RS-TCN-based V2G cluster output prediction method
Through the V2G cluster output prediction method based on RS-TCN, combined with the correlation analysis of Pearson's correlation coefficient and gray correlation degree, a one-dimensional full convolution network is used for charging and discharging feature extraction and simulation, which solves the problems of narrow coverage, poor generalization, weak real-time and insufficient feature fusion of V2G cluster charge and discharge load prediction in the prior art, achieving a more accurate and stable prediction effect.
Patent Information
- Application Number
- CN202510221940.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-27
- Publication Date
- 2025-05-30
AI Technical Summary
The existing V2G cluster charging and discharge load prediction methods have problems such as narrow coverage, poor generalization, weak real-time and insufficient feature fusion, making it difficult to effectively deal with the load fluctuations caused by electric vehicle network access.
Using the V2G cluster output prediction method based on RS-TCN, correlation analysis was performed through Pearson's correlation coefficient and gray correlation degree to extract features highly correlated with SOC. Then, based on the V2G cluster behavior data and charge and discharge behavior assumptions, the output simulation is performed by random sampling. Finally, a one-dimensional full convolution network is used to fuse causal convolution and expanding convolution, and a "hop connection" is introduced to form a residual block to build a TCN model suitable for charging and discharging prediction of V2G clusters.
It improves the accuracy and stability of V2G cluster output prediction, can better capture time characteristics and trends, and enhances the generalization ability and real-timeness of prediction.
Smart Images
Figure CN120073703A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of cluster output prediction, and relates to a method for predicting the output of a V2G cluster based on RS-TCN. Background Art
[0002] Due to the explosive growth of electric vehicles, the additional electric vehicle charging load during the peak period of the basic load of the regional power grid will surely impact the regional power grid, resulting in the situation of "peak on peak" in the regional power grid load. The accurate prediction of the charging and discharging load of the V2G cluster is an effective means to cope with the load fluctuations brought by the large-scale access of electric vehicles at present, and is of great significance for formulating reasonable electric vehicle charging and discharging strategies.
[0003] At present, the research on the prediction of the charging and discharging load of the V2G cluster is mainly divided into traditional methods and artificial intelligence methods.
[0004] Traditional methods for predicting the charging and discharging load of the V2G cluster mainly rely on mathematical models established based on the load change law, usually using the Monte Carlo method for random statistics, or analyzing in combination with the historical charging behavior data of electric vehicles. Traditional methods are restricted in load prediction in many aspects: ① Narrow coverage: The models proposed by traditional methods are difficult to cover all factors affecting the charging and discharging load of electric vehicles in reality, and are more dependent on human experience to model parameters; ② Poor generalization: The model verification of traditional methods requires a large amount of different types of data, so the generalization performance for predicting the charging and discharging load of electric vehicles in different scenarios is poor; ③ Weak real-time performance: Traditional methods for predicting the charging and discharging load of electric vehicles rely on preset models and simplified assumptions, and the continuous changes over time make it difficult for traditional models to adjust in time to cope with the complex laws behind the changes in the charging and discharging load; ④ Insufficient feature fusion: The characteristics of the diversity of information features, the huge amount of data, and the complexity of the data structure brought by the large number of electric vehicles accessing the network make it difficult for traditional methods to model.
[0005] Under the background of artificial intelligence deep learning, for time series prediction of V2G cluster output modeling, it can be achieved through recurrent neural network architectures such as long short-term memory network (LSTM) and gated recurrent unit (GRU), etc., but there are still problems and defects in some aspects: ① Complex structure: Although the influence of various factors on the load prediction accuracy is considered, the structure is complex, the training intensity is large, and it is difficult to evaluate the cumulative effect of errors; ② Incomplete capture of time features: Existing methods mainly rely on the long short-term memory network to process time series data, but it may encounter problems of gradient disappearance or gradient explosion when capturing long-term dependencies. In addition, existing methods do not fully consider time features such as periodicity and trend, which may lead to unstable prediction results; ③ Slow convergence: The gated recurrent unit is used to extract the important features of the time series characteristics to distinguish the importance of different time series, but the convergence speed is slow, affecting the prediction accuracy. Summary of the Invention
[0006] The object of the present invention is to overcome the defects of the prior art and provide a method for predicting the output of a V2G cluster based on RS-TCN. First, the Pearson correlation coefficient (PCC) and the grey relational grade (GRG) are used for correlation analysis to extract features highly correlated with the state of charge (SOC) from large-scale field data. Then, a large number of electric vehicles connected to the grid for charging and discharging are regarded as a V2G cluster. Based on the V2G cluster behavior data and the charging and discharging behavior assumptions, a certain number of V2G clusters are randomly sampled for V2G cluster output simulation. Finally, a one-dimensional fully convolutional network is adopted, which integrates causal convolution and dilated convolution, and introduces "skip connections" to form residual blocks to build a residual network. In a residual block, the TCN has two layers of dilated causal convolution and non-linearity, and the linear rectification unit ReLu activation function is improved by using the LeakyReLU function, and a TCN model suitable for predicting the charging and discharging of the V2G cluster is built.
[0007] To achieve the above object, the technical solution of the present invention is: a method for predicting the output of a V2G cluster based on RS-TCN, which constructs a charging and discharging prediction model for the output of a V2G cluster based on RS-TCN to achieve the prediction of the output of the V2G cluster.
[0008] In an embodiment of the present invention, the method specifically includes:
[0009] Using the Pearson correlation coefficient PCC and the grey relational grade GRG for correlation analysis to extract features highly correlated with the SOC from large-scale field data;
[0010] Regarding a large number of electric vehicles connected to the grid for charging and discharging as a V2G cluster, and based on the V2G cluster behavior data and the charging and discharging behavior assumptions, randomly sampling a predetermined number of V2G clusters for V2G cluster output simulation;
[0011] The temporal convolutional network TCN adopts a one-dimensional fully convolutional network, which integrates causal convolution and dilated convolution, and introduces "skip connections" to form residual blocks to build a residual network, and builds a TCN prediction model for the charging and discharging output of the V2G cluster.
[0012] In an embodiment of the present invention, using the Pearson correlation coefficient PCC and the grey relational grade GRG for correlation analysis to extract features highly correlated with the SOC from large-scale field data is specifically implemented as follows:
[0013] Step 1.1, data cleaning: eliminating invalid data, including vehicle data with a charging amount greater than 90 KWh, vehicle data with a residence time less than 15 minutes, and vehicle data with a residence time greater than 24 hours;
[0014] Step 1.2, Data Conversion: The prediction of the output of the V2G cluster is carried out in time steps. Time step 1 is 0:00, time step 2 is 0:15, and so on. The start charging time, end charging time, and vehicle stay time in the original data after data cleaning are converted into time steps;
[0015] Step 1.3, Conduct a correlation analysis to select the feature data that is highly correlated with the SOC during the charging process.
[0016] In an embodiment of the present invention, step 1.3 is specifically implemented as follows:
[0017] Step 1.3.1, Based on the Pearson correlation coefficient PCC, provide the direction and strength of linear correlation; the specific formula is as follows:
[0018]
[0019] In the formula, x i is the i-th value of the input feature x sequence, y i is the i-th value of the target sequence, y is the SOC of the battery pack, is the average value of the input feature sequence x, is the average value of the SOC sequence;
[0020] Step 1.3.2, Based on the grey relational grade GRG, provide a quantitative measure of the system's dynamic evolution; the specific formula is as follows:
[0021]
[0022] In the formula, ξ i (k) is the grey correlation coefficient of the i-th feature at the k-th point; the calculation formula of ξ i (k) is as follows,
[0023]
[0024] In the formula, ρ is the resolution coefficient;
[0025] Step 1.3.3, Set PCC greater than 0.5 and GRG greater than 0.8 as the features highly correlated with the SOC. At the same time, set the PCC and GRG between two features greater than 0.9 as highly autocorrelated. If two features are highly autocorrelated, one of them must be discarded to reduce feature redundancy.
[0026] In an embodiment of the present invention, a large number of electric vehicles connected to the grid for charging and discharging are regarded as a V2G cluster. Based on the V2G cluster behavior data and the charging and discharging behavior assumptions, a predetermined number of V2G clusters are randomly sampled for V2G cluster output simulation. The specific implementation is as follows:
[0027] Step 2.1. V2G cluster simulates discharge output;
[0028] Step 2.1.1. Determine whether the initial SOC of the vehicle is greater than the preset discharge threshold SOC f ;
[0029] Step 2.1.2. Let t i,in be the vehicle access time, t i,out be the vehicle end charging time, and loop from t i,in to t i,out to end; record the discharge power P d_i,t of each electric vehicle at each time step, and calculate the SOC i，t at each time step after discharge, and the lowest threshold for discharge is SOC min ; if SOC i，t < SOC min , then the vehicle stops discharging to the grid, and record the state of charge SOC_f i after each vehicle ends discharging;
[0030] Step 2.1.3. Repeat steps 2.1.1 to 2.1.3 from the electric vehicle cluster V2G 1 to V2G m cluster, and superimpose the discharge power of each electric vehicle at each time step to obtain the V2G cluster simulated discharge output, where m represents the V2G cluster with a sampling quantity of m;
[0031] Step 2.2. V2G cluster simulates charging output;
[0032] Step 2.2.1. Assume that each electric vehicle is fully charged when unplugging the charging pile, that is, SOC out = 1;
[0033] Step 2.2.2. Loop from t i,out to t i,in to end, calculate the state of charge SOC i,t-1 at the previous time step, and record the charging power P c_i,t at this time step, SOC_f i is the state of charge after ending discharge. If SOC i,t-1 < SOC_f i , then the vehicle ends charging;
[0034] Step 2.2.3. Repeat steps 2.2.1 to 2.2.3 from the electric vehicle cluster V2G 1 to V2G m cluster, and superimpose the charging power of each electric vehicle at each time step to obtain the V2G cluster simulated charging output.
[0035] In one embodiment of the present invention, the Temporal Convolutional Network (TCN) adopts a one-dimensional fully convolutional network, which integrates causal convolution and dilated convolution, and introduces "skip connections" to form residual blocks to build a residual network, and a TCN model suitable for V2G cluster charging and discharging prediction is built. The specific implementation is as follows:
[0036] Step 3.1: The TCN adopts a one-dimensional fully convolutional network;
[0037] Step 3.2: Add causal dilated convolution, add an interval dilation factor d to the convolutional kernel. Given a filter F=(f 1 , f 2 , …, f K ), and a sequence X=(x 1 , x 2 , …, x t ), the dilated convolution at x t is defined as:
[0038]
[0039] In the formula, K is the filter length, k is the convolutional kernel size, k = 1, 2, …, K, d is the dilation factor, and * d is the dilated convolution operation, indicating that the dilation factor d is inserted during the convolution process;
[0040] Step 3.3: Introduce "skip connections" to form residual blocks;
[0041] Step 3.4: Build a TCN prediction model for the discharging output of the V2G cluster, which is composed of 2 connected residual blocks;
[0042] Step 3.5: Build a TCN prediction model for the charging output of the V2G cluster, which is composed of 3 connected residual blocks.
[0043] In one embodiment of the present invention, in the one-dimensional fully convolutional network, the input and output lengths of the hidden layer are both n time steps, the convolutional kernel size k = 3, and the number of zero-padding is k - 1.
[0044] In one embodiment of the present invention, within a residual block, the TCN has two layers of dilated causal convolution and non-linearity. The linear correction unit ReLu activation function is improved by using the LeakyReLU function, and the expression is as follows:
[0045]
[0046] In the formula, α is a constant;
[0047] To build a residual network, the output O of the residual block is:
[0048] O = Activition(x + f(x))
[0049] Where x is the input, f(x) is the residual, and Activation() is the activation function. Different activation functions can be selected according to the situation.
[0050] The present invention also provides an electronic device, including a memory, a processor, and computer program instructions stored on the memory and capable of being run by the processor. When the processor runs the computer program instructions, the method steps as described above can be implemented.
[0051] The present invention also provides a computer-readable storage medium, on which computer program instructions capable of being run by the processor are stored. When the processor runs the computer program instructions, the method steps as described above can be implemented.
[0052] Compared with the prior art, the present invention has the following beneficial effects:
[0053] ① The present invention introduces the diversity of samples through random sampling to ensure that the samples used to simulate the output of the V2G cluster cover electric vehicles with different types, different parameters, and different charging and discharging behaviors, and can more conform to the randomness of the V2G cluster output under actual conditions.
[0054] ② The prediction basic data used in the present invention has been subjected to cleaning processing, correlation analysis feature extraction, and the influencing factors of the charging load have been summarized.
[0055] ③ The present invention uses the Pearson correlation coefficient (PCC) and the grey relational grade (GRG) to characterize the correlation between the input features and the target output of the data-driven model, and improves the linear correlation and the system evolution intensity.
[0056] ④ The present invention analyzes the characteristics of the time series V2G cluster output load, more accurately predicts the V2G cluster output through the time series, and uses historical data to reveal future trends.
[0057] ⑤ TCN uses causal dilated convolution, adding intervals in the convolutional kernel, which can increase the receptive field without increasing the number of convolutional kernels.
[0058] ⑥ The present invention uses a temporal convolutional network, which can overcome problems such as gradient vanishing and gradient explosion, and supports parallel computing. Compared with recurrent neural networks, the temporal convolutional network (TCN) has irreplaceable advantages in predicting time series data. Description of the Drawings
[0059] Figure 1 It is a flowchart of the method of the present invention.
[0060] Figure 2 It is a schematic diagram of the one-dimensional fully convolutional network structure.
[0061] Figure 3Schematic diagram of a one-dimensional fully convolutional network structure with causal dilated convolution added.
[0062] Figure 4 Schematic diagram of the residual block structure.
[0063] Figure 5 V2G cluster discharge output TCN prediction model.
[0064] Figure 6 V2G cluster charging output TCN prediction model. Specific implementation manner
[0065] The technical solution of the present invention will be specifically described below with reference to the accompanying drawings.
[0066] It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present application belongs.
[0067] It should be noted that the terms used herein are only for describing specific implementation manners and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular forms are also intended to include the plural forms. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.
[0068] The present invention provides a method for predicting the output of a V2G cluster based on RS-TCN, constructs a prediction model for the charging and discharging output of a V2G cluster based on RS-TCN, and realizes the prediction of the output of a V2G cluster. Specifically, it includes:
[0069] Using the Pearson correlation coefficient PCC and the grey relational grade GRG for correlation analysis, and extracting features highly correlated with SOC from large-scale field data;
[0070] Regarding a large number of electric vehicles connected to the grid for charging and discharging as a V2G cluster, based on the behavior data of the V2G cluster and the charging and discharging behavior assumptions, a predetermined number of V2G clusters are randomly sampled for V2G cluster output simulation;
[0071] The temporal convolutional network TCN adopts a one-dimensional fully convolutional network, fuses causal convolution and dilated convolution, and introduces "skip connections" to form residual blocks to build a residual network, and constructs a TCN prediction model for the charging and discharging output of a V2G cluster.
[0072] The following is the specific implementation process of the present invention.
[0073] As Figure 1As shown in the figure, this embodiment provides a method for predicting the output of a V2G cluster based on RS-TCN, including the following steps:
[0074] Step 1: Obtain the real charging data set of electric vehicles, which mainly includes the charging start date, grid connection time, grid disconnection time, charging end date, the amount of electricity charged by the electric vehicle this time, and the duration of the electric vehicle charging, and preprocess the data to extract features based on correlation.
[0075] Step 1.1: Data cleaning, eliminating invalid data: vehicles with a charged electricity greater than 90 KWh; vehicle data with a stay time less than 15 minutes; vehicle data with a stay time greater than 24 hours.
[0076] Step 1.2: Data conversion: The prediction of the V2G cluster output is carried out in a time step manner. Time step 1 is 0:00, time step 2 is 0:15, and so on. Therefore, it is necessary to convert the vehicle start charging time, end charging time, and vehicle stay time in the original data into time steps.
[0077] Step 1.3: Conduct a correlation analysis on the data to select feature data that is highly correlated with the SOC during the charging process.
[0078] Step 1.3.1: Based on the Pearson correlation coefficient (PCC), provide the direction and intensity of linear correlation. The specific formula is as follows:
[0079]
[0080] Where x i is the i-th value of the input feature x sequence, y i is the i-th value of the target sequence, y is the SOC of the battery pack, is the average value of the input feature sequence x, is the average value of the SOC sequence.
[0081] Step 1.3.2: Based on the grey relational grade (GRG), provide a quantitative measure of the system's dynamic evolution. The specific formula is as follows:
[0082]
[0083] Where is the grey correlation coefficient of the i-th feature at the k-th point. The calculation formula of ξ i (k) is as follows,
[0084]
[0085] Where ρ is the resolution coefficient, which is taken as 0.5 in this example.
[0086] Step 1.3.3: Set PCC greater than 0.5 and GRG greater than 0.8 as features highly correlated with SOC. At the same time, set the PCC and GRG between the two features to be greater than 0.9 as highly autocorrelated. If the two features are highly autocorrelated, one of the features must be discarded to reduce feature redundancy.
[0087] Step 2: Regard large-scale electric vehicles for grid-connected charging and discharging as a V2G cluster. Based on the V2G cluster behavior data and charging and discharging behavior assumptions, conduct random sampling to simulate the V2G cluster output. Randomly select a number of m electric vehicle clusters from the dataset.
[0088] Step 2.1: Simulate the discharging output of the V2G cluster
[0089] Step 2.1.1: Determine whether the initial SOC of the vehicle is greater than the preset discharging threshold SOC. In this example, set SOC f , this example sets SOC f = 0.5.
[0090] Step 2.1.2: Let t i,in be the vehicle grid connection time, and t i,out be the vehicle end charging time. The loop starts from t i,in and ends at t i,out . Record the discharging power P d_i,t of each electric vehicle at each time step, and calculate the SOC i，t at each time step after discharging. Set the lowest threshold SOC min of discharging to 0.2. If SOC i，t < SOC min , then the vehicle stops discharging to the grid, and record the state of charge SOC_f i after each vehicle ends discharging.
[0091] Step 2.1.3: Repeat steps 2.1.1 - 2.1.3 from the electric vehicle cluster V2G 1 to V2G m cluster. Superimpose the discharging power of each electric vehicle at each time step to obtain the simulated discharging output of the V2G cluster.
[0092] Step 2.2: Simulate the charging output of the V2G cluster
[0093] Step 2.2.1: Assume that each electric vehicle is fully charged when unplugging the charging pile, that is, SOC out = 1.
[0094] Step 2.2.2: Let t i,in be the vehicle grid connection time, and t i,out be the vehicle end charging time. The loop starts from t i,out and ends at t i,inEnd, calculate the state of charge SOC of the previous time step i,t-1 , and record the charging power P of this time step c_i,t , SOC_f i is the state of charge after the end of discharge. If SOC i,t-1 <SOC_f i , then the vehicle ends charging.
[0095] Step 2.2.3, from the electric vehicle cluster V2G 1 to V2G m cluster, repeat steps 2.2.1 - 2.2.3, and superimpose the charging power of each electric vehicle at each time step to obtain the simulated charging output of the V2G cluster.
[0096] Step 3, adopt a one - dimensional fully convolutional network, fuse causal convolution and dilated convolution, and introduce "skip connection" to form a residual block to build a residual network, and build a TCN model suitable for V2G cluster charge - discharge prediction.
[0097] The specific steps are as follows:
[0098] Step 3.1, as Figure 2 shown, TCN adopts a one - dimensional fully convolutional network (1D FCN). The input and output lengths of the hidden layer are both n time steps, the convolutional kernel size k = 3, and the number of zero - paddings is k - 1.
[0099] Step 3.2, as Figure 3 shown, add causal dilated convolution, add an interval dilation factor d to the convolutional kernel. Given a filter F=(f 1 , f 2 ,…, f K ), the dilated convolution of the sequence X=(x 1 , x 2 ,…, x t ) at x t can be defined as:
[0100]
[0101] In the formula, k is the convolutional kernel size, and d is the dilation factor.
[0102] Step 3.3, as Figure 4 shown, introduce "skip connection" to form a residual block.
[0103] Within a residual block, TCN has two layers of dilated causal convolution and non - linearity. The linear correction unit ReLu activation function is improved to adopt the LeakyReLU function, and the expression is as follows:
[0104]
[0105] Where α is a constant, and in this example, α is taken as 0.01.
[0106] Build a residual network. The output O of the residual block is:
[0107] O = Activition(x + f(x))
[0108] Where x is the input and f(x) is the residual.
[0109] Step 3.4, as Figure 5 shown, build a TCN prediction model for the discharging output of the V2G cluster, which is formed by connecting 2 residual blocks.
[0110] Step 3.5, as Figure 6 shown, build a TCN prediction model for the charging output of the V2G cluster, which is formed by connecting 3 residual blocks.
[0111] The present invention also provides an electronic device, including a memory, a processor, and computer program instructions stored on the memory and capable of being run by the processor. When the processor runs the computer program instructions, the method steps as described above can be implemented.
[0112] The present invention also provides a computer-readable storage medium, on which computer program instructions capable of being run by the processor are stored. When the processor runs the computer program instructions, the method steps as described above can be implemented.
[0113] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memories, CD-ROMs, optical memories, etc.) containing computer-usable program codes.
[0114] The present application is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.
[0115] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to work in a particular manner, such that the instructions stored in the computer-readable memory produce a manufacture including an instruction device that implements the functions specified in one or more processes and / or blocks Figure 1 of the process or processes and / or boxes Figure 1 or boxes.
[0116] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus, such that a series of operational steps are performed on the computer or other programmable apparatus to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable apparatus provide steps for implementing the functions specified in one or more processes and / or blocks Figure 1 of the process or processes and / or boxes Figure 1 or boxes.
[0117] As described above, it is only the preferred embodiment of the present invention, and it is not intended to limit the present invention in any other form. Any person skilled in the art may use the technical content disclosed above to make changes or modifications into equivalent embodiments with equivalent changes. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution content of the present invention still fall within the protection scope of the technical solution of the present invention.
Claims
1. A V2G cluster output prediction method based on RS-TCN, characterized in that: A V2G cluster output charging and discharging prediction model based on RS-TCN is constructed to realize V2G cluster output prediction.
2. The RS-TCN-based V2G cluster output prediction method according to claim 1, characterized in that: The method specifically comprises: Correlation analysis was performed using the Pearson correlation coefficient PCC and the grey relational grade GRG to extract features highly correlated with SOC from large-scale field data; Considering large-scale electric vehicles connected to the grid for charging and discharging as V2G clusters, a predetermined number of V2G clusters are randomly sampled to simulate V2G cluster output based on V2G cluster behavior data and charging and discharging behavior assumptions. The temporal convolutional network (TCN) adopts a one-dimensional full convolutional network, integrates causal convolution and dilated convolution, and introduces "skip connection" to form a residual block to build a residual network, and build a TCN prediction model for V2G cluster charging and discharging output.
3. The RS-TCN-based V2G cluster output prediction method according to claim 2, characterized in that: Correlation analysis was performed using the Pearson correlation coefficient PCC and the grey relational grade GRG to extract features highly correlated with SOC from large-scale field data. The specific implementation is as follows: Step 1.1, data cleaning: Eliminate invalid data, including data of vehicles with a charging capacity greater than 90KWh, vehicle data with a stay time of less than 15 minutes, and vehicle data with a stay time of more than 24 hours; Step 1.2, data conversion: V2G cluster output prediction is performed in time steps, with time step 1 being 0:00, time step 2 being 0:15, and so on. The vehicle charging start time, charging end time, and vehicle stay time in the original data after data cleaning are converted into time steps; Step 1.3: Perform correlation analysis to select feature data that is highly correlated with SOC during the charging process.
4. The RS-TCN-based V2G cluster output prediction method according to claim 3 is characterized in that: Step 1.3 is implemented as follows: Step 1.3.1, based on the Pearson correlation coefficient PCC, provide the direction and strength of linear correlation; the specific formula is as follows: In the formula, x i is the i-th value of the input feature x sequence, y i is the i-th value of the target sequence, y is the SOC of the battery pack, is the average value of the input feature sequence x, is the average value of the SOC series; Step 1.3.2: Based on the grey relational degree GRG, provide a quantitative measure of the dynamic evolution of the system; the specific formula is as follows: In the formula, ξ i (k) is the grey correlation coefficient of the i-th feature at the k-th point; ξ i The calculation formula of (k) is as follows, Where ρ is the resolution coefficient; Step 1.3.3: Set PCC greater than 0.5 and GRG greater than 0.8 as features that are highly correlated with SOC. At the same time, set PCC and GRG between the two features greater than 0.9 as highly autocorrelated. If there is a high autocorrelation between the two features, one of the features must be discarded to reduce feature redundancy.
5. The RS-TCN-based V2G cluster output prediction method according to claim 2, characterized in that: Considering large-scale electric vehicles connected to the grid for charging and discharging as V2G clusters, based on V2G cluster behavior data and charging and discharging behavior assumptions, a predetermined number of V2G clusters are randomly sampled to simulate V2G cluster output. The specific implementation is as follows: Step 2.1, V2G cluster simulates discharge output; Step 2.1.1: Determine whether the vehicle's initial SOC is greater than the preset discharge threshold SOC f ; Step 2.1.2, let t i,in is the vehicle access time, t i,out The charging time for the vehicle ends, and the cycle starts from t i,in Start to t i,out End; record the discharge power P of each electric vehicle at each time step d_i,t , and calculate the SOC at each time step after discharge i，t , the minimum discharge threshold is SOC min If SOC i，t <SOC min , the vehicle stops discharging to the grid and records the state of charge SOC_f of each vehicle after the discharge ends i ; Step 2.1.3: From electric vehicle cluster V2G1 to V2G m The cluster repeats steps 2.1.1 to 2.1.3, superimposing the discharge power of each electric vehicle at each time step to obtain the simulated discharge output of the V2G cluster, where m represents the number of V2G clusters extracted is m; Step 2.2, V2G cluster simulates charging output; Step 2.2.1: Assume that each electric vehicle is fully charged when it is unplugged from the charging pile, that is, SOC out =1; Step 2.2.2, loop from t i,out Start to t i,in End, calculate the state of charge SOC of the previous time step i,t-1 , and record the charging power P at this time step c_i,t , SOC_f i is the state of charge after the discharge is completed. If SOC i,t-1 <SOC_f i , the vehicle ends charging; Step 2.2.3: From electric vehicle cluster V2G1 to V2G m The cluster repeats steps 2.2.1 to 2.2.3, superimposing the charging power of each electric vehicle at each time step to obtain the simulated charging output of the V2G cluster.
6. The RS-TCN-based V2G cluster output prediction method according to claim 2, characterized in that: The temporal convolutional network (TCN) uses a one-dimensional full convolutional network, integrates causal convolution and dilated convolution, and introduces "skip connection" to form a residual block to build a residual network, and build a TCN model suitable for V2G cluster charging and discharging prediction. The specific implementation is as follows: Step 3.1, TCN uses a one-dimensional fully convolutional network; Step 3.2: Add causal dilation convolution, add the interval dilation factor d to the convolution kernel, and give the filter F = (f1, f2, ..., f K ), sequence X = (x1, x2, …, x t ) in x t The dilated convolution at is defined as: Where K is the filter length, k is the convolution kernel size, k = 1, 2, ... K, d is the dilation factor, * d It is a dilated convolution operation, which means that a dilation factor d is inserted in the convolution process; Step 3.3, introduce "skip connection" to form a residual block; Step 3.4: Build a TCN prediction model for V2G cluster discharge output, which is formed by connecting two residual blocks; Step 3.5: Build a TCN prediction model for V2G cluster charging output, which is composed of three residual blocks.
7. The RS-TCN-based V2G cluster output prediction method according to claim 2 or 6, characterized in that: In a one-dimensional fully convolutional network, the input and output lengths of the hidden layer are consistent with n time steps, the convolution kernel size is k=3, and the number of zero padding is k-1.
8. The RS-TCN-based V2G cluster output prediction method according to claim 2 or 6, characterized in that: In a residual block, TCN has two layers of dilated causal convolution and nonlinearity. The linear rectification unit ReLu activation function is improved to use the LeakyReLU function, which is expressed as follows: In the formula, α is a constant; Construct a residual network, the output O of the residual block is: O=Activition(x+f(x)) Where x is the input, f(x) is the residual, and Activation() is the activation function. Different activation functions can be selected according to the situation.
9. An electronic device, characterized in that: The method comprises a memory, a processor and computer program instructions stored in the memory and capable of being executed by the processor. When the processor executes the computer program instructions, the method steps as claimed in any one of claims 1 to 8 can be implemented.
10. A computer-readable storage medium having stored thereon computer program instructions that can be executed by a processor, and when the processor executes the computer program instructions, the method steps according to any one of claims 1 to 8 can be implemented.