Reservoir-computed multi-station charging load distributed prediction method and system

Abstract

The application discloses a kind of based on reservoir computing multi-station charging load distributed prediction method and system, the method first obtains the historical data of multiple electric vehicle charging stations charging load, and pretreatment is carried out;Then the spatiotemporal coupling characteristics between electric vehicle charging stations distributed in different places on the basis of similarity matrix of Gaussian kernel function is evaluated, and the distributed station set is constructed according to coupling characteristics;Finally, for the associated distributed station set, respectively establish the distributed prediction network based on reservoir computing paradigm, with the station load data in the distributed station set as input, as output, and use the distributed learning algorithm based on approximate synchronous parallel mechanism to carry out distributed training of prediction network;Real-time prediction is carried out using the trained prediction network.The application realizes the efficient cooperation of multi-station data and charging load dynamic evolution modeling, and has an important role for further promoting traffic electrification and grid intelligentization.

Description

Technical Field

[0001] This invention relates to the field of new energy vehicle charging station technology, and more specifically, it is a distributed prediction method and system for multi-station charging load based on reservoir computing. Background Technology

[0002] With the global energy structure transformation, the number of new energy vehicles has experienced explosive growth, with electric vehicles (EVs) dominating the market. Consequently, related industries such as EV charging equipment are also booming. Charging infrastructure includes AC slow charging stations, DC slow charging stations, DC fast charging stations, and DC supercharging stations. Among these, DC fast charging stations and DC supercharging stations are highly favored by consumers for their high charging power and fast charging speed, and are widely deployed in public charging stations.

[0003] With the continuous growth in electric vehicle sales, the surge in charging demand has led to increasingly prominent problems such as widening peak-valley differences in the power distribution network and overload operation in some areas. Accurate power prediction of charging loads at multiple charging stations within a region has become a core requirement for smart grid dispatching. Traditional prediction methods face significant challenges in addressing the spatiotemporal heterogeneity of multiple charging stations and the dynamic coupling characteristics of loads: 1. Insufficient spatiotemporal feature extraction. Existing models mostly employ single time series analysis, making it difficult to capture the load distribution characteristics of different charging stations due to differences in geographical location and user behavior; 2. Low data collaboration efficiency. Traditional centralized prediction frameworks suffer from data transmission delays and privacy risks between multiple charging stations, while single-station models cannot effectively utilize the load correlation between multiple charging stations; 3. Weak dynamic adaptability. Traditional statistical models (such as ARIMA) are sensitive to external disturbances such as sudden weather changes and holidays, while machine learning models (such as LSTM) can capture time series features, but parameter training is complex and difficult to adapt to heterogeneous data from multiple charging stations.

[0004] Current technologies, based on reservoir computation, have demonstrated advantages in time-series modeling. Employing the dynamic reservoir characteristics of reservoir computation allows for the preservation of the spatiotemporal features of input signals through random mapping, providing a new approach for multi-site load forecasting. However, existing distributed forecasting systems often employ federated learning architectures, which, while ensuring data privacy, cannot simultaneously guarantee real-time performance.

[0005] Therefore, it is urgent to build a distributed prediction system based on the reservoir computing paradigm to achieve efficient collaboration of data from multiple power stations and dynamic evolution modeling of charging load. This will play a crucial role in further developing my country's electric vehicle industry and promoting transportation electrification and power grid intelligence. Summary of the Invention

[0006] The purpose of this invention is to address the shortcomings of existing technologies by proposing a distributed prediction method and system for multi-site charging load based on reservoir computing.

[0007] The objective of this invention is achieved through the following technical solution: a distributed prediction method for multi-station charging load based on reservoir calculation, the method comprising:

[0008] Step 1: Obtain historical charging load data from multiple electric vehicle charging stations and perform preprocessing.

[0009] Step 2: Evaluate the spatiotemporal coupling characteristics between electric vehicle charging stations located in different spatial locations based on the similarity matrix of the Gaussian kernel function, and construct a distributed site set based on the coupling characteristics;

[0010] Step 3: Establish a distributed prediction network based on the reservoir computing paradigm, using the site load data in the distributed site set as input and the matrix weight as output, and use a distributed learning algorithm based on an approximate synchronous parallel mechanism for distributed training of the prediction network.

[0011] Step 4: For each set of associated distributed stations, construct and train a distributed prediction network in turn according to Step 3, and use the trained prediction network to make real-time predictions.

[0012] Furthermore, in step 1, the historical data of the charging load of the electric vehicle charging stations is the collection of load data for each charging station; the preprocessing includes: cleaning outliers and missing values ​​in the data, data normalization processing; and time alignment and interpolation processing of the historical data of multiple charging stations.

[0013] Furthermore, in step 2, based on the historical load time series data of charging stations from multiple sites, the Gaussian kernel function values ​​and similarities between load data points are calculated, and a similarity matrix is ​​constructed.

[0014] Furthermore, in step 3, an echo state network is established based on the reservoir computing paradigm as a single charging station charging load prediction model, and a distributed prediction network is constructed to address the load prediction problem of multiple charging station clusters.

[0015] Furthermore, in step 3, the charging station cluster needs to integrate historical load data from different stations to establish the augmented Lagrangian function of the prediction network. :

[0016]

[0017] in, Represents the first in the charging station cluster Historical data on the load of each charging station Representing the first in the charging station cluster The hidden layer output of the neural network for each charging station. For the first in the charging station cluster The output layer weight matrix of the neural network for each charging station. It is the target value of the output layer weight matrix. , For system hyperparameters, It is the parameter of the penalty term of the Lagrange function. The number of charging stations representing a distributed charging station cluster. Let T be a Lagrange multiplier vector, and let T denote the transpose.

[0018] The prediction network parameters are obtained by training it based on the target values ​​of the output layer weight matrix using the following method:

[0019]

[0020] in This is the estimated value of the output layer weight matrix during the k-th iteration of training.

[0021] Furthermore, a distributed average consensus protocol based on an approximate synchronous parallel mechanism is established to achieve distributed collaborative training of multiple charging station cluster systems. The specific process is as follows: randomly initialize the output layer weight matrix, target value, and Lagrange multiplier vector parameters. The output layer weight matrix is ​​calculated iteratively at the local site for the (k+1)th iteration. , To achieve local weight updates:

[0022]

[0023] in Let be the Lagrange multiplier vector for the k-th iteration. Represents the identity matrix.

[0024] Further, in step 3, define and Represent and The mean of the two is needed to obtain global information during distributed collaborative training. Global consensus is achieved among the sub-sites through a distributed average consensus protocol.

[0025]

[0026]

[0027] in, This represents the communication network topology connection coefficient between sub-site i and sub-site j, where n represents the number of communications. The maximum degree is determined based on the communication network topology diagram of the charging station cluster. Calculate the communication coefficient between different stations Conversely, when there is no communication between different stations, this value is 0; ,in Represents the node communication topology graph. The degree of each charging station node; then when the number of communications... Data for each node and It will converge to the mean of all data. and The charging station cluster communicates multiple times and follows a matrix pattern. The elements of the transmission and The value is then updated to obtain the mean.

[0028] Furthermore, in step 3, the distributed average consensus protocol introduces an Approximate Synchronous Parallel (ASP) mechanism. For each node, communication occurs when each update exceeds a set threshold, and distributed global consensus calculation is performed; otherwise, it is necessary to confirm whether the node communication is normal. During the iteration process, the threshold is dynamically adjusted.

[0029] Furthermore, in step 3, after obtaining global consensus through multiple communications, the global consensus variables are updated on the local site. and local multipliers Continue until the convergence condition is met;

[0030]

[0031]

[0032] During distributed training, and All updates are performed locally, eliminating the need to access global information and maintaining data privacy. Only updates are performed locally. When it is necessary to obtain global weight mean information, only the output weight matrix with partial information needs to be transmitted, and it does not participate in the transmission of directly sensitive data.

[0033] On the other hand, the present invention also provides a distributed forecasting system for multi-site charging load based on reservoir computing, the system comprising:

[0034] The data acquisition module is used to obtain historical load data from each electric vehicle charging station and perform preprocessing.

[0035] The selected module evaluates the spatiotemporal coupling characteristics between electric vehicle charging stations located in different spatial locations based on the similarity matrix of the Gaussian kernel function, and constructs a distributed site set based on the coupling characteristics.

[0036] The communication module is used for network communication during collaborative prediction in the distributed prediction system of electric vehicle charging station clusters.

[0037] The training module is used to establish a distributed prediction network based on the reservoir computing paradigm. It takes the site load data in the distributed site set as input, the matrix weight as output, and uses a distributed learning algorithm based on an approximate synchronous parallel mechanism to perform distributed training of the prediction network.

[0038] The prediction module is used to make real-time predictions using the trained prediction network.

[0039] The beneficial effects of this invention are:

[0040] 1. This invention adopts an echo state network prediction method based on reservoir computing. The system is lightweight and easy to deploy, and can be easily deployed and expanded in large-scale regional multi-site electric vehicle charging clusters.

[0041] 2. This invention designs a distributed network prediction model that considers both the security and efficiency of system information network transmission. During the process of obtaining global consensus, it does not participate in the transmission of directly sensitive historical data, but only transmits the anonymized output weight matrix data. This satisfies privacy protection requirements and the ability to deploy on resource-constrained edge devices.

[0042] 3. The multi-site electric vehicle charging cluster prediction system model of the present invention takes the system as a whole and is based on distributed collaborative training of local variable information interaction among multiple charging stations, which can effectively improve the prediction accuracy. Attached Figure Description

[0043] Figure 1 The flowchart shows the distributed prediction method for multi-station charging load based on reservoir calculation provided by the present invention.

[0044] Figure 2 The diagram shows the structure of the distributed prediction model based on reservoir computation, which is provided by this invention.

[0045] Figure 3 This is an implementation diagram of the multi-site distributed charging load prediction model based on reservoir calculation provided by the present invention. Detailed Implementation

[0046] The present invention will be further described below with reference to embodiments. The description of the embodiments below is only for the purpose of helping to understand the present invention. It should be noted that those skilled in the art can make several modifications to the present invention without departing from the principle of the present invention, and these improvements and modifications also fall within the protection scope of the claims of the present invention.

[0047] Example 1:

[0048] like Figure 1 As shown, this application provides a distributed forecasting method for multi-station charging load based on reservoir computing, which is applied to applications such as... Figure 3 As shown, the charging load forecast for five charging stations distributed in different geographical locations was established, and a system was built as follows: Figure 2 The distributed prediction network shown, based on an approximately synchronous parallel mechanism, specifically includes:

[0049] Step 1: Obtain historical load data for five geographically dispersed electric vehicle charging stations and preprocess the collected data.

[0050] In step 1, the preprocessing includes: handling missing values, time alignment, and data normalization.

[0051] The interpolation and time alignment processing for missing data includes: observing historical data, selecting a unified timestamp, and performing secondary interpolation processing on the missing data; discarding data with misaligned timestamps.

[0052] The normalization process uses the following formula:

[0053]

[0054] It represents the true value of each data point in the sequence data. , These represent the maximum and minimum values ​​of the sequence data, respectively. This represents the normalized data for each data point in the sequence. The historical load data from the five stations were normalized sequentially.

[0055] Step 2: Construct a similarity matrix based on the Gaussian kernel function to evaluate the spatiotemporal coupling characteristics between electric vehicle charging stations located in different spatial locations. Based on these coupling characteristics, construct a distributed site set, and use the load data of these sites as input to the prediction model.

[0056] The historical load data for each electric vehicle charging station is represented as follows: ,in, This indicates the serial number of the charging station. , Indicates the sequence number of the data point. Indicates charging station A set of historical load data points.

[0057] Specifically, coupling analysis includes: Gaussian kernel function calculation, similarity calculation, and similarity matrix calculation.

[0058] Iterative computation between sites Data similarity of each sampling point for:

[0059]

[0060] in, The first in the dataset The first historical load time series data of the charging station Each sampled value, It is a parameter that controls the range of action. For the two sites Gaussian kernel function values ​​between load data points These are the weight coefficients that are updated iteratively.

[0061] In step 2, the initial value of the similarity matrix is ​​calculated based on the data similarity. :

[0062]

[0063] in, It is the identity matrix. Represents the number of sampling points. It is a matrix of all ones. These are the calibration parameters. The iterative formula for the similarity matrix is:

[0064]

[0065] in, This represents the first historical load time series data of site m and site n. The similarity of each sampled value is represented by an element in the similarity matrix. For the threshold, These are the iteration parameters. After multiple iterations, when the coupling between sites is strong, the similarity increases with the number of iterations. It will gradually converge to 1; when the coupling between sites is weak, the similarity will... It will gradually converge to 0. Based on the similarity matrix based on the Gaussian kernel function, the spatiotemporal coupling relationship between various stations can be analyzed relatively well.

[0066] Step 3: Establish the echo state network for the reservoir computation paradigm. It consists of a randomly generated input layer, a reservoir, and an output layer to be trained, with the reservoir exhibiting recurrent neural network characteristics. The specific network structure is as follows:

[0067]

[0068]

[0069]

[0070] in for The input vector at time t, for The state vector at time t, for The reservoir output vector at time t. It is a nonlinear function. for The output vector at time step 1 These are the input layer weight matrix, the reservoir weight matrix, and the output layer weight matrix, respectively.

[0071] Based on the reservoir computation paradigm, an echo state network is established to build a single power station charging load prediction model. The output layer of the mathematical model of this prediction network is represented as follows:

[0072]

[0073] In the formula, This is the output of the hidden layer of the neural network. This is the output layer weight matrix of the neural network. Represents network output.

[0074] To improve the robustness of the prediction network model, a regularization term is introduced, as shown below:

[0075]

[0076] In the formula, This represents the historical load data of the charging stations. These represent the hyperparameters of the model. In this embodiment, there are five charging stations, therefore... A distributed prediction network model was constructed:

[0077]

[0078]

[0079] In the formula, Represents the first in the charging station cluster Historical data on the load of each charging station Representing the first in the charging station cluster The hidden layer output of the neural network for each charging station. The first in the charging station cluster The output layer weight matrix of the neural network for each charging station. It is the target value of the output layer weight matrix. For system hyperparameters, This represents the number of charging stations in a distributed charging station cluster.

[0080] Step 4: The charging station cluster needs to integrate historical load data from different stations. To balance the requirements of big data training and data privacy protection, the augmented Lagrangian function of this prediction network is established as follows:

[0081]

[0082] in, , It is the parameter of the penalty term of the Lagrange function. Let be the Lagrange multiplier vector. It can be seen that if the target value of the output layer weight matrix is ​​estimated, then for the For each site, the prediction network can be trained and its parameters obtained using the following method:

[0083]

[0084] in This is the estimated value of the output layer weight matrix during the k-th iteration of training.

[0085] A learning algorithm using a distributed average consensus protocol based on an approximately synchronous parallel mechanism transforms complex problems into solving problems about variables. , and The three sub-problems are addressed by optimizing three variables respectively, enabling distributed collaborative training of multiple charging station cluster systems. After a sufficient number of iterations, the optimal weight matrix of the neural network is finally obtained. The training process of this neural network model is as follows:

[0086] 1. Randomly initialize the output layer weight matrix, target value, and Lagrange multiplier vector parameters. The output layer weight matrix is ​​calculated iteratively at the local site for the (k+1)th iteration. , Implement subsites Local weight updates in the prediction network:

[0087]

[0088] 2. During distributed collaborative training, it is necessary to calculate the mean of both methods to obtain global information. To achieve good distributed prediction results, each sub-site updates the global consensus variable through a distributed averaging consensus protocol:

[0089]

[0090] in, and Represent and The mean. Therefore, each station needs to iterate through communication as follows to obtain the result:

[0091]

[0092]

[0093] in, This is the topology connection matrix for the communication network between sub-sites. Represents the node communication topology graph. The degree of each charging station node, and This represents the maximum degree within the charging station cluster communication network topology. (When the number of communications...) Data for each node and It will converge to the mean of all data. and The charging station cluster communicates multiple times and follows a matrix pattern. The elements of the transmission and The value is then updated to obtain the mean.

[0094] To improve communication efficiency, the distributed average consensus protocol introduces an Approximate Synchronous Parallel (ASP) mechanism. When an update is less than a certain value, the update is delayed to reduce communication overhead. When an event is triggered, distributed global consensus computation is performed; if the conditions are not met, a lightweight heartbeat packet (a 1-bit flag) is sent to confirm whether node communication is normal. The triggering conditions are:

[0095]

[0096] in, It is a threshold parameter, which is dynamically updated with the number of iterations, allowing the communication traffic to decrease exponentially as the algorithm converges.

[0097]

[0098] 3. After achieving global consensus through multiple communications, update the global consensus variables on the local site. and local multipliers :

[0099]

[0100] The above training process is used to iteratively calculate the prediction model for the distributed electric vehicle charging station cluster multiple times to obtain the required weight matrix. During this distributed training process, and All updates are performed locally, without needing to access global information, thus maintaining data privacy. Only updates... When obtaining global weight mean information, only the output weight matrix containing partial information needs to be transmitted; it does not participate in the transmission of directly sensitive data. This is necessary for achieving global consensus. The distributed average consensus protocol is used to enable the site cluster to achieve automatic iterative calculation without a central node, and finally a network model for load prediction of distributed electric vehicle charging station cluster is obtained.

[0101] Neural networks were used to mine the nonlinear relationships between the loads of five selected electric vehicle charging stations. A continuous 30-day historical dataset was selected as the training set, and load predictions were made for the next 1, 3, and 7 days, respectively. The results are shown in Tables 1-3. Since previous step information was unavailable during testing, the output data from the first 100 steps was discarded, and the predicted data was padded to the next step. A grid search method was used to find the optimal hyperparameters on the training set. .

[0102] Table 1. Daily forecast results of distributed charging load at multiple charging stations

[0103]

[0104] Table 2. Three-day forecast results of distributed charging load at multiple charging stations

[0105]

[0106] Table 3. Seven-day forecast results of distributed charging load at multiple charging stations

[0107]

[0108] The accuracy metrics used in model training include normalized root mean square error (RMSE), mean square error (MSE), and mean absolute error (MAE), and are expressed as follows:

[0109]

[0110] in, Represents the number of samples. The first representing electric vehicle charging stations One load forecast value, The first representing electric vehicle charging stations The smaller the error index, the smaller the prediction error.

[0111] Step 5: Real-time prediction is performed using the trained model. The prediction results are shown in Tables 1-3, with the data showing the best overall performance highlighted in bold. The data source is the open-source ST-EVCDP V2 dataset for charging stations in a certain city, and RMSE is the primary error metric. The control groups are L-RC and C-RC, representing centralized iterative solution training and centralized analytical solution training of the reservoir calculation model, respectively. The reservoir calculation model for distributed multi-site electric vehicle charging load prediction in this invention is D-RC. Without obtaining all the data for local training, D-RC performs better than L-RC and is roughly equivalent to C-RC.

[0112] Example 2:

[0113] Based on Example 1, this application provides a distributed forecasting system for multi-station charging load based on reservoir computing, including:

[0114] The data acquisition module is used to obtain historical load data from each electric vehicle charging station and perform preprocessing.

[0115] A module is selected to construct a similarity matrix based on the Gaussian kernel function to test the correlation between the load data of each electric vehicle charging station. A distributed site set is constructed based on the coupling characteristics, and the load data of the sites in this set is used as the input of the prediction model.

[0116] The communication module is used for network communication during collaborative prediction in the distributed prediction system of electric vehicle charging station clusters.

[0117] The training module is used to calculate the weights of the output matrix in the training process of a distributed neural network model based on an approximate synchronous parallel mechanism.

[0118] The prediction module is used to make real-time predictions using the trained model.

[0119] Specifically, the system provided in this embodiment is the same as the system corresponding to the method provided in embodiment 1. Therefore, the parts in this embodiment that are the same as or similar to those in embodiment 1 can be referred to each other and will not be described again in this application.

[0120] The above embodiments are used to explain and illustrate the present invention, but not to limit the present invention. Any modifications and changes made to the present invention within the spirit and scope of the claims shall fall within the protection scope of the present invention.

Claims

1. A distributed forecasting method for charging load at multiple charging stations based on reservoir computing, characterized in that, The method includes: Step 1: Obtain historical charging load data from multiple electric vehicle charging stations and perform preprocessing. Step 2: Evaluate the spatiotemporal coupling characteristics between electric vehicle charging stations located in different spatial locations based on the similarity matrix of the Gaussian kernel function, and construct multiple distributed charging station clusters based on the coupling characteristics; Step 3: Based on the reservoir computing paradigm, an echo state network is established as a single charging load prediction model for a charging station. For the load prediction problem of multiple distributed charging station clusters, a distributed prediction network is constructed. This distributed prediction network, based on the reservoir computing paradigm, uses station load data from the distributed station set as input and matrix weights as output. A distributed learning algorithm based on an approximately synchronous parallel mechanism is used for distributed training of the prediction network. The distributed charging station cluster needs to integrate historical load data from different stations to establish the augmented Lagrangian function of the prediction network. : in, Represents the first in a distributed charging station cluster Historical data on the load of each charging station Represents the first in the distributed charging station cluster The hidden layer output of the neural network for each charging station. For the first in a distributed charging station cluster The output layer weight matrix of the neural network for each charging station. It is the target value of the output layer weight matrix. , For system hyperparameters, It is the parameter of the penalty term of the Lagrange function. The number of charging stations representing a distributed charging station cluster. Let T be a Lagrange multiplier vector, and let T denote the transpose. The prediction network parameters are obtained by training it based on the target values ​​of the output layer weight matrix using the following method: in This is an estimate of the output layer weight matrix during the k-th iteration of training; Step 4: For multiple distributed charging station clusters, build and train distributed prediction networks in turn according to Step 3, and use the trained prediction networks to make real-time predictions.

2. The distributed forecasting method for multi-station charging load based on reservoir computing according to claim 1, characterized in that, In step 1, the historical data of the charging load of the electric vehicle charging stations is the collection of load data for each charging station; the preprocessing includes: cleaning outliers and missing values ​​in the data, data normalization, and time alignment and interpolation processing of the historical data of multiple charging stations.

3. The distributed forecasting method for multi-station charging load based on reservoir computing according to claim 1, characterized in that, In step 2, based on the historical load time series data of charging stations from multiple sites, the Gaussian kernel function values ​​and similarities between load data points are calculated, and a similarity matrix is ​​constructed.

4. The distributed forecasting method for multi-site charging load based on reservoir computing according to claim 1, characterized in that, A distributed average consensus protocol based on an approximate synchronous parallel mechanism is established to achieve distributed collaborative training of multiple distributed charging station clusters. The specific process is as follows: randomly initialize the output layer weight matrix, target value, and Lagrange multiplier vector parameters. The output layer weight matrix is ​​calculated iteratively at the local site for the (k+1)th iteration. , To achieve local weight updates: in Let be the Lagrange multiplier vector for the k-th iteration. Represents the identity matrix.

5. The distributed forecasting method for multi-site charging load based on reservoir computing according to claim 4, characterized in that, In step 3, define and Represent and The mean of the two is needed to obtain global information during distributed collaborative training. Global consensus is achieved among the sub-sites through a distributed average consensus protocol. in, This represents the communication network topology connection coefficient between sub-site i and sub-site j, where n represents the number of communications. The maximum degree is determined based on the communication network topology diagram of the charging station cluster. Calculate the communication coefficient between different stations Conversely, when there is no communication between different stations, this value is 0; ,in Represents the node communication topology graph. The degree of each charging station node; then when the number of communications... Data for each node and It will converge to the mean of all data. and The charging station cluster communicates multiple times and follows a matrix pattern. The elements of the transmission and The value is then updated to obtain the mean.

6. The distributed forecasting method for multi-site charging load based on reservoir computing according to claim 4, characterized in that, In step 3, the distributed average consensus protocol introduces an Approximate Synchronous Parallel (ASP) mechanism. For each node, communication occurs when each update exceeds a set threshold, and distributed global consensus calculation is performed. Otherwise, it is necessary to confirm whether the node communication is normal. During the iteration process, the threshold is dynamically adjusted.

7. The distributed forecasting method for multi-site charging load based on reservoir calculation according to claim 5, characterized in that, In step 3, after obtaining global consensus through multiple communications, the global consensus variables are updated on the local site. and local multipliers Continue until the convergence condition is met; During distributed training, and All updates are performed locally, eliminating the need to access global information and maintaining data privacy. Only updates are performed locally. When it is necessary to obtain global weight mean information, only the output weight matrix with partial information needs to be transmitted, and it does not participate in the transmission of directly sensitive data.

8. A distributed forecasting system for multi-station charging load based on reservoir computing, implementing the distributed forecasting method according to any one of claims 1-7, characterized in that, The system includes: The data acquisition module is used to obtain historical load data from each electric vehicle charging station and perform preprocessing. The selected module evaluates the spatiotemporal coupling characteristics between electric vehicle charging stations located in different spatial locations based on the similarity matrix of the Gaussian kernel function, and constructs multiple distributed charging station clusters based on the coupling characteristics. The communication module is used for network communication during collaborative prediction in the distributed prediction system of electric vehicle charging station clusters. The training module is used to establish a distributed prediction network based on the reservoir computing paradigm. It takes the site load data in the distributed charging station cluster as input, the matrix weight as output, and adopts a distributed learning algorithm based on an approximate synchronous parallel mechanism to perform distributed training of the prediction network. The prediction module is used to make real-time predictions using the trained prediction network.

Images