Data-driven prediction method for electric vehicle charging capacity and charging time

Abstract

The present invention discloses a data-driven method for predicting the charging capacity and charging duration of an electric vehicle, comprising the following steps: Step 1, preprocessing the historical charging data of the electric vehicle; Step 2, extracting features from the preprocessed charging segments and assigning labels; Step 3, implementing label prediction based on an LSTM classification model; Step 4, implementing charging capacity and charging duration prediction based on an RF regression model. The present invention efficiently and accurately predicts the charging capacity and charging duration of an electric vehicle, which can not only optimize the operating efficiency of charging stations and improve the user's charging experience, but also make greater contributions to smart grids and sustainable transportation systems. Through accurate predictions, charging stations can better arrange the use of charging facilities, reduce waiting time in queues, and improve the utilization rate of charging piles; users can reasonably plan charging time based on the prediction information, avoid long waiting times, and improve the flexibility and convenience of travel plans.

Description

Technical Field

[0001] The invention belongs to the technical field of electric vehicle charging and discloses a data-driven method for predicting the charging amount and charging time of an electric vehicle. Background Art

[0002] As the proportion of renewable energy generation increases annually, large-scale flexible resources are needed to maintain the smooth operation of the power system. In this context, vehicle-grid interaction, as a new energy regulation method, not only provides flexible resource support to ensure the stable operation of the power system, but also promotes the large-scale application of renewable energy and the development of intelligent energy management. Within the framework of vehicle-grid interaction, efficient and accurate prediction of electric vehicle charging capacity and charging time is particularly important. Accurate prediction of charging capacity and charging time not only helps optimize charging station operational efficiency, reduces charging queues, and improves user experience, but is also crucial for grid dispatch. This predictive information enables more efficient allocation of power resources, rationally adjusts load during peak demand periods, and further promotes the efficient use of renewable energy.

[0003] Therefore, there is an urgent need for a reasonable and feasible solution to accurately predict the charging amount and charging time of electric vehicle users, so as to ensure the stability of the power system and optimize the driver's charging experience. Summary of the Invention

[0004] This paper analyzes historical electric vehicle operating data, extracts data features representative of charging behavior, and establishes a GMM-LSTM-RF prediction model for electric vehicles. Using this model, we can accurately predict the charge capacity and charging duration of electric vehicles in the future.

[0005] The present invention discloses a data-driven electric vehicle charging capacity and charging time prediction method, which comprises the following steps:

[0006] Perform data preprocessing on historical charging data of electric vehicles;

[0007] Extract features from pre-processed historical charging data and assign labels to generate charging behavior pattern features and add them to electric vehicle charging segment data;

[0008] Predict charging behavior patterns based on electric vehicle charging segment data and LSTM classification model;

[0009] The charging amount and charging time are predicted based on the predicted charging behavior pattern, electric vehicle charging segment data and RF regression model.

[0010] Furthermore, the data preprocessing of the historical charging data of the electric vehicle includes the following steps:

[0011] Clean the historical charging data of electric vehicles to handle data missing, abnormal, and duplicated data;

[0012] Standardize the data;

[0013] The charging cycle is divided. When the battery current is negative and the vehicle speed is zero, it is considered that the electric vehicle is in the charging state.

[0014] Furthermore, the feature extraction and labeling of the pre-processed historical charging data includes the following steps:

[0015] Each charging segment is extracted from the pre-processed historical charging data as:

[0016] S i ={F1,F2,F3,F4,F5,F6,Num n}

[0017] Among them, F1-F6 correspond to the proposed features charging start time, charging start date, charging start SOC, charging method, charging duration and charging power respectively. n Indicates the corresponding electric vehicle number;

[0018] According to the electric vehicle number, the set of all charging segments for each vehicle is:

[0019] V n ={S1,…,S i ,…} T

[0020] Among them, S i Indicates the charging segment characteristics;

[0021] Based on the set of all charging segments of all electric vehicles, the electric vehicle charging segment feature database is constructed as follows:

[0022] Data={V1,…,V n ,…} T

[0023] V n represents the charging segment feature set;

[0024] The optimal number of clusters is determined according to the BIC criterion, which is expressed as:

[0025]

[0026] Among them, BIC represents the score calculated by the BIC criterion, p(x n |Θ) is the observed data sample x under the given model parameter Θ nThe probability density function of , N is the number of samples, n represents the sample number, and K is the number of clusters;

[0027] GMM clustering is performed by extracting features and the optimal number of clusters; the Num n Column deletion to obtain the feature matrix:

[0028]

[0029] Where N is the number of samples and f is the number of features;

[0030] The Gaussian distribution model is determined by solving the optimal Gaussian distribution model parameters for each sample using the expectation maximization method. The expectation maximization method is:

[0031] D={μ i ,∑ i ,α i}

[0032] Among them, μ i ,Σ i , α i Represent the mean vector, covariance matrix, and mixing coefficient of the Gaussian distribution respectively; each sample determines its most likely clustering according to the responsibility weight of each Gaussian distribution to which it belongs;

[0033] The clustering results are added to the electric vehicle charging segment data as charging behavior pattern features;

[0034] All charging segments of all vehicles are labeled with four types of labels representing four charging behavior modes: short-term low-power charging mode, balanced charging mode, long-term high-power charging mode, and high-efficiency fast charging mode;

[0035] Then according to Num n Divide all segments of all vehicles into a set of charging segments with each electric vehicle labeled by its charging behavior:

[0036] V n ′={S1′,…,S i ′,…} T

[0037] V n ′ represents the set of charging segments without vehicle numbers, S i ' represents the charging segment without the vehicle number.

[0038] Furthermore, the method of predicting charging behavior patterns based on electric vehicle charging segment data and an LSTM classification model includes the following steps:

[0039] Extract each charging segment S i′ in F1, F2, F3, F4, F7 to obtain a new charging segment x t ={F1,F2,F3,F4,F7}, where F1-F6 correspond to the proposed features of charging start time, charging start date, charging start SOC, charging method, charging duration and charging power respectively, and F7 is the charging behavior mode. Then all new charging segments x t Combination to get X={F1,…,x t ,…} T ;

[0040] For each new charging segment, the charging behavior pattern is obtained by calculating it as the input of the LSTM unit. The calculation process of the LSTM unit includes the forget gate, input gate, memory unit state update and output gate. The calculation of each unit can be described as follows:

[0041]

[0042] Among them, W f 、W i 、W C 、W o They are the weight matrices of the forget gate, input gate, candidate memory unit state, and output gate, respectively, b f 、b i 、b C 、b o They are the forget gate, input gate, candidate memory unit state, and output gate bias, respectively. t-1 represents the hidden state of the previous time step, σ represents the sigmoid activation function, C t-1 Represents the memory cell state of the previous time step;

[0043] Before training begins, all weight matrices and bias items of the LSTM unit are randomly initialized. During the training process, the weight matrices and bias items are updated using the backpropagation through time algorithm; finally, the hidden state h of the LSTM unit is obtained. t and memory cell state C t Transfer to the next time step, the hidden state h t and memory cell state C t Each is a vector of length l, where l is the dimension of the LSTM hidden layer;

[0044] After processing all time steps in the time series, the hidden state h of the last time step is used T As the final feature vector for classification:

[0045] y=softmax(W y ·h T +b y )

[0046] Among them, y is the probability distribution vector of the classification, W y is the weight matrix of the output layer, b y is the bias term, h T Represents the hidden state of the last time step; the weight matrix and bias terms are randomly initialized before training begins, and then the backpropagation algorithm is used to update the weight matrix and bias terms through time;

[0047] Finally, the charging behavior pattern F7′ corresponding to the y with the highest probability is taken as the prediction result of the label.

[0048] Furthermore, the prediction of charging capacity and charging duration is achieved based on the predicted charging behavior pattern, EV charging segment data, and RF regression model, including the following steps:

[0049] Extract each charging segment S i The charging start time F1, charging start date F2, charging start SOCF3, charging method F4, charging duration F5, charging power F6 in ′ are combined with F7′ predicted by the LSTM classification model to obtain a new charging segment x t ′={F1,F2,F3,F4,F7′,F5,F6}, and then all new charging segments x t ′ combination to obtain X′={x1′,…,x t ′,…} T ;

[0050] Randomly extract multiple subsamples from X′, and each subsample is used to train a decision tree;

[0051] Each decision tree predicts the input sample. For a sample x t ′, the prediction process of random forest can be expressed as:

[0052]

[0053] in, is the final predicted value, M is the number of decision trees, is the mth decision tree for sample x t ′’s prediction results;

[0054] Finally, the charging capacity and charging time are obtained through the final predicted values.

[0055] Furthermore, for the extracted historical charging data, if there is a gap of more than half an hour between two consecutive charging data frames, they are considered to be two independent charging cycles.

[0056] The present invention also provides a data-driven electric vehicle charging capacity and charging time prediction device, which includes a data preprocessing module, a feature extraction module, an LSTM classification module and an RF regression module;

[0057] The data preprocessing module is used to preprocess the historical charging data of the electric vehicle;

[0058] The feature extraction module is used to extract features from the pre-processed historical charging data and assign labels to generate data with charging behavior pattern features added to the electric vehicle charging segment;

[0059] The LSTM classification module is used to predict charging behavior patterns based on the data of electric vehicle charging segments and the LSTM classification model;

[0060] The RF regression module is used to predict the charging amount and charging time based on the predicted charging behavior pattern and electric vehicle charging segment data through the RF regression model.

[0061] An electronic device is also provided, comprising a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set, or an instruction set, and the at least one instruction, at least one program, the code set, or the instruction set is loaded and executed by the processor to implement a data-driven method for predicting the charge amount and charging time of an electric vehicle.

[0062] A computer storage medium is also provided, in which at least one instruction, at least one program, code set or instruction set is stored. The at least one instruction, at least one program, code set or instruction set is loaded and executed by the processor to implement a data-driven electric vehicle charging amount and charging time prediction method.

[0063] The beneficial effects achieved by the present invention are:

[0064] This invention efficiently and accurately predicts the charge capacity and duration of electric vehicles, not only optimizing charging station operations and enhancing the user charging experience, but also contributing significantly to smart grids and sustainable transportation systems. Accurate predictions allow charging stations to better manage charging facilities, reducing waiting times and increasing the utilization of charging piles. Users can also use this prediction information to rationally plan charging times, avoiding long waits and improving the flexibility and convenience of travel plans.

[0065] This method uses a GMM clustering algorithm combined with the BIC criterion to determine the optimal number of clusters and uses an LSTM model to predict future charging behavior. This process makes the predictions more consistent with actual drivers' charging habits, reducing time and waste of charging station resources. This improved prediction accuracy also enhances the user's charging experience, allowing users to more rationally plan their charging schedules, avoiding the inconvenience caused by charging uncertainty, and improving charging facility utilization and user satisfaction.

[0066] This invention uses RF regression models to predict charging capacity and duration, providing data support for the construction and maintenance of charging infrastructure. This allows charging stations to dynamically adjust their service strategies based on the predictions, such as optimizing charging pile layout and scheduling charging power, thereby improving the operational efficiency and service level of the charging infrastructure. Furthermore, accurate charging capacity predictions help power companies rationally allocate power resources, reduce power waste, and achieve effective management and sustainable development of green energy. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 A flow chart of the data-driven electric vehicle charging capacity and charging time prediction method provided by the present invention;

[0068] Figure 2 This is a flow chart of electric vehicle data preprocessing in the data-driven electric vehicle charging capacity and charging time prediction method provided by the present invention;

[0069] Figure 3 A flow chart of the GMM clustering labeling charging behavior pattern in the data-driven electric vehicle charging capacity and charging time prediction method provided by the present invention;

[0070] Figure 4 A flowchart of the LSTM classification prediction of charging behavior patterns in the data-driven electric vehicle charging capacity and charging time prediction method provided by the present invention;

[0071] Figure 5 This is a flow chart of RF regression prediction of charging capacity and charging time in the data-driven electric vehicle charging capacity and charging time prediction method provided by the present invention. DETAILED DESCRIPTION

[0072] The present invention will be further described below with reference to specific embodiments, and the advantages and features of the present invention will become clearer as the description proceeds. However, these embodiments are merely exemplary and do not constitute any limitation to the scope of the present invention. It should be understood by those skilled in the art that the details and forms of the technical solutions of the present invention may be modified or replaced without departing from the spirit and scope of the present invention, and such modifications and replacements fall within the scope of protection of the present invention.

[0073] The core concept of this invention is to collect and analyze historical EV charging data to understand drivers' charging preferences and predict their charging behavior patterns. This allows for precise predictions of charging capacity and duration. Based on this data, a GMM-LSTM-RF prediction model is developed to predict drivers' charging capacity and duration within a specific future timeframe.

[0074] Example 1

[0075] like Figure 1 As shown, the present invention proposes a data-driven electric vehicle charging capacity and charging time prediction method, which includes the following steps:

[0076] Step 1, data preprocessing;

[0077] Due to noise pollution and signal transmission problems, the original sample data may be slightly missing, abnormal, or repeated. Therefore, it is necessary to clean and segment the data, which includes the following steps:

[0078] As attached Figure 2 As shown, the electric vehicle historical data is first cleaned to address missing, anomaly, and duplication issues. The data is then normalized. Finally, the charging cycle is divided. When the battery current is negative and the vehicle speed is zero, the electric vehicle is considered to be in a charging state. For the extracted charging data, if there is a gap of more than half an hour between two consecutive charging data frames, they are considered to be two independent charging cycles.

[0079] Step 2: Extract charging segment features and assign labels;

[0080] As attached Figure 3 As shown, the following steps are also included:

[0081] Step 21, extract features: each charging segment is obtained as:

[0082] S i ={F1,F2,F3,F4,F5,F6,Num n}

[0083] Among them, F1-F6 correspond to the proposed features charging start time, charging start date, charging start SOC, charging method, charging duration and charging power respectively. n Represents the corresponding electric vehicle number. Based on the electric vehicle number, the set of all charging segments for each vehicle is:

[0084] V n ={S1,…,S i ,…} T

[0085] Among them, Si Represents the charging segment characteristics.

[0086] Based on the set of all charging segments of all electric vehicles, the electric vehicle charging segment feature database is constructed as follows:

[0087] Data={V1,…,V n ,…} T

[0088] V n Represents a charging segment feature set.

[0089] Step 22: Determine the optimal number of clusters based on the BIC criterion: In this study, the number of Gaussian mixture model clusters is determined by using the BIC criterion, which is expressed as:

[0090]

[0091] Among them, p(x n |Θ) is the observed data sample x under the given model parameter Θ n The probability density function of N is the number of samples, n represents the sample number, and K is the number of clusters. From the analysis of the data, it can be concluded that as the number of clusters increases, the fitting accuracy of the model to the data shows an upward trend, which is manifested in a gradual decrease in the BIC score and the optimization of the clustering effect. However, when the number of clusters exceeds 4, the change in the BIC score tends to be flat, and the improvement in model accuracy slows down significantly. At the same time, the computational complexity and time cost rise sharply. In view of this, weighing the accuracy gain and computational cost, the present invention sets the optimal number of clusters to 4, that is, K=4, as the basis for subsequent cluster analysis.

[0092] Step 23, perform GMM clustering based on the extracted features and the optimal number of clusters: The core concept of GMM clustering is to flexibly simulate the data distribution by using a set of linearly superimposed single Gaussian components, thereby achieving refined coverage of the sample space. n Column deletion to obtain the feature matrix:

[0093]

[0094] Where N is the number of samples, f is the number of features, and according to the above, f = 6. Assuming that all samples are independently distributed, the expectation maximization method is used to solve the optimal Gaussian distribution model parameter D = {μ i ,∑ i ,α i} to determine the Gaussian distribution model. Among them, μ i ,∑ i , α iThey represent the mean vector, covariance matrix, and mixing coefficient of the Gaussian distribution respectively. Each statistic determines its most likely clustering according to the responsibility weight of each Gaussian distribution to which it belongs.

[0095] Step 24: Add the clustering results as new features to the electric vehicle charging segment data: label all charging segments of all vehicles, and the four labels represent four charging behavior modes, namely short-term low-power charging mode, balanced charging mode, long-term high-power charging mode and high-efficiency fast charging mode. n Divide all segments of all vehicles into a set of charging segments with each electric vehicle labeled by its charging behavior:

[0096] V n ′={S1′,…,S i ′,…} T

[0097] Step 3: LSTM classification model implements label prediction;

[0098] As attached Figure 4 As shown, the following steps are also included:

[0099] Step 31, feature extraction: extract each charging segment S i ′ in F1, F2, F3, F4, F7 to get x t ={F1, F2, F3, F4, F7}, where F7 is the charging behavior mode, and then all new charging segments x t Combination to get X={x1,…,x t ,…} T .

[0100] Step 32: The charging segment is calculated by the timing unit: For each charging segment, the calculation process of the LSTM unit includes the forget gate, input gate, memory unit state update and output gate. The calculation of each unit can be described as follows:

[0101]

[0102] Among them, W f 、W i 、W C 、W o They are the weight matrices of the forget gate, input gate, candidate memory unit state, and output gate, respectively, b f 、b i 、b C 、b oThey are the forget gate, input gate, candidate memory unit state, and output gate bias. Before training begins, all weight matrices and bias items are randomly initialized, and then backpropagation is used to update the weight matrix and bias items through the time algorithm. Finally, the hidden state h of this unit is obtained. t and memory cell state C t Transfer to the next time step, the hidden state h t and memory cell state C t It is a vector of length l, where l is the dimension of the LSTM hidden layer.

[0103] Step 33, the output layer realizes the prediction of charging behavior pattern and adds the corresponding electric vehicle charging segment: After processing all time steps of the time series, the hidden state h of the last time step is used T As the final feature vector for classification:

[0104] y=softmax(W y ·h T +b y )

[0105] Among them, y is the probability distribution vector of the classification, W y is the weight matrix of the output layer, b y y is the bias term. Before training begins, both the weight matrix and the bias term are randomly initialized and then updated using a backpropagation algorithm. Finally, the charging behavior pattern F7′ corresponding to the highest probability y is selected as a feature and added to the corresponding EV charging segment.

[0106] Step 4: RF regression model is used to predict charging capacity and charging time;

[0107] As attached Figure 5 As shown, the following steps are also included:

[0108] Step 41, feature extraction: extract each charging segment S i ′ and F7′ predicted by the LSTM classification model to obtain x t ′={F1,F2,F3,F4,F7′,F5,F6}, and then all new charging segments x t ′ combination to obtain X′={x1′,…,x t ′,…} T .

[0109] Step 42, extract subsets and build a decision tree: Multiple subsamples are randomly sampled from X′, and each subsample is used to train a decision tree. During the decision tree construction process, each time a node splits, a subset of features is randomly selected from all features for split selection. Each leaf node in the decision tree represents a predicted value, typically the mean of the samples in the leaf node.

[0110] Step 43, regression prediction of charging capacity and charging time: Each decision tree will make a prediction for the input sample, and the final prediction result is the average of the prediction results of all trees. Specifically, for a sample x t ′, the prediction process of random forest can be expressed as:

[0111]

[0112] in, is the final predicted value, M is the number of decision trees, is the mth decision tree for sample x t ′’s prediction results.

[0113] Example 2

[0114] The present invention also provides a data-driven electric vehicle charging capacity and charging time prediction device, which includes a data preprocessing module, a feature extraction module, an LSTM classification module and an RF regression module;

[0115] The data preprocessing module is used to preprocess the historical charging data of electric vehicles;

[0116] The feature extraction module extracts features from the pre-processed historical charging data and assigns labels to generate data with charging behavior pattern features and adds them to the electric vehicle charging segment;

[0117] The LSTM classification module adds charging behavior pattern features to the data of electric vehicle charging segments and uses the LSTM classification model to predict the charging behavior pattern;

[0118] The RF regression module uses the RF regression model to predict the charging amount and charging time based on the predicted charging behavior pattern and electric vehicle charging segment data.

[0119] This embodiment also provides an electronic device, including a processor and a memory, wherein the memory stores at least one instruction, at least one program, code set or instruction set, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by the processor to predict the charging amount and charging time of an electric vehicle driven by the data in Example 1.

[0120] This embodiment also provides a computer storage medium, which stores at least one instruction, at least one program, code set or instruction set. The at least one instruction, at least one program, code set or instruction set is loaded and executed by a processor to implement the data-driven electric vehicle charging amount and charging time prediction method as described in Example 1.

[0121] The above are only specific steps of the present invention and do not constitute any limitation to the scope of protection of the present invention; any technical solutions formed by equivalent transformation or equivalent replacement fall within the scope of protection of the present invention; the parts not elaborated in detail in the present invention belong to the common knowledge of those skilled in the art.

Claims

1. A data-driven method for predicting the charging capacity and charging time of an electric vehicle, characterized in that: The data-driven electric vehicle charging capacity and charging time prediction method includes the following steps: Perform data preprocessing on historical charging data of electric vehicles; Extract features from pre-processed historical charging data and assign labels to generate charging behavior pattern features and add them to electric vehicle charging segment data; Predict charging behavior patterns based on electric vehicle charging segment data and LSTM classification model; Predict charging capacity and charging duration based on predicted charging behavior patterns, EV charging segment data, and RF regression models; The feature extraction and labeling of the pre-processed historical charging data includes the following steps: Each charging segment is extracted from the pre-processed historical charging data as: S i ={F1,F2,F3,F4,F5,F6,Num n } Among them, F1-F6 correspond to the proposed features charging start time, charging start date, charging start SOC, charging method, charging duration and charging power respectively. n Indicates the corresponding electric vehicle number; According to the electric vehicle number, the set of all charging segments for each vehicle is: V n ={S1,…,S i ,…} T Among them, S i Indicates the charging segment characteristics; Based on the set of all charging segments of all electric vehicles, the electric vehicle charging segment feature database is constructed as follows: Data={V1,…,V n ,…} T V n represents the charging segment feature set; The optimal number of clusters is determined according to the BIC criterion, which is expressed as: Among them, BIC represents the score calculated by the BIC criterion, p(x n |Θ) is the observed data sample x under the given model parameter Θ n The probability density function of , N is the number of samples, n represents the sample number, and K is the number of clusters; GMM clustering is performed by extracting features and the optimal number of clusters; the Num n Column deletion to obtain the feature matrix: Where N is the number of samples and f is the number of features; The Gaussian distribution model is determined by solving the optimal Gaussian distribution model parameters for each sample using the expectation maximization method. The expectation maximization method is: D={μ i ,∑ i ,a i } Among them, μ i ,∑ i , α i Represent the mean vector, covariance matrix, and mixing coefficient of the Gaussian distribution respectively; each sample determines its most likely clustering according to the responsibility weight of each Gaussian distribution to which it belongs; The clustering results are added to the electric vehicle charging segment data as charging behavior pattern features; All charging segments of all vehicles are labeled with four types of labels representing four charging behavior modes: short-term low-power charging mode, balanced charging mode, long-term high-power charging mode, and high-efficiency fast charging mode; Then according to Num n Divide all segments of all vehicles into a set of charging segments with each electric vehicle labeled by its charging behavior: V n ′={S1′,…,S i ′,…} T V n ′ represents the set of charging segments without vehicle numbers, S i ' represents the charging segment without the vehicle number.

2. The data-driven electric vehicle charging capacity and charging time prediction method according to claim 1 is characterized in that: The data preprocessing of the historical charging data of the electric vehicle includes the following steps: Clean the historical charging data of electric vehicles to handle data missing, abnormal, and duplicated data; Standardize the data; The charging cycle is divided. When the battery current is negative and the vehicle speed is zero, it is considered that the electric vehicle is in the charging state.

3. The data-driven electric vehicle charging capacity and charging time prediction method according to claim 1 is characterized in that: The method of predicting charging behavior patterns based on electric vehicle charging segment data and an LSTM classification model includes the following steps: Extract each charging segment S i ′ in F1, F2, F3, F4, F7 to obtain a new charging segment x t ={F1,F2,F3,F4,F7}, where F1-F6 correspond to the proposed features of charging start time, charging start date, charging start SOC, charging method, charging duration and charging power, respectively. F7 is the charging behavior mode. Then, all new charging segments x t Combination to get X={x1,…,x t ,…} T ; For each new charging segment, the charging behavior pattern is obtained by calculating the input of the LSTM unit. The calculation process of the LSTM unit includes the forget gate, input gate, memory unit state update and output gate. The calculation of each unit is described as follows: Among them, W f 、W i 、W C 、W o They are the weight matrices of the forget gate, input gate, candidate memory unit state, and output gate, respectively, b f 、b i 、b C 、b o They are the forget gate, input gate, candidate memory unit state, and output gate bias, respectively. t-1 represents the hidden state of the previous time step, σ represents the sigmoid activation function, C t-1 Represents the memory cell state of the previous time step; Before training begins, all weight matrices and bias items of the LSTM unit are randomly initialized. During the training process, the weight matrices and bias items are updated using the backpropagation through time algorithm; finally, the hidden state h of the LSTM unit is obtained. t and memory cell state C t Transfer to the next time step, the hidden state h t and memory cell state C t Each is a vector of length l, where l is the dimension of the LSTM hidden layer; After processing all time steps in the time series, the hidden state h of the last time step is used T As the final feature vector for classification: y =softmax(W y ·h T +b y ) Among them, y is the probability distribution vector of the classification, W y is the weight matrix of the output layer, b y is the bias term, h T Represents the hidden state of the last time step; the weight matrix and bias terms are randomly initialized before training begins, and then the backpropagation algorithm is used to update the weight matrix and bias terms through time; Finally, the charging behavior pattern F7′ corresponding to the y with the highest probability is taken as the prediction result of the label.

4. The data-driven electric vehicle charging capacity and charging time prediction method according to claim 1 is characterized in that: The method uses the predicted charging behavior pattern, EV charging segment data, and RF regression model to predict the charging amount and charging duration, including the following steps: Extract each charging segment S i The charging start time F1, charging start date F2, charging start SOCF3, charging method F4, charging duration F5, charging power F6 in ′ are combined with F7′ predicted by the LSTM classification model to obtain a new charging segment x t ′={F1,F2,F3,F4,F7′,F5,F6}, and then all new charging segments x t ′ combination to obtain X′={x1′,…,x t ′,…} T ; Randomly extract multiple subsamples from X′, and each subsample is used to train a decision tree; Each decision tree predicts the input sample. For a sample x t ′, the prediction process of random forest is expressed as: in, is the final predicted value, M is the number of decision trees, is the mth decision tree for sample x t ′’s prediction results; Finally, the charging capacity and charging time are obtained through the final predicted values.

5. The data-driven electric vehicle charging capacity and charging time prediction method according to claim 2 is characterized in that: For the extracted historical charging data, if there is a gap of more than half an hour between two consecutive charging data frames, they are considered to be two independent charging cycles.

6. A data-driven electric vehicle charging capacity and charging time prediction device based on the data-driven electric vehicle charging capacity and charging time prediction method according to any one of claims 1 to 5, characterized in that: The device includes a data preprocessing module, a feature extraction module, an LSTM classification module and an RF regression module; The data preprocessing module is used to preprocess the historical charging data of the electric vehicle; The feature extraction module is used to extract features from the pre-processed historical charging data and assign labels to generate data with charging behavior pattern features added to the electric vehicle charging segment; The LSTM classification module is used to predict charging behavior patterns based on the data of electric vehicle charging segments and the LSTM classification model; The RF regression module is used to predict the charging amount and charging time based on the predicted charging behavior pattern and electric vehicle charging segment data through the RF regression model.

7. An electronic device, characterized in that: The device includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the data-driven electric vehicle charging amount and charging time prediction method as described in any one of claims 1-5.

8. A computer storage medium, characterized in that The computer storage medium stores at least one instruction, at least one program, a code set, or an instruction set, and the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by the processor to implement the data-driven electric vehicle charging amount and charging time prediction method as described in any one of claims 1 to 5.

Images