hours.

10. The AC charging pile power distribution system according to claim 6, characterized in that: The charging priority setting module is specifically used to: Obtain historical charging data for all user accounts that have charged at AC charging stations in the community. Based on the historical charging data, extract the charging behavior of each user account, including at least charging frequency, average charging power, and charging duration. Set the charging priority of each user account based on the charging frequency, average charging power, and charging duration: Charging priority = a / charging frequency + b*average charging power + c / charging time; Among them, a, b, and c all represent weight coefficients; The power distribution module is specifically used for: Obtain the billing rules for electricity prices, set a load threshold, verify the electricity load forecast report based on the load threshold, select a time period in the next 6 hours when the electricity load exceeds the load threshold, and dynamically allocate the power of the AC charging pile within the time period based on the charging priority and billing rules. If the electricity load cannot be met, the photovoltaic energy storage system is used for power supply. If the photovoltaic energy storage system also cannot meet the electricity load, the power output of the AC charging pile corresponding to the user account with the lowest charging priority is shut down in turn until the electricity load is less than the load threshold.

Images