A dynamic evaluation method for electric vehicle user charging satisfaction

Abstract

The application relates to the technical field of electric vehicle charging service, and discloses a dynamic evaluation method for electric vehicle user charging satisfaction. The method continuously collects multi-source data such as vehicles, charging facilities, user interaction and preset parameters during the charging process, and extracts three types of features, i.e. charging efficiency, facility availability and cost rationality, in real time. After normalization, the features are spliced into a multi-dimensional dynamic feature vector, and a time sequence feature matrix is constructed and input into a pre-trained dynamic weight model. The model dynamically generates a three-dimensional weight vector based on the feature evolution trend, performs weighted fusion on the current features, and finally maps the fusion value to a real-time satisfaction score, so that fine and dynamic evaluation conforming to the subjective experience of users is realized.

Description

Technical Field

[0001] This invention relates to the field of electric vehicle charging service technology, and specifically to a dynamic evaluation method for electric vehicle user charging satisfaction. Background Technology

[0002] With the increasing popularity of electric vehicles, user satisfaction with charging services has become crucial for measuring the operational quality of charging facilities and improving user experience. However, traditional charging satisfaction assessments often rely on post-charging questionnaires or simple single-instance metric calculations (such as charging success rate and failure rate). These static and lagging assessment methods have significant shortcomings: First, they fail to capture dynamic changes during the charging process, such as fluctuations in charging efficiency, the real-time status of charging facilities (e.g., abnormal interface temperature, communication delays), and the real-time psychological impact of cost changes on users. Second, these methods often use fixed weights to weight a limited number of indicators, ignoring the dynamic changes in user focus on charging efficiency, facility availability, and cost reasonableness at different stages of the charging process. For example, users may be more concerned about startup speed and efficiency at the beginning of charging, while they may be more concerned about cost achievement later. This static assessment model struggles to accurately quantify the real-time changes in users' subjective feelings throughout the entire charging event.

[0003] Existing technologies lack a refined, real-time evaluation method that can integrate multi-source real-time data (vehicles, facilities, user interactions) and dynamically adjust the weights of each evaluation dimension based on the evolution of charging process characteristics. This results in a discrepancy between the evaluation results and the user's actual experience, failing to provide operators with timely and accurate service optimization basis. Summary of the Invention

[0004] To address the aforementioned technical shortcomings, the purpose of this invention is to provide a dynamic evaluation method for electric vehicle user charging satisfaction, solving the problem that existing charging evaluation systems are static and singular, and cannot accurately reflect the dynamic user experience in real time.

[0005] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: In a first aspect, the present invention provides a dynamic evaluation method for electric vehicle user charging satisfaction, the method comprising: Step S1: During the charging process, continuously collect multi-source monitoring data of the target charging event at a preset sampling period. The multi-source monitoring data includes vehicle charging process data, charging facility status data, user interaction behavior data, and user-set target charging parameters. Step S2: For each sampling time, based on the multi-source monitoring data in the current and historical windows, extract the charging efficiency features, facility availability status features, and charging cost rationality features, and then normalize each feature and concatenate them in order to generate the multi-dimensional dynamic feature vector corresponding to that sampling time. Step S3: Stack the multi-dimensional dynamic feature vectors corresponding to the current sampling time and the previous preset number of historical sampling times in chronological order to form a temporal feature matrix, and input it into the pre-trained dynamic weight model; the dynamic weight model dynamically generates a three-dimensional dynamic weight vector to characterize the current weights of charging efficiency features, facility availability status features and charging cost rationality features based on the feature evolution trend reflected by the temporal feature matrix. Step S4: Perform weighted fusion of the multidimensional dynamic feature vector and the three-dimensional dynamic weight vector at the current sampling time to obtain a weighted fusion value, and map the weighted fusion value to a preset scoring range to output the real-time satisfaction score at the current sampling time.

[0006] Preferably, in one possible implementation of the first aspect, step S1 specifically includes: Vehicle charging process data is collected through the vehicle controller local area network bus according to the sampling period. The vehicle charging process data includes the time series of real-time charging current, real-time charging voltage, real-time temperature of battery pack, and battery state of charge. By using the open charging protocol interface of the charging facility, the status data of the charging facility can be obtained. The status data of the charging facility includes the output power of the charging pile, the temperature of the charging interface, the status of the charging cable, and the network communication latency. The system captures user interaction data through the camera and touch screen logs of the charging pile. This data includes the time interval between the user's operation of the charging pile interface and the start of charging, the number of times the user actively queries during the charging process, and the status indicators of whether the user stays within the vehicle's preset range for a preset time before charging is completed, determined by image recognition. The system obtains the user-defined target charging parameters through the charging pile's touchscreen or the user's mobile terminal that communicates with the charging pile. These target charging parameters include the target state of termination of charge and the estimated charging cost.

[0007] Preferably, in one possible implementation of the first aspect, the acquisition of vehicle charging process data via the vehicle controller local area network bus includes preprocessing the acquired time-series data. Preprocessing includes: using a sliding time window to perform median filtering on the real-time temperature data of the battery pack to eliminate impulse noise, and using linear interpolation to fill in the missing points in the battery state of charge sequence caused by data transmission, thereby generating a smooth vehicle charging process data sequence.

[0008] Preferably, in one possible implementation of the first aspect, step S2, extracting charging efficiency features includes: calculating the theoretical state of charge increment per unit time based on the battery pack rated energy parameters, charging pile output power and charging time; calculating the actual state of charge increment per unit time based on the battery state of charge time series; calculating the ratio of the actual increment to the theoretical increment as the instantaneous charging efficiency ratio, and obtaining the standard deviation of the instantaneous charging efficiency ratio sequence over the entire charging event; The features of facility availability status are extracted as follows: the duration of charging interface temperature exceeding the preset safety threshold, the cumulative number of times the charging cable status is marked as abnormal, and the frequency of network communication delay exceeding the preset response threshold are normalized and then summed according to the pre-calibrated weight coefficients. Extracting the reasonableness characteristics of charging costs includes: predicting the electricity cost required from the current moment to the target end state of charge based on the real-time electricity price, the amount of electricity already charged, and the current state of battery charge; calculating the ratio of the deviation between the predicted electricity cost and the budgeted charging cost; and obtaining the cost deviation coefficient.

[0009] Preferably, in one possible implementation of the first aspect, in step S2, the normalized features are concatenated sequentially, specifically by mapping the standard deviation of the charging efficiency feature, the weighted sum of the facility availability status feature, and the cost deviation coefficient to a sigmoid function. The intervals are then concatenated into a three-dimensional vector according to a fixed order of charging efficiency, facility availability, and cost reasonableness, serving as a multi-dimensional dynamic feature vector.

[0010] Preferably, in one possible implementation of the first aspect, in step S3, the dynamic weight model is a network model based on a temporal attention mechanism, including a sequentially connected dynamic perceptual coding layer and a temporal weight generation layer. The dynamic perceptual coding layer receives the temporal feature matrix, processes the temporal data of each feature dimension in parallel, and then fuses them to output a hidden state matrix with global temporal awareness. The temporal weight generation layer is used to perform attention convergence and gating transformation on the hidden state matrix, and outputs a three-dimensional dynamic weight vector corresponding to the current sampling time.

[0011] Preferably, in one possible implementation of the first aspect, the dynamic sensing coding layer includes multiple parallel feature subnets and a feature fusion module; Each feature subnetwork corresponds to one feature dimension of the multidimensional dynamic feature vector. Its structure includes, in sequence: a one-dimensional convolutional module for extracting local temporal patterns, a layer normalization module, and a temporal self-attention module for enhancing key temporal information. The feature fusion module is used to dynamically aggregate the outputs of multiple feature subnets using learnable parameters to generate a hidden state matrix.

[0012] Preferably, in one possible implementation of the first aspect, the time-series weight generation layer includes a sequentially connected soft selection unit, a feedforward transformation unit, and a gated normalization unit; The soft selection unit performs a weighted summation of the hidden state matrix at each time step using a learnable temporal context vector to obtain a comprehensive temporal context vector, and then concatenates it with the hidden state at the current time step. The feedforward transformation unit includes at least one fully connected layer, which is used to transform the concatenated vector into a three-dimensional unnormalized weight vector. The gated normalization unit generates scaling and bias factors through the Sigmoid function to perform affine transformation on the unnormalized weight vector, and then normalizes it through the Softmax function to obtain the three-dimensional dynamic weight vector.

[0013] Preferably, in one possible implementation of the first aspect, in step S4, the multidimensional dynamic feature vector at the current sampling time is weighted and fused with the three-dimensional dynamic weight vector, and the calculation process is as follows: The three-dimensional dynamic weight vector is multiplied element-wise with the multi-dimensional dynamic feature vector at the same sample time, and the sum of all elements of the result vector is obtained to obtain the weighted fusion value.

[0014] Preferably, in one possible implementation of the first aspect, mapping the weighted fusion value to a preset scoring interval specifically involves: The weighted fusion value is input into a pre-calibrated piecewise linear mapping function, and then linearly interpolated to map it into a specific value in the range of 0 to 100, which serves as the real-time satisfaction score.

[0015] The beneficial effects of this invention are as follows: by collecting multi-source data such as vehicle, facility, user interaction and preset parameters at high frequency during the charging process, and extracting three core features in real time: charging efficiency, facility availability and cost rationality, the charging process status can be comprehensively depicted.

[0016] A pre-trained dynamic weight model based on temporal attention mechanism is introduced. This model can analyze the temporal evolution of feature vectors and dynamically generate weight vectors that represent the current importance of each feature, reflecting the dynamic changes in the user's focus at different stages of charging.

[0017] The dynamic evaluation mechanism makes the output of real-time satisfaction scores more in line with users' subjective psychological feelings, thereby improving the accuracy and timeliness of the evaluation.

[0018] At the same time, it provides charging operators with high-time-resolution user experience insights, which helps to identify service shortcomings in the charging process in a timely manner, and to provide early warnings and interventions, thereby optimizing operational strategies and improving overall service quality and user satisfaction. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This application provides a flowchart of a dynamic evaluation method for electric vehicle user charging satisfaction. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Example 1: As Figure 1 As shown, the present invention provides a dynamic evaluation method for electric vehicle user charging satisfaction, comprising: Step S1: During the charging process, continuously collect multi-source monitoring data of the target charging event at a preset sampling period. The multi-source monitoring data includes vehicle charging process data, charging facility status data, user interaction behavior data, and user-set target charging parameters.

[0023] In this embodiment, the preset sampling period is 1 second. Vehicle charging process data is collected via a connection to the vehicle controller's local area network bus. Specifically, time-series data of the battery pack's real-time charging current, real-time charging voltage, real-time battery pack temperature, and battery state of charge (SOC) are collected. The collected time-series data is preprocessed. For the battery pack's real-time temperature data, a sliding time window of 5 seconds is used for median filtering to eliminate impulse noise. This filtering process is as follows: at each sampling moment, five temperature data points (the current moment and the previous four historical moments) are taken, sorted by value, and the median is taken as the filtered temperature value for the current moment. For the battery SOC sequence, due to potential missing points in data transmission, linear interpolation is used to fill in the gaps. For a missing battery SOC value, a linear function is constructed using the preceding and following valid data points to calculate the corresponding battery SOC value at the missing moment, thereby generating a smooth vehicle charging process data sequence.

[0024] The status data of the charging facility is obtained through the open charging protocol interface of the charging facility. Specifically, it obtains the real-time output power of the charging pile, the real-time temperature of the charging interface, the status identifier of the charging cable (the charging cable status identifier is a binary status, 0 indicates that the cable connection is normal, and 1 indicates that the cable connection is abnormal), and the network communication latency between the charging pile and the backend server.

[0025] User interaction data is captured through the cameras and touchscreen logs of the charging pile. The interval between the user's operation of the charging pile interface and the start of charging is calculated by recording the timestamp of the user clicking the "Start Charging" button on the touchscreen and the timestamp of the charging relay actually closing successfully as reported by the charging pile controller. The number of times the user actively queries during the charging process is obtained by counting the number of log entries in the touchscreen logs for user-triggered queries (including queries for charging status, fees, and remaining time) after charging starts and before charging ends. Whether the user remains within a preset range around the vehicle for a preset time before charging is completed is determined by capturing video streams from the charging pile's cameras and running an image recognition algorithm. The preset time is 5 minutes, and the preset range is 3 meters around the vehicle. The image recognition algorithm uses frame difference-based personnel presence detection. The specific process is as follows: starting 5 minutes before the end of charging, the camera image is sampled every second, and the absolute difference between pixels between two consecutive frames is calculated. If the difference exceeds a set threshold, it is determined that there is personnel movement. Combined with a preset area mask, it is determined whether the personnel are within 3 meters of the vehicle. If more than 50% of the sampling moments in the 5 minutes before the end of charging are detected that the personnel are within range, the stationary status is marked as 1 (indicating stationary), otherwise it is marked as 0 (indicating no stationary).

[0026] The target charging parameters set by the user are obtained through the charging pile's touchscreen or the user's mobile terminal application that establishes a communication connection with the charging pile. The target charging parameters include a target state of charge (SOC) as a percentage value and a charging cost budget as a specific monetary amount.

[0027] Step S2: For each sampling time, based on the multi-source monitoring data in the current and historical windows, extract the charging efficiency features, facility availability status features, and charging cost rationality features, and then normalize each feature and concatenate them in order to generate the multi-dimensional dynamic feature vector corresponding to that sampling time.

[0028] In this embodiment, for each sampling time, based on the current time and a length of Feature extraction is performed on multi-source monitoring data within a historical window of seconds. The specific process for extracting charging efficiency features is as follows: First, based on the rated energy parameters of the vehicle battery pack... Charging pile output power and unit sampling period Calculate the theoretical state of charge increment per unit time. The formula is:

[0029] Secondly, based on the preprocessed battery state-of-charge time series Calculate the actual state of charge increment per unit time. The formula is:

[0030] Next, the ratio of the actual increment to the theoretical increment is calculated as the instantaneous charging efficiency ratio at that sampling moment. The formula is:

[0031] Finally, this charging event is analyzed from its start to the current sampling time. The time series consisting of all the instantaneous charging efficiency ratios generated Calculate its standard deviation The standard deviation is the characteristic value of charging efficiency, which characterizes the stability of the charging process efficiency. The formula is:

[0032] in This is the mean of the sequence.

[0033] The specific process for extracting facility availability status characteristics involves calculating three sub-indicators. The first sub-indicator is that the charging interface temperature exceeds a preset safety threshold. Duration percentage At the current sampling time, the historical data window is statistically analyzed. All internal temperature data Exceed Number of data points ,calculate The second sub-metric is the cumulative number of times the charging cable status is marked as abnormal. In the history window Inside, the cumulative charging cable status indicator The number of times the value changes from 0 to 1. The third sub-metric is the network communication latency exceeding a preset response threshold. frequency In the history window Internally, statistical network communication delay data Exceed The number of data points. Then, normalization is performed on each of these three sub-indicators. and By dividing by the historical window length Normalization is performed to obtain and Cumulative number of times Through a preset maximum possible number of times Normalization is performed to obtain The normalized values ​​of the three sub-indicators are then adjusted according to pre-defined weighting coefficients. , , Perform a weighted summation to obtain the weighted sum value. This is the characteristic value of facility availability status, and the formula is:

[0034] The specific process for extracting the reasonableness characteristics of charging costs is as follows: First, based on the real-time electricity price... The current cumulative charge amount and the current state of battery charge. Predicts the charging time from the current moment to the user-defined target state of charge. Required electricity cost The predicted electricity cost equals the energy required to charge the battery pack from its current state of charge to the target end state of charge, multiplied by the real-time electricity price. The required energy is based on the battery pack's rated capacity. The difference between the target termination state of charge and the current state of charge is calculated. The formula for predicting electricity charges is:

[0035] Next, the predicted electricity cost is calculated and compared with the user's set charging cost budget. The ratio of the deviations between them. Cost deviation coefficient. This is the characteristic value of the reasonableness of charging costs, and the formula is:

[0036] After calculating the above three feature values, feature normalization and vector concatenation are performed. The standard deviation of the charging efficiency feature is then calculated. Weighted sum of facility availability status characteristics and cost deviation coefficient through respectively The function maps to the interval between 0 and 1. The specific form of the function is ,in The input feature values ​​are used. After mapping, we get:

[0037] Following a fixed order of charging efficiency characteristics, facility availability characteristics, and charging cost reasonableness characteristics, the three values ​​are concatenated into a three-dimensional vector. This three-dimensional vector represents the current sampling time. The corresponding multidimensional dynamic feature vector.

[0038] Step S3: Stack the multi-dimensional dynamic feature vectors corresponding to the current sampling time and the previous preset number of historical sampling times in chronological order to form a temporal feature matrix, and input it into the pre-trained dynamic weight model; the dynamic weight model dynamically generates a three-dimensional dynamic weight vector to characterize the current weights of charging efficiency features, facility availability status features and charging cost rationality features based on the feature evolution trend reflected by the temporal feature matrix.

[0039] In this embodiment, the preset number of historical sampling times The value is 60. The process of constructing the time-series feature matrix is ​​as follows: for the current sampling time... Get from time At the time Continuous The multidimensional dynamic feature vectors corresponding to each sampling time point. The three-dimensional vectors corresponding to each sampling time point are stacked from top to bottom in chronological order to form a multidimensional... The matrix is ​​denoted as This matrix is ​​the time series feature matrix.

[0040] The pre-trained dynamic weight model is a network model based on a temporal attention mechanism, and its input is a temporal feature matrix. The output is a three-dimensional dynamic weight vector. The model consists of three components that dynamically represent the current-moment weights of charging efficiency, facility availability, and charging cost reasonableness. The model comprises a sequentially connected dynamic sensing encoding layer and a temporal weight generation layer.

[0041] The dynamic perceptual coding layer is used to receive the temporal feature matrix. After processing the temporal data of each feature dimension in parallel, the data is fused to output a hidden state matrix with global temporal awareness. This layer contains three parallel feature subnetworks and one feature fusion module. Each feature subnetwork corresponds to one feature dimension of the multidimensional dynamic feature vector, specifically handling the length of that dimension. The time series data is analyzed. Each feature subnetwork consists of: a one-dimensional convolutional module, a layer normalization module, and a temporal self-attention module. The one-dimensional convolutional module uses multiple convolutional kernels of size 3 and stride 1, sliding along the time axis to extract local temporal patterns of the feature sequence. The layer normalization module normalizes the convolutional output across all channels at each time step, calculated by normalizing the feature vector at a single time step. Calculate the mean With variance Then standardize: ,in A very small constant is used for numerical stability. The temporal self-attention module is used to enhance key temporal information in the sequence and model long-range dependencies. Its calculation process is as follows: First, the input sequence is... Through a learnable weight matrix , , Linear projections into query matrices respectively Key matrix Sum matrix .

[0042] Then, the attention score matrix is ​​calculated: ,in Let be the dimension of the key vector. Finally, calculate the self-attention output: The three feature subnets each output three time-series representation matrices. ,in To hide the state dimension, this embodiment sets it to 64. The feature fusion module uses a learnable parameter matrix. The three representations are dynamically aggregated to generate the final hidden state matrix. :

[0043] Temporal weight generation layer for hidden state matrix Attention convergence and gating transformation are performed to output a three-dimensional dynamic weight vector. This layer consists of sequentially connected soft selection units, feedforward transformation units, and gated normalization units. The soft selection unit first uses a learnable time context vector... ,calculate Attention weights at each time step :

[0044] in yes The Row vectors. Then, the hidden states at all time steps are weighted and summed to obtain the comprehensive temporal context vector. Next, Representing the current moment Hidden state The vectors are concatenated to form an enhanced representation vector. .

[0045] The feedforward transform unit contains at least one fully connected layer, which concatenates the enhanced representation vector. Mapped to a three-dimensional unnormalized weight vector :

[0046] in and These are learnable parameters.

[0047] The gating normalization unit first passes through Function generates scaling factor and bias factor : , ,in for function, These are learnable parameters. Next, an affine transformation is performed on the unnormalized weight vector: ,in This indicates element-wise multiplication. Finally, through... The function normalizes the transformed vector to ensure that the sum of all weights is 1, thus obtaining the final three-dimensional dynamic weight vector:

[0048] The dynamic weight model is trained based on historical charging event data. The training dataset consists of a large number of charging process samples, each containing a temporal feature matrix. and its corresponding expert-annotated true 3D weight vector The goal of model training is to minimize the mean squared error loss function between the predicted weights and the true weights.

[0049] in For batch size. Model used. The optimizer updates its parameters and learns the mapping relationship from temporal feature evolution patterns to feature importance weights through backpropagation. After training, the model parameters are fixed for inference. During real-time evaluation, the constructed temporal feature matrix is ​​input into this pre-trained model, which dynamically generates the three-dimensional dynamic weight vector corresponding to the current sampling time.

[0050] Step S4: Perform weighted fusion of the multidimensional dynamic feature vector and the three-dimensional dynamic weight vector at the current sampling time to obtain a weighted fusion value, and map the weighted fusion value to a preset scoring range to output the real-time satisfaction score at the current sampling time.

[0051] In this embodiment, the specific calculation process of weighted fusion is as follows: The current sampling time is... The corresponding multidimensional dynamic feature vector is denoted as The corresponding three-dimensional dynamic weight vector generated by the dynamic weight model is denoted as... The weighted fusion operation is defined as the element-wise multiplication of a three-dimensional dynamic weight vector with a multi-dimensional dynamic feature vector, followed by summing all elements of the resulting vector to obtain a scalar value, namely the weighted fusion value St. Its mathematical expression is:

[0052] in, and They represent vectors respectively and The Each element. This calculation achieves a weighted summation of the original feature values ​​based on the dynamic importance of the features.

[0053] Obtain the weighted fusion value Then, it is mapped to a preset rating interval to output a real-time satisfaction score. This embodiment uses a pre-calibrated piecewise linear mapping function. After completing this mapping, the function will weight the fused values. Transformed into a specific value within the range of 0 to 100 through linear interpolation. Piecewise linear mapping function. The definition is based on multiple pre-calibrated boundary points. This embodiment sets a weighted fusion value. The theoretical range is And set two dividing points within this interval. and and the corresponding scoring boundaries and The specific form of the mapping function is as follows:

[0054] in, This is the real-time satisfaction score at the current sampling moment. (Function) The dividing point , and the finish line The calibration was obtained through historical charging event data. The calibration method involved collecting the final weighted fusion values ​​of a large number of charging events and their corresponding user survey satisfaction scores. Regression analysis was then used to determine these key points, ensuring the highest consistency between the mapped scores and users' subjective evaluations. During the calculation of this piecewise linear mapping process, the first step was to determine… The interval to which it belongs is then used to calculate the final real-time satisfaction score based on the linear interpolation formula for that interval. And output it.

[0055] Example 2: This invention provides a dynamic evaluation method for electric vehicle user charging satisfaction, and also includes the training and verification of a dynamic weight model.

[0056] Specifically, the training of the dynamic weight model is based on a dataset constructed from historical charging event data. Each sample contains a temporal feature matrix composed of stacked multi-dimensional dynamic feature vectors, and a three-dimensional true weight vector labeled by experts based on the comprehensive performance of the charging process.

[0057] The dynamic weight model is trained in a supervised manner by minimizing the mean squared error loss between the predicted dynamic weight vector and the true weight vector. The optimizer uses the Adam algorithm and employs an early stopping strategy to prevent overfitting.

[0058] After the dynamic weight model is trained, it is evaluated on an independent validation set. The accuracy and rationality of its dynamic allocation of feature weights are measured by calculating the mean cosine similarity between the predicted weights and the expert-annotated weights. Only after the validation is passed can it be deployed for real-time inference.

[0059] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A dynamic evaluation method for electric vehicle user charging satisfaction, characterized in that, The method includes: Step S1: During the charging process, continuously collect multi-source monitoring data of the target charging event at a preset sampling period. The multi-source monitoring data includes vehicle charging process data, charging facility status data, user interaction behavior data, and user-set target charging parameters. Step S2: For each sampling time, based on the multi-source monitoring data in the current and historical windows, extract the charging efficiency features, facility availability status features, and charging cost rationality features, and then normalize each feature and concatenate them in order to generate the multi-dimensional dynamic feature vector corresponding to that sampling time. Step S3: Stack the multi-dimensional dynamic feature vectors corresponding to the current sampling time and the previous preset number of historical sampling times in chronological order to form a temporal feature matrix, and input it into the pre-trained dynamic weight model; the dynamic weight model dynamically generates a three-dimensional dynamic weight vector to characterize the current weights of charging efficiency features, facility availability status features and charging cost rationality features based on the feature evolution trend reflected by the temporal feature matrix. Step S4: Perform weighted fusion of the multidimensional dynamic feature vector and the three-dimensional dynamic weight vector at the current sampling time to obtain a weighted fusion value, and map the weighted fusion value to a preset scoring range to output the real-time satisfaction score at the current sampling time.

2. The dynamic evaluation method for electric vehicle user charging satisfaction as described in claim 1, characterized in that, Step S1 specifically includes: Vehicle charging process data is collected through the vehicle controller local area network bus according to the sampling period. The vehicle charging process data includes the time series of real-time charging current, real-time charging voltage, real-time temperature of battery pack, and battery state of charge. By using the open charging protocol interface of the charging facility, the status data of the charging facility can be obtained. The status data of the charging facility includes the output power of the charging pile, the temperature of the charging interface, the status of the charging cable, and the network communication latency. The system captures user interaction data through the camera and touch screen logs of the charging pile. This data includes the time interval between the user's operation of the charging pile interface and the start of charging, the number of times the user actively queries during the charging process, and the status indicators of whether the user stays within the vehicle's preset range for a preset time before charging is completed, determined by image recognition. The system obtains the user-defined target charging parameters through the charging pile's touchscreen or the user's mobile terminal that communicates with the charging pile. These target charging parameters include the target state of termination of charge and the estimated charging cost.

3. The dynamic evaluation method for electric vehicle user charging satisfaction as described in claim 2, characterized in that, The process of collecting vehicle charging process data through the vehicle controller local area network bus includes preprocessing the collected time series data. Preprocessing includes: using a sliding time window to perform median filtering on the real-time temperature data of the battery pack to eliminate impulse noise, and using linear interpolation to fill in the missing points in the battery state of charge sequence caused by data transmission, thereby generating a smooth vehicle charging process data sequence.

4. The dynamic evaluation method for electric vehicle user charging satisfaction as described in claim 1, characterized in that, In step S2, extracting charging efficiency features includes: calculating the theoretical state of charge increment per unit time based on the battery pack's rated energy parameters, the charging pile's output power, and the charging time; calculating the actual state of charge increment per unit time based on the battery's state of charge time series; calculating the ratio of the actual increment to the theoretical increment as the instantaneous charging efficiency ratio, and obtaining the standard deviation of the instantaneous charging efficiency ratio sequence over the entire charging event. The features of facility availability status are extracted as follows: the duration of charging interface temperature exceeding the preset safety threshold, the cumulative number of times the charging cable status is marked as abnormal, and the frequency of network communication delay exceeding the preset response threshold are normalized and then summed according to the pre-calibrated weight coefficients. Extracting the reasonableness characteristics of charging costs includes: predicting the electricity cost required from the current moment to the target end state of charge based on the real-time electricity price, the amount of electricity already charged, and the current state of battery charge; calculating the ratio of the deviation between the predicted electricity cost and the budgeted charging cost; and obtaining the cost deviation coefficient.

5. The dynamic evaluation method for electric vehicle user charging satisfaction as described in claim 4, characterized in that, In step S2, the features are normalized and then concatenated in sequence. Specifically, the standard deviation of the charging efficiency feature, the weighted sum of the facility availability status feature, and the cost deviation coefficient are mapped to a sigmoid function. The intervals are then concatenated into a three-dimensional vector according to a fixed order of charging efficiency, facility availability, and cost reasonableness, serving as a multi-dimensional dynamic feature vector.

6. The dynamic evaluation method for electric vehicle user charging satisfaction as described in claim 1, characterized in that, In step S3, the dynamic weight model is a network model based on the temporal attention mechanism, which includes a sequentially connected dynamic perceptual encoding layer and a temporal weight generation layer. The dynamic perceptual coding layer receives the temporal feature matrix, processes the temporal data of each feature dimension in parallel, and then fuses them to output a hidden state matrix with global temporal awareness. The temporal weight generation layer is used to perform attention convergence and gating transformation on the hidden state matrix, and outputs a three-dimensional dynamic weight vector corresponding to the current sampling time.

7. The dynamic evaluation method for electric vehicle user charging satisfaction as described in claim 6, characterized in that, The dynamic sensing coding layer includes multiple parallel feature subnets and a feature fusion module. Each feature subnetwork corresponds to one feature dimension of the multidimensional dynamic feature vector. Its structure includes, in sequence: a one-dimensional convolutional module for extracting local temporal patterns, a layer normalization module, and a temporal self-attention module for enhancing key temporal information. The feature fusion module is used to dynamically aggregate the outputs of multiple feature subnets using learnable parameters to generate a hidden state matrix.

8. The dynamic evaluation method for electric vehicle user charging satisfaction as described in claim 7, characterized in that, The temporal weight generation layer includes a sequentially connected soft selection unit, a feedforward transformation unit, and a gated normalization unit; The soft selection unit performs a weighted summation of the hidden state matrix at each time step using a learnable temporal context vector to obtain a comprehensive temporal context vector, and then concatenates it with the hidden state at the current time step. The feedforward transformation unit includes at least one fully connected layer, which is used to transform the concatenated vector into a three-dimensional unnormalized weight vector. The gated normalization unit generates scaling and bias factors through the Sigmoid function to perform affine transformation on the unnormalized weight vector, and then normalizes it through the Softmax function to obtain the three-dimensional dynamic weight vector.

9. The dynamic evaluation method for electric vehicle user charging satisfaction as described in claim 1, characterized in that, In step S4, the multidimensional dynamic feature vector and the three-dimensional dynamic weight vector at the current sampling time are weighted and fused. The calculation process is as follows: The three-dimensional dynamic weight vector is multiplied element-wise with the multi-dimensional dynamic feature vector at the same sample time, and the sum of all elements of the result vector is obtained to obtain the weighted fusion value.

10. The dynamic evaluation method for electric vehicle user charging satisfaction as described in claim 9, characterized in that, The process of mapping the weighted fusion value to a preset scoring range specifically involves: The weighted fusion value is input into a pre-calibrated piecewise linear mapping function, and then linearly interpolated to map it into a specific value in the range of 0 to 100, which serves as the real-time satisfaction score.

Images