A charging pile charging fee optimization method and system based on dynamic rate

Abstract

This invention relates to the field of electric vehicle charging technology, and discloses a method and system for optimizing charging fees for charging piles based on dynamic rates. The method includes: collecting multi-dimensional feature data and user charging data, performing preprocessing and feature extraction; constructing and training a charging demand prediction model to predict charging demand in real time; constructing a price response model and combining it with a charging load prediction curve for dynamic rate optimization; performing user cluster analysis and designing personalized incentive strategies; setting rate information release rules and dynamically scheduling charging resources; collecting operational data and user response data in real time, and optimizing the charging demand prediction model and price response model based on the charging load prediction curve; and performing data encryption and privacy protection. This invention, through a personalized incentive mechanism, formulates differentiated incentive strategies for three types of users: price-sensitive, time-sensitive, and convenience-first.

Description

Technical Field

[0001] This invention relates to the field of electric vehicle charging technology, and more specifically, to a method and system for optimizing charging fees for charging piles based on dynamic rates. Background Technology

[0002] With the rapid growth of electric vehicle ownership, the scale of charging infrastructure construction is constantly expanding, but the uneven spatial and temporal distribution of charging demand is becoming increasingly prominent. In areas with high charging demand, such as urban business districts and residential communities, charging load exhibits obvious morning and evening peak characteristics, with peak demand occurring between 8-9 am and 6-7 pm. During working hours (10 am to 4 pm) and at night, charging demand decreases, leaving many charging stations idle. This uneven demand distribution leads to a shortage of charging stations during peak hours, resulting in long waiting times for users, while during off-peak hours, the utilization rate of charging stations is extremely low, leading to inefficient resource allocation.

[0003] Existing charging pile management systems mostly adopt a fixed-rate model, which cannot dynamically adjust according to changes in grid load and charging demand, making it difficult to leverage the guiding role of price signals. When peak charging times coincide with peak regional electricity consumption, it causes significant impact on the power grid, increases grid operation risks, and also drives up electricity costs. Traditional charging pile scheduling methods are mainly based on a first-come, first-served principle, lacking in-depth analysis of user charging behavior characteristics and personalized service capabilities, and thus failing to effectively guide users to charge during off-peak hours.

[0004] Furthermore, existing systems often rely on simple historical data statistics for charging demand forecasting, failing to adequately consider the combined impact of multiple factors such as weather conditions, traffic conditions, and user behavior, resulting in limited forecast accuracy. Regarding user incentive mechanisms, there is a lack of differentiated strategies designed for different user types, leading to ineffective incentives. These problems hinder the efficient utilization of charging infrastructure and impede the healthy development of the electric vehicle industry. Summary of the Invention

[0005] This invention provides a charging pile charging fee optimization method and system based on dynamic rates, which solves the technical problems of uneven spatial and temporal distribution of charging demand, large fluctuations in power grid load, and low utilization rate of charging resources in related technologies.

[0006] This invention provides a method for optimizing charging station fees based on dynamic rates, comprising the following steps: Collect multidimensional feature data and user charging data, perform preprocessing and feature extraction to obtain a standardized multidimensional feature dataset and user charging behavior features; Based on a standardized multidimensional feature dataset, a charging demand prediction model is constructed and trained to predict charging demand in real time and obtain a charging load prediction curve. A price response model is constructed based on user charging behavior characteristics, and dynamic rate optimization is performed by combining the charging load prediction curve to obtain a dynamic rate scheme. Based on user charging behavior characteristics, user clustering analysis is conducted to design personalized incentive strategies and obtain differentiated incentive schemes for different user types. Based on dynamic rate schemes and differentiated incentive schemes for different user types, rate information release rules are set, charging resources are dynamically scheduled, and rate release schemes and resource scheduling results are obtained. Based on the rate release scheme and resource scheduling results, real-time operational data and user response data are collected. Combined with the charging load prediction curve, the charging demand prediction model and price response model are optimized to obtain optimized charging demand prediction parameters and corrected price response parameters. Based on optimized charging demand forecasting parameters and corrected price response parameters, data encryption and privacy protection are implemented to obtain tamper-proof transaction credentials.

[0007] In a preferred embodiment, the process of collecting multidimensional feature data and user charging data, and performing preprocessing and feature extraction includes: The charging pile's operating parameters are read and transmitted in real time through the IoT data acquisition module to obtain the charging pile's real-time operating data stream. By connecting to the regional power grid monitoring system through a data proxy server, real-time data on the regional power grid load can be obtained. Use the meteorological open data API and the transportation open data API to obtain environmental impact factor data; Extract historical charging records from the user behavior database and calculate user charging behavior characteristics; Time alignment is performed on multidimensional feature data and user charging behavior features. The multidimensional feature data includes real-time operation data streams of charging piles, real-time data of regional power grid load, and data on environmental influencing factors. Abnormal data is identified and outliers are repaired to obtain a cleaned dataset. Time series samples are constructed based on the cleaned dataset, and time features, periodic features, and trend features are extracted and scaled to a preset range to obtain a standardized multidimensional feature dataset. The standardized multidimensional feature dataset is then divided into training set, validation set, and test set according to time order.

[0008] In a preferred embodiment, retrieving historical charging records from the user behavior database includes: Extract user charging data from the user behavior database, including user ID, charging time, charging location, charging duration, charging amount, charging cost, vehicle model, and battery capacity. Statistical analysis is performed on the historical charging records of individual users to calculate user charging behavior characteristics; Normalize the characteristics of user charging behavior.

[0009] In a preferred embodiment, the step of constructing and training a charging demand prediction model to predict charging demand in real time includes: Construct a charging demand prediction model that integrates convolutional neural networks and long short-term memory networks; The training set from the standardized multidimensional feature dataset is input into the charging demand prediction model for training, resulting in a trained charging demand prediction model. The performance of the trained charging demand prediction model is evaluated using a test set of a standardized multidimensional feature dataset, and a performance evaluation report is generated. Multidimensional feature data is collected in real time and input into the trained charging demand prediction model. Multi-step prediction is performed using a rolling prediction method, and the prediction results are smoothed to obtain the charging load prediction curve.

[0010] In a preferred embodiment, the dynamic rate optimization includes: Calculate the statistical characteristics of the load curve, including maximum load, minimum load, load peak-to-valley difference, and load factor, and use the load variance as the optimization objective function; Establish a price response model to quantify the relationship between rates and demand, estimate different demand price elasticity coefficients for different user groups, and establish a demand transfer matrix to describe the demand transfer relationship between different time periods. A multi-objective optimization function is constructed that simultaneously considers peak shaving and valley filling of power grid load and reduction of user charging costs. The weighted summation method is used to transform the multi-objective into a single objective. Set the rate optimization constraints to form the constraint set of the optimization problem and define the feasible solution space; The particle swarm optimization algorithm is used to solve the multivariable nonlinear constrained optimization problem. The particle swarm is initialized and the initial particle positions are randomly generated in the feasible solution space. The new velocity and new position of the particles are calculated to obtain the dynamic rate scheme.

[0011] In a preferred embodiment, estimating different price elasticity of demand coefficients for different user groups includes: Users are categorized into price-sensitive users, time-sensitive users, and convenience-priority users; the absolute values ​​of the price elasticity coefficients of demand for price-sensitive users, time-sensitive users, and convenience-priority users are then set. Calculate the change in demand after the rate adjustment based on the rate adjustment range and the demand price elasticity coefficient of the corresponding user type; Establish a demand transfer matrix to describe the demand transfer relationships between different time periods.

[0012] In a preferred embodiment, the personalized incentive strategy includes: Design a points reward mechanism for price-sensitive users, linking the number of points to the rate discount, setting up a points accumulation mechanism and a redemption mechanism, and establishing a points tier system; A priority service mechanism is designed for time-sensitive users. Users can choose to pay a priority rate to obtain charging priority. A priority charging queue and reservation mechanism are established, and the supply and demand balance is adjusted through the rate adjustment. A smart recommendation mechanism is designed for users who prioritize convenience. A comprehensive scoring model for charging stations is established, and the scores of each dimension are weighted, summed, and recommended in order of score.

[0013] In a preferred embodiment, the step of setting rate information publishing rules and dynamically scheduling charging resources includes: Establish a rate information release rule engine, including setting rate forecast release rules, rate locking rules, emergency rate adjustment rules, and rate recovery rules; Establish a rate subscription and reminder service, allow users to set personalized rate reminder conditions, push reminder service, and establish a push frequency control mechanism; A smart scheduling model for charging resources is constructed to match user charging needs with charging pile resources, thereby obtaining the optimal matching relationship between users and charging piles. A decision matrix is ​​established, and the decision matrix is ​​normalized to calculate the weighted comprehensive score to determine the optimal matching charging pile. Establish a charging reservation and resource locking mechanism to achieve precise scheduling of charging resources through time window allocation and resource locking, and obtain resource scheduling results.

[0014] In a preferred embodiment, the data encryption and privacy protection includes: Establish a data encryption storage mechanism to encrypt and store user identity information, vehicle information, charging records, location information, and payment information; Establish user authentication and access control mechanisms to implement multi-factor authentication and role-based access control. The charging transaction record is constructed using blockchain technology. The transaction record is stored in a distributed manner. After the charging transaction is completed, a transaction record is generated and hashed. The transaction settlement is automatically executed by smart contract to obtain a tamper-proof transaction certificate.

[0015] In a preferred embodiment, a charging pile charging fee optimization system based on dynamic rates is used to execute the above-described charging pile charging fee optimization method based on dynamic rates, including: The data acquisition module is used to collect multidimensional feature data and user charging data, perform preprocessing and feature extraction, and obtain standardized multidimensional feature datasets and user charging behavior features. The demand forecasting module constructs and trains a charging demand forecasting model based on a standardized multidimensional feature dataset, forecasts charging demand in real time, and obtains a charging load forecast curve. The rate optimization module constructs a price response model based on user charging behavior characteristics, combines it with the charging load prediction curve, and performs dynamic rate optimization to obtain a dynamic rate scheme. The user analysis module performs user clustering analysis based on user charging behavior characteristics, designs personalized incentive strategies, and obtains differentiated incentive schemes for different user types. The resource scheduling module sets rules for publishing rate information based on dynamic rate schemes and differentiated incentive schemes for different user types, dynamically schedules charging resources, and obtains rate publishing schemes and resource scheduling results. The optimization module, based on the rate release scheme and resource scheduling results, collects operational data and user response data in real time, and combines them with the charging load prediction curve to optimize the charging demand prediction model and the price response model, thereby obtaining optimized charging demand prediction parameters and corrected price response parameters. The security protection module, based on optimized charging demand forecasting parameters and corrected price response parameters, performs data encryption and privacy protection to obtain tamper-proof transaction credentials.

[0016] The beneficial effects of this invention are as follows: This invention constructs a charging demand prediction model that integrates multi-dimensional feature data. This model comprehensively considers charging pile operation data, grid load data, meteorological and traffic data, and user charging behavior characteristics. Employing deep learning technology, it achieves high-precision spatiotemporal charging demand prediction, providing a scientific basis for dynamic pricing. Simultaneously, it establishes a grid-friendly dynamic pricing algorithm. Through multi-objective optimization, it achieves the dual goals of peak shaving and valley filling of grid load and reducing user charging costs, effectively alleviating grid pressure and improving the efficiency of charging resource utilization.

[0017] This invention designs a personalized incentive mechanism based on user profiles, and formulates differentiated incentive strategies for three types of users: price-sensitive, time-sensitive, and convenience-first. These strategies include points rewards, priority services, and intelligent recommendations to enhance users' enthusiasm for participating in off-peak charging. Through a distributed computing architecture with edge cloud collaboration and blockchain technology, the system's real-time response capability and data security are ensured, enabling intelligent scheduling and optimized configuration of charging load. This provides a complete technical solution for the efficient operation of electric vehicle charging infrastructure. Attached Figure Description

[0018] Figure 1 This is a flowchart of the main process of a charging pile charging fee optimization method based on dynamic rates in this invention. Figure 2 This is a detailed flowchart of a charging pile charging fee optimization method based on dynamic rates in this invention; Figure 3This is a block diagram of a charging pile charging fee optimization system based on dynamic rates in this invention. Detailed Implementation

[0019] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, some features described in the examples may be combined in other examples.

[0020] At least one embodiment of the present invention discloses a charging pile charging fee optimization method based on dynamic rates, such as... Figures 1 to 2 As shown, it includes the following steps: Step 1: Collect multidimensional feature data and user charging data, perform preprocessing and feature extraction to obtain a standardized multidimensional feature dataset and user charging behavior features; Step 1.1: Deploy the charging pile IoT data acquisition module to transmit operational data and obtain the real-time operational data stream of the charging pile; An IoT data acquisition module is integrated into each charging pile controller. This module connects to the main controller of the charging pile via a CAN bus interface to read the charging pile's operating parameters in real time, obtaining a real-time operating data stream. The collected operating parameters include charging pile number, charging pile type, charging power, charging current, charging voltage, charging capacity, charging start time, charging end time, charging status identifier, and fault codes. The charging status identifier includes status types such as idle, charging in progress, fault, and maintenance.

[0021] The IoT data acquisition module communicates with the edge server using MQTT (Message Queuing Telemetry Transport). The MQTT protocol is based on a publish-subscribe model, with the charging pile acting as an MQTT client and the edge server acting as an MQTT broker. The charging pile publishes operational data to the edge server every 5 seconds. The data is encapsulated in JSON format, and the JSON object contains a timestamp field, a device identifier field, a data type field, and a numeric field. The timestamp field records the precise time the data was generated, using a Unix timestamp format with millisecond precision. The device identifier field uses a globally unique identifier to ensure accurate differentiation between data from different charging piles.

[0022] When the network connection is interrupted, the IoT data acquisition module activates the local caching mechanism to temporarily store the data in the local storage. The storage capacity can support 48 hours of offline caching. After the network connection is restored, the cached data is automatically uploaded to the edge server in chronological order to ensure data integrity.

[0023] Step 1.2: Connect to the power grid monitoring system interface and use the data proxy service conversion protocol to obtain real-time data on the regional power grid load; The system connects to the regional power grid monitoring system via a data interface to obtain real-time load data. To achieve protocol conversion, a data proxy server is deployed. One end of the data proxy server connects to the regional power grid monitoring system using the IEC61850 protocol, while the other end provides a RESTful API interface. The data proxy server reads real-time power grid operating parameters such as load data, voltage data, frequency data, and power factor data from the regional substations from the regional power grid monitoring system, converts the data format to JSON format, and responds to data requests from the system via the HTTP protocol.

[0024] The power grid data collected from the regional power grid monitoring system includes the total regional load power, load curves for each time period, real-time voltage values, power grid frequency, and time-of-use electricity price information. Among them, the time-of-use electricity price information includes the current electricity price, the electricity price for each future time period, and the division of peak, flat, and valley periods.

[0025] The data update frequency is set to 1 minute, meaning the latest data is requested from the regional power grid monitoring system every minute. Considering the security requirements of the regional power grid monitoring system, the data interface uses digital certificate authentication and SSL encrypted transmission to ensure secure data transmission. A data interface health monitoring mechanism is established to monitor interface response time and data update latency in real time. When the interface response time exceeds a set threshold or data updates are interrupted, an alarm message is sent to the system administrator to promptly troubleshoot interface faults.

[0026] Step 1.3: Obtain meteorological and traffic information, and use data analysis methods to obtain environmental influencing factor data; The system collects meteorological and traffic data to aid in predicting charging demand. It obtains meteorological information, including temperature, humidity, rainfall, wind speed, and other weather conditions, by calling an open meteorological data API provided by the meteorological department. Temperature directly impacts electric vehicle battery performance and energy consumption; low temperatures reduce battery capacity and increase charging demand. Rainfall and other severe weather conditions can affect users' willingness to travel, thus influencing charging demand. The open meteorological data API uses an HTTP GET request method, with request parameters including geographical location (latitude and longitude) and time range. The API returns meteorological data in JSON format.

[0027] Traffic information is obtained by calling the traffic open data API provided by the traffic management department. This traffic data includes indicators such as road traffic flow, average speed, congestion index, and parking lot occupancy rate. Traffic flow reflects the activity level of vehicles in the area and is positively correlated with charging demand; parking lot occupancy rate reflects the number of parked vehicles, with vehicles parked for longer periods being more likely to charge. The traffic open data API also uses HTTP GET requests and returns data in JSON format.

[0028] The JSON data returned by the meteorological and traffic open data APIs is parsed, the required data fields are extracted, and converted into a unified data structure. An API call retry mechanism is established to automatically retry requests that fail, with a maximum of three retries. If the request still fails after retries, the historical average value is used to replace the currently missing data. Environmental impact factor data, including traffic and meteorological information, is obtained.

[0029] Step 1.4: Extract historical charging records from the user behavior database and use feature engineering methods to obtain user charging behavior features; User charging data is extracted from the user behavior database. This data includes user identifier, charging time, charging location, charging duration, charging amount, charging cost, vehicle model, and battery capacity. Statistical analysis is performed on the historical charging records of individual users to calculate their charging behavior characteristics. These characteristics are organized into a feature vector, comprising charging frequency, average charging duration, frequently used charging time periods, charging location preference, price sensitivity, and vehicle battery capacity. The feature vector dimension is the number of users multiplied by the number of features. Sparse matrix storage is used to save storage space. The feature vector is normalized using Z-score standardization, converting the feature values ​​to a standard normal distribution with a mean of 0 and a standard deviation of 1, thus eliminating the influence of differences in feature dimensions. The charging frequency feature is defined as the number of times a user charges within the statistical period, reflecting the intensity of the user's charging demand; the average charging duration feature is defined as the average charging duration of the user's historical charging records, reflecting the user's charging habits; the frequently used charging time period feature is obtained by clustering user charging time to identify the user's preferred charging time periods; the charging location preference feature is obtained by statistically analyzing the number of times a user charges at different charging stations to identify the user's frequently used charging locations; the price sensitivity feature is obtained by analyzing user charging behavior during different tariff periods, calculating the proportion of users choosing to charge during low-rate periods, with a higher proportion indicating higher price sensitivity; the vehicle battery capacity feature is obtained from the vehicle information database, and the battery capacity determines the demand for a single charge.

[0030] Step 1.5: Perform time alignment and quality inspection on the multidimensional feature data, and use the Laida criterion to identify outlier data to obtain the cleaned dataset; The multidimensional feature data includes real-time operation data streams of charging piles, real-time data of regional power grid load, and environmental influencing factor data. Combined with the user charging behavior features extracted in step 1.4, the multidimensional feature data is time-aligned. The timestamp of the real-time operation data stream of charging piles is selected as the base time, and data from other data sources are aligned with the base time according to the nearest time matching principle. Data with time differences within the allowable range are directly correlated, while data with time differences exceeding the allowable range are marked as missing values.

[0031] The time-aligned data undergoes quality inspection, and the Raida criterion is used to identify outliers. The Raida criterion assumes the data follows a normal distribution. For numerical features, the mean and standard deviation of the feature in historical data are calculated. If the current value deviates from the mean by more than three times the standard deviation, it is considered an outlier. For charging power data, if the power value is negative or exceeds the rated power of the charging pile, it is considered an outlier. For grid load data, if the load value exceeds the capacity limit of the regional substation, it is considered an outlier.

[0032] For meteorological information, if parameters such as temperature and humidity exceed reasonable physical ranges, they are judged as abnormal values.

[0033] The identified abnormal data is repaired using interpolation methods. For numerical features, Lagrange interpolation is used, which uses normal data points before and after the outlier to perform polynomial interpolation to estimate the reasonable value of the outlier. For categorical features, mode filling is used, which uses the historical mode of the feature to fill out the outlier values.

[0034] A data quality scoring mechanism is established to calculate a quality score for each data record. The quality score takes into account dimensions such as data integrity, data consistency, and data timeliness. Data records with a quality score below the threshold are removed and not included in subsequent analysis, resulting in a cleaned dataset.

[0035] Step 1.6: Construct time series samples based on the cleaned dataset, perform feature extraction and normalization, and obtain a standardized multidimensional feature dataset; Based on the cleaned dataset output from step 1.5, a sliding window technique is used to construct time series samples. The window length is set to 72, meaning that data from the past 72 time steps are used to predict the charging demand at future time steps, with each time step lasting 15 minutes. The sliding window slides forward with a step size of 1, generating a large number of overlapping training samples and increasing the scale of the training data. Each sample contains an input feature matrix and an output label. The input feature matrix has a dimension of 72 multiplied by the number of features, and the output label is the charging load value at a future time step.

[0036] Deep feature extraction is performed on the constructed time-series samples to extract time features, periodic features, trend features, and interaction features. Time features include hourly features, weekday features, monthly features, and holiday features; hourly features represent the 24 hours of a day using one-hot encoding; weekday features represent the 7 days of the week, with differences in charging demand patterns between weekdays and weekends; holiday features use binary identifiers to indicate whether the current date is a statutory holiday, as charging demand patterns differ from weekdays. Periodic features are extracted, as charging demand exhibits multi-scale periodicity, including daily and weekly periods, and Fourier transform is used to extract the periodic components. Trend features are extracted by calculating moving averages of historical load data to obtain the long-term trend components of the load. Interaction features are extracted by calculating the products and ratios between different features to uncover interactions between them; for example, the interaction between temperature and time period can capture the differentiated impact of temperature on charging demand at different times.

[0037] The extracted features are normalized using a minimum-maximum normalization method, scaling the feature values ​​to the range of 0 to 1. The normalization calculation is to subtract the minimum value from the feature value and then divide by the difference between the maximum and minimum values, resulting in a standardized multidimensional feature dataset. This standardized multidimensional feature dataset is then divided chronologically into training, validation, and test sets. The first 70% of the data is used as the training dataset, the middle 15% as the validation set, and the last 15% as the test set, ensuring the temporal continuity of the data partition and preventing future data leakage.

[0038] In some embodiments, due to significant differences in meteorological conditions across different regions, a uniform feature extraction method may not be adequately adapted to regional characteristics. An adaptive feature selection method can be employed to select the most relevant feature subsets for different regions, thereby improving the model's generalization ability. Specifically, a recursive feature elimination method is used for feature selection. The base model is trained using all features, and the importance score of each feature is calculated. The importance score measures the degree to which a feature affects the model's prediction error. The feature with the lowest importance score is removed, and the model is retrained using the remaining features. This process is iterated until the number of features drops to a preset threshold or the model performance no longer improves. The final retained feature set is the optimal feature subset for that region. Using the optimal feature subset for demand prediction in that region, compared to using the full feature set, improves model training speed, reduces overfitting risk, and enhances prediction generalization ability.

[0039] Step 2: Based on the standardized multidimensional feature dataset, construct and train a charging demand prediction model to predict charging demand in real time and obtain a charging load prediction curve. Step 2.1: Design a charging demand prediction model by using a multi-layer stacking method to obtain the charging demand prediction model; Charging demand prediction considers both spatial correlation and temporal dependence, constructing a charging demand prediction model that integrates Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM). This combination fully leverages the spatiotemporal characteristics of charging demand. The overall architecture of the charging demand prediction model adopts a cascaded approach. Input data is processed through convolutional layers for feature extraction. The output of these convolutional layers is used as input to the LSTM layer for time series modeling, and the prediction result is output through a fully connected layer. The convolutional layers employ one-dimensional convolution operations, with the convolutional kernels sliding along the time dimension to extract local pattern features of the time series. Three convolutional layers are used, each containing 64 convolutional kernels of sizes 3, 5, and 7. These multi-scale kernels capture local features across different time spans. Following the convolutional layers are batch normalization layers and activation layers. The batch normalization layer normalizes the convolutional outputs, accelerating model convergence, while the activation layer uses the ReLU activation function to introduce nonlinear transformation capabilities. Max pooling layers are used for downsampling between convolutional layers, with a pooling window size of 2, to reduce feature dimensionality and computational cost. LSTM layers are used to model long-term dependencies in time series. LSTMs control the flow of information through gating mechanisms, alleviating the gradient vanishing problem of traditional recurrent neural networks and capturing dependency patterns over long periods. Two LSTM layers are used, each containing 128 hidden units. The output sequence of the first LSTM layer serves as the input to the second LSTM layer. Stacking multiple LSTM layers allows for the learning of more abstract temporal features. A Dropout layer is connected after the LSTM layers, with a Dropout ratio of 0.3, randomly dropping 30% of neurons during training to prevent overfitting and improve generalization ability. A fully connected layer maps the LSTM output to the final predicted value. The fully connected layer has only one neuron and outputs a predicted charging load value at a future time.

[0040] Step 2.2: Train the charging demand prediction model by using time series cross-validation and an early stop strategy to obtain the trained charging demand prediction model. The training set from the standardized multidimensional feature dataset output in step 1.6 is input into the charging demand prediction model architecture for training, minimizing the error between the predicted and true values ​​to obtain the trained charging demand prediction model. Mean Squared Error (MSE) is chosen as the loss function. MSE calculates the sum of squared differences between the predicted and true values, imposing a higher penalty on larger errors, which helps the model converge to a more accurate solution. An adaptive learning rate optimization algorithm is used for parameter updates. This algorithm combines the advantages of momentum and adaptive learning rates, automatically adjusting the learning rate of each parameter to accelerate convergence. The initial learning rate is set to 0.001, gradually decreasing as training progresses. The training batch size is set to 64, meaning 64 samples are used per iteration to calculate the gradient and update the parameters. A smaller batch size introduces more randomness, helping to escape local optima. The maximum number of iterations is set to 200, meaning the training set is traversed 200 times. Time-series cross-validation is used to evaluate model performance. This method considers the temporal order of data, dividing the training set into multiple folds in chronological order. Earlier folds serve as training data, and later folds serve as validation data, avoiding the use of future data to predict past data. After each training round, the validation loss of the model is calculated using the validation set in the standardized multidimensional feature dataset output in step 1.6, monitoring the model's performance on unseen data. An early stopping strategy is employed to prevent overfitting. A patience parameter of 10 is set, meaning training is terminated early when the validation loss no longer decreases after 10 consecutive rounds. This prevents the model from overfitting on the training set and losing generalization ability. The model parameters with the minimum validation loss are saved as the final model, resulting in a trained charging demand prediction model.

[0041] Step 2.3: Use the test set to evaluate the prediction performance, calculate using multiple error metrics, and obtain a performance evaluation report; Based on the trained charging demand prediction model output in step 2.2, the model's predictive performance is evaluated using the test set from the standardized multidimensional feature dataset output in step 1.6. This test set data was not used during training and can objectively reflect the model's performance in real-world scenarios. The test set data is then input into the trained model to obtain the prediction results for the test set.

[0042] The model performance is quantified by calculating various error evaluation indicators, including root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAS%). RMSE calculates the square root of the mean of the squared differences between the predicted and actual values, with the same unit as the prediction target, directly reflecting the absolute magnitude of the prediction error. MAE calculates the mean of the absolute differences between the predicted and actual values, and is less sensitive to outliers compared to RMSE. MAS calculates the average percentage of the relative error between the predicted and actual values, eliminating the influence of units and facilitating horizontal comparisons between different models.

[0043] Plotting a curve comparing predicted and actual values ​​visually demonstrates the model's predictive performance, allowing observation of whether the predicted curve accurately follows the trend of the actual curve, and identifying periods where the model performs well and where there are significant deviations. Drawing a histogram of prediction error distribution analyzes the statistical characteristics of the error, verifying whether it follows a zero-mean normal distribution. A systematic shift in error indicates model bias requiring further adjustment. Calculating prediction accuracy across different prediction time spans analyzes the model's differences in short-term, medium-term, and long-term prediction capabilities. Typically, prediction uncertainty and error increase with the prediction time span. Generating a performance evaluation report includes numerical values ​​for various error indicators, prediction curves, and error distribution plots, providing a basis for decision-making regarding the model's practical deployment.

[0044] Step 2.4: Input the real-time collected multidimensional feature data into the trained model and use the rolling prediction method to obtain the charging load prediction curve; Based on the trained charging demand prediction model output from step 2.2 and the real-time data streams of charging pile operation, regional power grid load, environmental influencing factors, and user charging behavior characteristics collected in step 1, the trained prediction model is deployed to the actual system for real-time prediction. The latest multi-dimensional feature data is obtained from the data acquisition system, including charging pile operation data, power grid load data, meteorological data, traffic data, and user charging behavior characteristics from the last 72 time steps. The real-time data undergoes the same preprocessing operations as the training data, including outlier handling, missing value imputation, feature extraction, and normalization, ensuring the input data format is consistent with the training phase. The preprocessed data is input into the prediction model, which outputs the predicted charging load value for the next 15-minute time step. A rolling prediction method is used for multi-step prediction, adding the predicted value of the current time step to the historical data window as input for the next time step, and so on, recursively predicting the charging load for each future time step. After 96 rolling prediction steps, a charging load prediction curve (for the next 24 hours) is obtained, with a time granularity of 15 minutes and containing 96 prediction points.

[0045] The prediction results are post-processed, employing a moving average filter to smooth the charging load prediction curve, eliminating short-term fluctuations and making the curve smoother and more reasonable. Upper and lower limits are set for the predicted values: they cannot be lower than 0 and cannot exceed the total capacity of charging piles in the area; predicted values ​​exceeding these limits are truncated. The charging load prediction curve is stored in a database for subsequent dynamic rate optimization. Confidence intervals for the predicted values ​​are established; in addition to the point prediction values, confidence intervals are calculated. These intervals reflect the uncertainty of the prediction; a wider interval indicates greater uncertainty.

[0046] In some embodiments, since errors accumulate gradually during rolling forecasting, leading to a decrease in long-term forecast accuracy, a hybrid forecasting strategy can be adopted. This strategy combines the advantages of direct multi-step forecasting and rolling forecasting to improve long-term forecast accuracy. Specifically, two forecasting models are trained. The first model uses the aforementioned rolling forecasting method, suitable for short-term forecasting; the second model uses direct multi-step forecasting, directly outputting predicted values ​​for multiple future time steps, avoiding error accumulation, and is suitable for long-term forecasting. For short-term forecasts within the next hour, the results of the rolling forecasting model are used; for long-term forecasts from the next hour to 24 hours, the results of the direct multi-step forecasting model are used. At the boundary between the two models, a weighted average is used to fuse the forecast results. The weights are dynamically adjusted according to the forecast time distance: the closer to the current time, the greater the weight of the rolling forecasting model; the farther away, the greater the weight of the direct forecasting model. This hybrid forecasting strategy combines the advantages of both methods, ensuring high accuracy in short-term forecasting while avoiding error accumulation in long-term forecasting, thus improving overall forecast performance.

[0047] Step 3: Construct a price response model based on user charging behavior characteristics, combine it with the charging load prediction curve, and perform dynamic rate optimization to obtain a dynamic rate scheme. Step 3.1: Based on the charging load forecast curve and the grid load data, the total grid load curve is obtained by using the load superposition method; The charging load forecast curve output from step 2.4 reflects the charging demand for various time periods in the future under the current tariff conditions. The basic load forecast curve for the next 24 hours is obtained from the power grid monitoring system. Basic load refers to other electricity loads besides charging load, including industrial, commercial, and residential electricity consumption. The total load curve of the power grid is calculated using a load superposition method. For each time point, the total load equals the base load plus the charging load. The load values ​​at each time point are calculated, and a 24-hour total load curve is plotted. The peak-valley distribution of the load can be visually observed from the curve, identifying peak and off-peak periods. Statistical characteristics of the load curve are calculated, including maximum load, minimum load, average load, peak-valley difference, and load factor. The peak-valley difference is defined as the difference between the maximum and minimum load; a larger peak-valley difference indicates more severe load fluctuations. The load factor is defined as the ratio of the average load to the maximum load; a higher load factor indicates a smoother load curve and higher utilization of power grid equipment. The total load curve is compared with the capacity limits of regional substations to check for overload risks. If the total load exceeds the capacity limit during a certain period, there is a risk of power grid overload during that period, requiring tariff adjustments to suppress charging demand. The variance of the load curve is calculated; the smaller the variance, the more stable the load and the better the peak-shaving and valley-filling effect. The load variance is used as one of the objective functions for dynamic tariff optimization, with the optimization objective being to minimize the load variance and make the load curve as smooth as possible.

[0048] Step 3.2: Establish a price response model of charging rates to user demand, and use the price elasticity of demand coefficient to obtain the change in demand after rate adjustment; Based on the user charging behavior characteristics output in step 1.4, user charging demand is affected by charging rates. Demand decreases when rates rise and rises when rates fall. A price response model is established to quantify the relationship between rates and demand, described by the price elasticity of demand coefficient. Price elasticity of demand is defined as the ratio of the rate of change in demand to the rate of change in price. The price elasticity of demand coefficient is negative; a larger absolute value indicates greater price sensitivity. By analyzing historical charging records and statistically analyzing charging demand under different rate levels, a linear regression method is used to fit the relationship curve between demand and rates, estimating the price elasticity of demand coefficient. Since different user groups have different price sensitivities, users are classified, and different price elasticity of demand coefficients are estimated for different user categories, resulting in user classification results, including price-sensitive users, time-sensitive users, and convenience-priority users.

[0049] Price-sensitive users have a larger absolute value for the price elasticity of demand, while time-sensitive users have a smaller absolute value. Specifically, the absolute value of the price elasticity of demand for price-sensitive users is set between 0.8 and 1.2, indicating that for every 10% change in the rate, their charging demand changes by 8% to 12%. The absolute value of the price elasticity of demand for time-sensitive users is set between 0.2 and 0.4, indicating that for every 10% change in the rate, their charging demand changes by only 2% to 4%. The absolute value of the price elasticity of demand for convenience-first users is set between 0.4 and 0.6, falling between the first two categories. The specific value of the price elasticity of demand is personalized based on the degree of price responsiveness in users' historical charging behavior data. The degree of responsiveness is quantified by statistically analyzing the frequency of users' charging time adjustments and the magnitude of changes in charging volume during historical rate changes.

[0050] During rate optimization, the change in demand after the rate adjustment is calculated based on the rate adjustment magnitude and the demand price elasticity coefficient for the corresponding user type. For a given period, if the rate increases relative to the benchmark rate, demand decreases during that period. The decrease equals the original demand multiplied by the rate change rate and then multiplied by the demand price elasticity coefficient for that user type. If the rate decreases, demand increases, and the increase is calculated using the same method. Demand changes do not disappear or arise out of thin air, but rather shift between periods; that is, the decrease in demand during high-rate periods shifts to low-rate periods, resulting in an increase in demand during low-rate periods. A demand transfer matrix is ​​established to describe the demand transfer relationship between different periods. Matrix elements represent the proportion of demand shifted from one period to another. Demand transfer prioritizes adjacent periods, with the proportion decreasing as the transfer distance increases, reflecting the limited flexibility of users' charging time. The demand transfer matrix is ​​constructed by statistically analyzing historical data on users' actual charging time changes after rate adjustments, calculating the proportion of demand shifted from each period to other periods, and forming a transfer probability matrix. A sum of 1 in each row of the matrix indicates that the demand for that period has completely shifted to all periods. Based on the price response model and demand shift matrix, the charging demand for each time period after the tariff scheme is adjusted is calculated. This demand is used as the input for load calculation to evaluate the effectiveness of the tariff scheme.

[0051] Step 3.3: Construct a multi-objective optimization function by using a weighted summation of grid load variance and total user cost to obtain a single-objective optimization function; Based on the total grid load curve output in step 3.1 and the demand change after rate adjustment output in step 3.2, dynamic rate optimization needs to consider multiple optimization objectives simultaneously, mainly including two core objectives: peak shaving and valley filling of grid load and reduction of user charging costs. The peak shaving and valley filling objective is achieved by minimizing the variance of the total grid load curve. A smaller load variance indicates a smoother load curve, reducing peak load and increasing valley load, thus achieving peak shaving and valley filling. The load variance is calculated by taking 96 time points over the next 24 hours, calculating the sum of squared deviations between the load value at each time point and the average load, and then dividing by the number of time points to obtain the load variance value. The user charging cost minimization objective is measured by calculating the total charging cost for all users. The total cost equals the sum of the charging amount multiplied by the corresponding rate for each time period. Rate optimization should not sacrifice user interests; it should reduce user costs while achieving grid objectives. Since the two objectives have different dimensions and numerical ranges, normalization is required. Each objective value is divided by its baseline value to obtain a relative objective value, eliminating the influence of dimensions. A weighted summation method is used to transform multiple objectives into a single objective. The single-objective optimization function is equal to the weighted sum of each normalized objective value multiplied by its corresponding weight coefficient. The weight coefficients reflect the importance of each objective and are set according to actual operational needs. If peak shaving and valley filling of the power grid are given greater importance, the load variance objective has a larger weight; if user benefits are given greater importance, the cost objective has a larger weight. The sum of the weight coefficients is 1 to ensure the standardization of the objective function. In practical applications, the weight coefficients can be flexibly adjusted according to the operational priorities of different periods to achieve dynamic adjustment of operational strategies. The single-objective optimization function is used as the objective of the optimization algorithm to find the rate scheme that minimizes the objective function value.

[0052] Step 3.4: Set rate optimization constraints, using charging pile capacity constraints and power grid security constraints to obtain the feasible solution space; Based on the single-objective optimization function output in step 3.3, rate optimization needs to meet a series of constraints to ensure the feasibility and rationality of the optimization results. These constraints include: 1. Charging pile capacity constraints: The charging load in each time period cannot exceed the total capacity of the charging piles in the area. The total capacity of the charging piles equals the number of charging piles multiplied by the rated power of each pile. If the optimized charging demand in a certain time period exceeds the capacity limit, the rate scheme is not feasible and needs adjustment. 2. Grid safety constraints: The total grid load in each time period cannot exceed the capacity limit of the substation in the area. When the total load approaches the capacity limit, it triggers the risk of grid overload, and charging demand must be suppressed by increasing the rate. 3. Rate fluctuation range constraints: The adjustment range of the rate relative to the benchmark rate cannot exceed the set upper and lower limits to avoid excessive rate fluctuations that could cause user dissatisfaction and regulatory problems. Typically, the rate increase should not exceed 50% of the benchmark rate, and the rate decrease should not exceed 30% of the benchmark rate to ensure that the rate fluctuates within a reasonable range. 4. Rate smoothness constraints: The rate change between adjacent time periods should not be too large to avoid abrupt rate changes. The rate change rate between adjacent time periods is limited to within 20% to ensure a smooth transition of the rate curve. To satisfy the demand constraint, the total charging demand after the rate adjustment should be roughly equivalent to the original total demand, ensuring that users' charging needs are met; the demand is simply redistributed over time, rather than the total demand decreasing. To ensure normal charging service charges, a non-negative rate constraint should be set, requiring the rate value to be greater than 0, preventing negative or zero rates. These constraints are formalized as mathematical inequalities or equations, forming the constraint set of the optimization problem. This constraint set defines the feasible solution space for the rate scheme; only rate schemes that satisfy all constraints are feasible solutions.

[0053] Step 3.5: Solve the rate optimization model using the particle swarm optimization algorithm, and obtain the dynamic rate scheme by adopting an iterative optimization strategy; Based on the feasible solution space output in step 3.4, the rate optimization problem is a multivariate nonlinear constrained optimization problem, which is difficult to solve using traditional gradient optimization methods. Therefore, a heuristic intelligent optimization algorithm is adopted. The Particle Swarm Optimization (PSO) algorithm is selected. This algorithm simulates the foraging behavior of a flock of birds, finding the optimal solution in the solution space through information sharing and cooperative search among particles. It has the advantages of simple implementation and fast convergence. The rate values ​​for 96 time periods in the next 24 hours are used as optimization variables, with each rate value representing an optimization dimension, resulting in a 96-dimensional optimization problem. The particle swarm is initialized with 50 particles, each representing a candidate rate scheme. The particle's position vector represents the rate value for each time period. Initial particle positions are randomly generated within the feasible solution space to ensure that the initial rate schemes satisfy the constraints. A velocity vector is randomly initialized for each particle, representing the particle's movement direction and step size in the solution space. Algorithm parameters are set, including the maximum number of iterations, inertia weight, individual learning factor, and social learning factor. The iterative optimization process is as follows: For each particle, based on the rate scheme represented by its position vector, the charging demand for each time period is calculated using the price response model, thereby calculating the total grid load and total user cost. The objective function value is calculated, and the constraint conditions are checked for satisfaction. If the constraints are not satisfied, a penalty term is applied. The optimal position found by each particle so far is recorded, i.e., the individual optimal solution, and the optimal position found by the entire particle swarm so far is recorded, i.e., the global optimal solution. Based on the particle's current position, individual optimal position, global optimal position, and algorithm parameters, the particle's new velocity and new position are calculated. The velocity update calculation includes three parts: inertia term, individual cognition term, and social cognition term. The inertia term keeps the particle maintaining its original motion trend, the individual cognition term moves the particle closer to its historical optimal position, and the social cognition term moves the particle closer to the global optimal position. The positions and velocities of all particles are updated, and the next iteration begins. A termination condition is set: the algorithm terminates when the maximum number of iterations is reached or the global optimal solution no longer improves after multiple consecutive iterations. The rate scheme corresponding to the global optimal solution is output, which is the optimized dynamic rate scheme.

[0054] In some embodiments, since particle swarm optimization (PSO) is prone to getting trapped in local optima, leading to unstable optimization results, an adaptive PSO algorithm can be employed. The aim is to improve global search capability and optimization stability by dynamically adjusting algorithm parameters. Specifically, an adaptive inertia weight strategy is used. The inertia weight is set to a large value in the early stages of iteration, giving particles strong global exploration capabilities. In the later stages of iteration, the inertia weight is gradually reduced, causing particles to concentrate on fine-grained searches in local regions. The inertia weight decreases linearly from 0.9 to 0.4. A mutation operator is introduced. When a decrease in particle swarm diversity is detected, i.e., particle positions tend to concentrate, random perturbations are applied to some particles, causing them to escape local optima and re-explore the solution space. The mutation probability is set to 0.1. A multi-swarm collaborative optimization strategy is adopted, dividing the particle swarm into multiple sub-swarms. Each sub-swarm evolves independently, and at regular intervals, the sub-swarms exchange optimal solution information. This multi-swarm strategy increases solution diversity and reduces the risk of getting trapped in local optima. Compared to the basic PSO algorithm, the adaptive algorithm provides higher optimization result quality and better stability, making it suitable for complex rate optimization problems.

[0055] Step 4: Based on user charging behavior characteristics, conduct user cluster analysis, design personalized incentive strategies, and obtain differentiated incentive schemes for different user types. Step 4.1: Extract user feature vectors and use the K-means clustering algorithm to obtain user group classification results; Based on the user charging behavior characteristics output in step 1.4, and combined with the need for user price sensitivity analysis in step 3.2, key user features are extracted from the user charging behavior feature vector for user clustering analysis. Key user features include dimensions such as charging frequency, average charging duration, commonly used charging time periods, price sensitivity, charging location preference, and vehicle type. The extracted key feature vectors are standardized to eliminate the influence of different feature dimensions and ensure that each feature has a relatively equal weight in the clustering process.

[0056] The K-means clustering algorithm is used to classify users. The K-means algorithm iteratively optimizes the algorithm to minimize intra-cluster distances and maximize inter-cluster distances, dividing users into K distinct groups. A suitable number of clusters K is selected, and the elbow rule is used to determine the optimal K value. The sum of squares within each cluster is calculated for different K values, and a curve showing the relationship between K and the sum of squares within each cluster is plotted. The inflection point on the curve corresponds to the optimal K value. Based on charging demand characteristics and price response traits, users are divided into three categories, resulting in user group classification results. The first category is price-sensitive users, who charge frequently, prefer charging during low-rate periods, respond positively to price changes, and have a large absolute value for their price elasticity coefficient, corresponding to the range of 0.8 to 1.2 set in step 3.2. The second category is time-sensitive users, whose charging time is fixed, mostly for commuting, with low charging time flexibility, are not price-sensitive, and focus more on charging convenience and speed, resulting in a small absolute value for their price elasticity coefficient, corresponding to the range of 0.2 to 0.4 set in step 3.2. The third category is convenience-priority users, who charge less frequently, prefer charging stations that are nearby and have high availability, are not sensitive to either price or time, and have a moderate absolute value for their price elasticity coefficient, corresponding to the range of 0.4 to 0.6 set in step 3.2. An interpretive analysis is performed on the user classification results, statistically analyzing the size percentage, typical characteristics, and charging behavior patterns of each user category to form user profiles. The user classification results are stored in the user database to provide a basis for the user type in the price response model in step 3.2, and to provide a foundation for subsequent personalized services.

[0057] Step 4.2: Design a points reward mechanism for price-sensitive users, using points accumulation and redemption rules to obtain a points incentive scheme; Based on the user group classification results output in step 4.1, price-sensitive users pay high attention to charging costs, and their charging behavior can be effectively guided through economic incentives. A points reward mechanism is designed, where users earn points for charging during low-rate periods, with the number of points linked to the rate discount—the lower the rate, the more points earned. A points accumulation mechanism is set up, where points earned from each charging session are added to the user's account and are valid indefinitely, encouraging continuous participation in off-peak charging. A points redemption mechanism is also set up, where users can redeem charging coupons, free charging credits, or other value-added services after accumulating a certain number of points, with transparent redemption rules. A points tier system is established, dividing users into different levels such as regular members, silver members, gold members, and platinum members based on their accumulated points, with different levels enjoying different point multipliers and exclusive benefits; the higher the level, the faster the points are accumulated, incentivizing continued user participation. The points calculation rule is that the points earned for a single charge are equal to the amount of electricity charged multiplied by the points coefficient. The points coefficient is proportional to the discount rate of the current period rate relative to the base rate. The discount rate is calculated by subtracting the current rate from the base rate and then dividing by the base rate. The higher the discount rate, the larger the points coefficient.

[0058] The mobile application displays users' points balance, points acquisition history, and points redemption options in real time, enhancing the visualization and immediacy of incentives; regular points doubling events are held, where users can earn double points by charging at specific times or stations, further strengthening the incentive effect.

[0059] Step 4.3: Design a priority service mechanism for time-sensitive users, adopting a strategy of exchanging higher rates for priority access to obtain a priority service plan; Based on the user group classification results output in step 4.1, time-sensitive users have low charging time flexibility, pay more attention to the timeliness and convenience of charging, and are willing to pay higher fees for fast charging. A priority service mechanism is designed, allowing users to choose to pay a priority rate higher than the standard rate to obtain charging priority. When charging station resources are scarce, users paying the priority rate will be allocated charging stations first, without having to wait in line. The priority rate is set at 1.2 to 1.5 times the standard rate, and users can choose whether to use the priority service based on the urgency of their charging needs. A priority charging queue is established, with the system maintaining two queues: a regular charging queue and a priority charging queue. Users in the priority queue are allocated charging stations first, while users in the regular queue can only obtain a charging station when the priority queue is empty. A priority reservation mechanism is also implemented, allowing users to reserve charging stations for specific time slots in advance. Reservations require a reservation fee, and the system locks the charging station for the user upon successful reservation. Users can start charging immediately upon arrival at their reserved time slot, avoiding waiting. The priority service mechanism satisfies the immediate charging needs of time-sensitive users while regulating charging demand during peak hours through a higher rate, achieving a balance between supply and demand. The mobile application displays the current waiting time and priority service options for charging stations, allowing users to choose flexibly according to their needs.

[0060] Step 4.4: Design an intelligent recommendation mechanism for convenience-first users, using a comprehensive scoring and ranking of charging piles to obtain personalized recommendation schemes; Based on the user group classification results output in step 4.1, convenience-first users prefer charging stations that are close, have high vacancy rates, and offer good charging environments, and are not sensitive to rates. An intelligent recommendation mechanism is designed to recommend the most suitable charging station based on the user's current location, charging needs, and the real-time status of the charging station. A comprehensive charging station scoring model is established, with scoring dimensions including distance cost, waiting time, rate level, charging power, and environmental evaluation. Distance cost is calculated based on the straight-line distance between the user's location and the charging station; the closer the distance, the higher the score. Waiting time is estimated based on the number of vehicles currently queuing at the charging station; the shorter the queue, the higher the score. Rate level is calculated based on the current rate; the lower the rate, the higher the score, but this dimension has a lower weight for convenience-first users. Charging power is calculated based on the charging station's rated power; the higher the power, the faster the charging speed, and the higher the score. Environmental evaluation is calculated based on user ratings and facility conditions at the charging station; the better the evaluation, the higher the score. The scores for each dimension are weighted and summed, with weights set according to user preferences; convenience-first users have higher weights set for distance and waiting time. The system calculates a comprehensive score for each candidate charging station, sorts them from highest to lowest score, and recommends the top three charging stations to the user. Detailed information about the recommended charging stations is displayed via a mobile application, including location, navigation route, current availability, estimated wait time, and pricing information. Users can navigate to the recommended charging stations with a single click. This intelligent recommendation system reduces user search costs and improves charging convenience and user satisfaction.

[0061] In some embodiments, since different users' preferences and needs change dynamically over time, static user classification may not accurately reflect the user's current state. A dynamic user profile update mechanism can be adopted to adjust user classification and incentive strategies in real time based on the user's latest charging behavior, thereby improving the accuracy of personalized services. Specifically, a user behavior tracking module is established to record information such as the time, location, rate, and response behavior of each user's charging in real time. A sliding window mechanism is used to calculate the user's behavioral characteristics over the past month and compare them with the user's historical characteristics to detect whether the characteristics have changed. When the change in the user's charging behavior characteristics exceeds a set threshold, the user profile update process is triggered to recalculate the user's cluster affiliation and determine whether the user has moved from one category to another. Based on the updated user profile, the incentive plan pushed to the user is dynamically adjusted to ensure that the incentive strategy matches the user's current needs. The dynamic profile mechanism enables the system to adapt to the evolution of user behavior and improves the sustainability of the incentive effect.

[0062] Step 5: Based on the dynamic rate scheme and differentiated incentive schemes for different user types, set the rate information release rules, dynamically schedule charging resources, and obtain the rate release scheme and resource scheduling results; Step 5.1: Design a rate information publishing rule engine, adopt trigger conditions and publishing strategies, and obtain an intelligent rate push solution; Based on the dynamic rate scheme output in step 3.5, a rate information release rule engine is established. The rule engine automatically executes rate information release tasks according to preset trigger conditions, including setting rate preview release rules, setting rate locking rules, setting emergency rate adjustment rules, and setting rate recovery rules.

[0063] Specifically, the system has several rules: First, it automatically sends a rate adjustment notification to users two hours before the adjustment takes effect. This notification includes the adjusted rate, the effective date, a brief explanation of the reason for the adjustment, and suggested charging times. This gives users ample time to adjust their charging plans, increasing user acceptance. Second, it locks the rate at the moment a user plugs in to start charging. This rate remains unchanged throughout the charging process, even if the rate changes, preventing user dissatisfaction caused by rate increases. Third, it triggers an emergency rate increase mechanism when the grid load exceeds a safe threshold. Within five minutes, the rate is raised to the highest level, and a load warning is sent to all users, indicating excessive grid pressure and suggesting a temporary halt to charging. This emergency rate quickly suppresses charging demand through price signals, ensuring grid safety. Finally, it gradually reduces the rate as the grid load returns to a safe range, using a step-by-step recovery rather than a sudden drop to avoid a secondary impact caused by a surge in charging demand.

[0064] Establish a multi-channel rate release mechanism. Rate information is simultaneously released through various channels such as charging pile displays, mobile application push notifications, WeChat official account messages, and SMS reminders to ensure information reach. Design a visual interface for rate information, using a line graph to display the rate change curve for the next 24 hours, and using red, yellow, and green to indicate high, medium, and low rate periods, so that users can quickly identify the optimal charging time and obtain an intelligent rate push solution.

[0065] Step 5.2: Establish a user rate subscription and reminder service, and use personalized threshold settings to obtain a rate publication plan; Based on the intelligent rate push solution output in step 5.1, to reduce the cost for users to actively query rates, a rate subscription and reminder service is established to obtain a precise push solution. Users can set personalized rate reminder conditions in the mobile application, and the system will automatically push a notification when the rate meets the conditions. These reminder conditions include rate threshold reminders, where users set a rate upper limit; when the rate is lower than the threshold during a certain period, the system pushes a notification to remind users that the rate is favorable for charging during that period; specific time period reminders, where users can subscribe to rate information for specific time periods, such as subscribing to rates at 8 am and 6 pm daily, and the system pushes rate information before these times; and charging station availability reminders, where users can follow specific charging stations, and when the charging station changes from busy to available and the rate is low, the system pushes a notification to remind users to go there to charge.

[0066] The subscription service transforms rate information delivery from a broadcast model to a targeted push model, allowing users to receive only the information they care about, avoiding information overload and improving user experience. A push frequency control mechanism is established to limit the number of push notifications a single user can receive per day, preventing frequent pushes from causing user annoyance. Push notifications are concise and clear, highlighting core information. Users can click on the notification to view detailed information and perform subsequent actions, such as navigating to a charging station or booking a charging service. This step outputs a rate publishing solution, including an intelligent rate push solution and a targeted push solution.

[0067] Step 5.3: Construct an intelligent scheduling model for charging resources and use a multi-attribute decision-making method to obtain the optimal matching relationship between users and charging piles; Based on the differentiated incentive schemes for different user types output in step 4, an intelligent scheduling model for charging resources is established to achieve optimal matching between user charging needs and charging pile resources.

[0068] When a user initiates a charging request through the mobile application, the system obtains information such as the user's current location, remaining vehicle battery power, expected charging time, and required charging power. The system also queries the real-time status of all charging stations in the area, including station location, availability, current tariff, charging power, and number of vehicles in the queue.

[0069] For each candidate charging station, a comprehensive evaluation index for user use of the charging station is calculated. The comprehensive evaluation index includes three dimensions: distance, time, and cost. The distance index is calculated based on the actual road distance between the user's location and the charging station, with a higher index value for greater distances. The time index includes the travel time to the charging station and the charging queue waiting time, estimated based on current traffic conditions and charging station queue status. The cost index is calculated based on the expected charging capacity and the current charging rate.

[0070] A multi-attribute decision-making method is employed to comprehensively evaluate candidate charging stations, establishing a decision matrix where rows represent candidate charging stations, columns represent evaluation indicators, and matrix elements represent the scores of each charging station on each indicator. The decision matrix is ​​normalized to eliminate the influence of different indicator dimensions. Weights for each indicator are set according to user preferences, with different weights assigned to different user types: price-sensitive users have higher weights for cost indicators, while time-sensitive users have higher weights for time indicators.

[0071] The specific calculation process for multi-attribute decision-making is as follows: convert each indicator value into a standardized score between 0 and 1. For cost-related indicators, a reciprocal transformation is used, that is, the standardized score equals 1 minus the ratio of the indicator value to the maximum value. For benefit-related indicators, a ratio transformation is directly used, that is, the standardized score equals the ratio of the indicator value to the maximum value. Calculate the weighted comprehensive score of each charging pile. The score is equal to the sum of the products of the standardized scores of each indicator and their corresponding weights. The charging pile with the highest score is the optimal matching charging pile for the user.

[0072] The system recommends the best charging station to the user, displays detailed information about the charging station and navigation route. After the user confirms, the system reserves the charging station for the user or adds the user to the charging station's queue, thus obtaining the optimal matching relationship between the user and the charging station.

[0073] Step 5.4: Establish a charging reservation and resource locking mechanism, and use a time window allocation method to obtain the resource scheduling results; Based on the optimal matching relationship between users and charging piles output in step 5.3, a charging reservation mechanism is established to improve the certainty of user charging. This mechanism achieves precise scheduling of charging resources through time window allocation and resource locking.

[0074] Users select "Reserve Charging" in the mobile application, entering information such as desired charging time, charging duration, and charging location. The system queries the occupancy status of charging stations in the target area during the reservation period, identifying those with available capacity. For charging stations with available capacity, the system assigns a reservation time window to the user, including the start and end times. Users must arrive and begin charging within the time window; reservations are automatically cancelled if the user misses the window. After a successful reservation, the system locks the charging station's capacity for the corresponding time period, preventing other users from reserving that time slot.

[0075] A dynamic pricing strategy is adopted, with lower fees for reservations during off-peak hours and higher fees during peak hours, using price leverage to incentivize users to reserve charging stations during off-peak periods. A reservation priority mechanism is established, giving paid reservation users priority over those queuing on-site, protecting the rights of reservation users. A reservation cancellation and modification mechanism is also in place, allowing users to cancel or modify their reservations free of charge within a certain period before the reservation takes effect; cancellations after this period incur a penalty, thus regulating user behavior and preventing malicious reservations that hoard resources.

[0076] The system monitors the reservation execution status in real time. When a user fails to arrive within the reserved time window, the system automatically releases the reserved resources and reallocates them to other users, improving resource utilization efficiency. Through the combination of the reservation mechanism and real-time scheduling, the system achieves optimized configuration of charging resources in the time dimension and intelligent allocation in the spatial dimension, forming resource scheduling results, including user charging pile matching schemes, time window allocation schemes, resource locking status, and dynamic adjustment strategies. This provides users with deterministic charging services while maximizing the utilization rate of charging piles.

[0077] In some embodiments, the fixed-time-window reservation mechanism lacks flexibility, and users' charging time may be delayed due to factors such as traffic, leading to reservation failure. A flexible reservation mechanism can be adopted to improve the flexibility of the reservation system and the user's fault tolerance. Specifically, the concept of a flexible time window is introduced. When making a reservation, users can choose a standard time window or a flexible time window. The standard time window is 30 minutes, requiring users to arrive within 30 minutes, with a lower reservation fee; the flexible time window is 60 minutes, allowing users to arrive at any time within 60 minutes, with a slightly higher reservation fee. The system dynamically adjusts the availability of the flexible window based on the charging pile's load. More flexible reservations are allowed when the load is low, while the number of flexible reservations is limited when the load is high. A reservation extension mechanism is established, allowing users to apply for an extension before the reservation time window expires. The system determines whether an extension is possible based on the subsequent reservation status of the charging pile. If there are no subsequent reservations, the extension is automatically approved; if there are subsequent reservations, the user is prompted to expedite their arrival or cancel the reservation. The flexible reservation mechanism improves user experience and reservation success rate while ensuring resource utilization efficiency.

[0078] Step 6: Based on the rate release plan and resource scheduling results, collect operational data and user response data in real time, and combine them with the charging load prediction curve to optimize the charging demand prediction model and price response model, so as to obtain optimized charging demand prediction parameters and corrected price response parameters. Step 6.1: Collect operational data and user response data in real time, and use data aggregation and storage methods to obtain an operational data warehouse; Based on the rate publication scheme and resource scheduling results output in step 5, a system operation data collection mechanism is established to collect various types of data generated during the operation of the dynamic rate system in real time, including the actual charging power of each charging pile in each time period, the actual number of charging vehicles, the distribution of charging duration, and the utilization rate of charging piles. Actual load data of the regional power grid is collected to compare the load curve changes before and after the rate adjustment and to evaluate the peak shaving and valley filling effect. User charging behavior data is collected, including user charging time selection, response to rate adjustments, reservation usage, and participation in incentive mechanisms. User feedback data is collected through the satisfaction evaluation function built into the mobile application to gather user evaluations and suggestions on the rate scheme, incentive mechanisms, and charging services. System operation logs are collected to record information such as rate adjustment events, abnormal events, and system failures, providing a basis for problem diagnosis.

[0079] A data aggregation platform is established, where data collected from various data sources is uploaded to the platform via a unified interface. The platform performs preprocessing on the data, including format conversion, quality checks, and time alignment. A data warehouse is built, employing a distributed storage architecture to store massive amounts of historical data, supporting rapid querying and analysis. The data warehouse organizes data according to subject domains, including charging operation, power grid load, user behavior, and system performance, facilitating multi-dimensional analysis. A data security mechanism is established, encrypting sensitive data, setting data access permissions, ensuring data security and protecting user privacy, resulting in a functional data warehouse.

[0080] Step 6.2: Calculate the system operation effect evaluation indicators, and use comparative analysis methods to obtain a dynamic rate implementation effect evaluation report; Based on the operational data warehouse output in step 6.1, quantitative indicators of system operation performance are calculated to evaluate the implementation effect of the dynamic rate scheme. The quantitative indicators of system operation performance include grid load peak shaving and valley filling indicators, charging pile utilization rate indicators, user charging cost indicators, user service quality indicators, and system operation revenue indicators.

[0081] Specifically, the system calculates peak-shaving and valley-filling indicators for the power grid, including the rate of change of peak-valley load difference, the rate of change of load curve variance, the load decrease during peak hours, and the load increase during off-peak hours. By comparing the load curves before and after the implementation of dynamic pricing, the improvement of each indicator is calculated; a larger improvement indicates a better peak-shaving and valley-filling effect. The system also calculates charging pile utilization indicators, including average utilization rate, utilization rate standard deviation, peak-hour utilization rate, and off-peak-hour utilization rate. Increased utilization rate indicates improved efficiency in charging resource allocation, while a decrease in the utilization rate standard deviation indicates a more balanced load distribution among different charging piles. Furthermore, the system calculates user charging cost indicators, including average user charging cost, cost savings, and cost savings percentage, comparing changes in user charging costs before and after the implementation of dynamic pricing to assess the economic benefits gained by users. Finally, the system calculates user service quality indicators, including average waiting time, reservation success rate, charging success rate, and user satisfaction rating; shorter waiting times and increased satisfaction indicate improved service quality. Finally, the system calculates system operating revenue indicators, including total charging service revenue, charging volume, and revenue per unit of electricity, to assess the impact of dynamic pricing on operating revenue.

[0082] Using a comparative analysis method, a period before the implementation of dynamic rates was selected as the baseline period, and the same duration after implementation was selected as the comparison period. By controlling for other variables, the net effect of dynamic rates was isolated, and a dynamic rate implementation effect evaluation report was generated. The report includes numerical values ​​of various indicators, trend charts, effect analysis conclusions, etc., providing decision-makers with intuitive evaluation basis.

[0083] Step 6.3: Analyze the distribution of charging demand prediction errors, and use error source tracing and model correction methods to obtain optimized charging demand prediction parameters; Based on the operational data warehouse output in step 6.1 and the charging load prediction curve output in step 2.4, the prediction accuracy of the charging demand prediction model is continuously monitored. Predicted and actual values ​​are collected, and the prediction error is calculated. The prediction error is defined as the predicted value minus the actual value. A positive error indicates that the prediction is too high, and a negative error indicates that the prediction is too low.

[0084] The distribution characteristics of the statistical prediction error are analyzed, and statistical measures such as the error mean, standard deviation, maximum error, and error quantiles are calculated. An error distribution histogram is plotted to check whether the error follows a zero-mean normal distribution. If the error mean deviates from 0, it indicates that there is a systematic bias in the model, which needs to be corrected. The differences in prediction errors under different time periods, different charging piles, and different weather conditions are analyzed to identify scenarios with large prediction errors and to trace the source of the errors.

[0085] Error source tracing methods include feature importance analysis, which calculates the contribution of each input feature to the prediction error, identifies the features that have the greatest impact on the error, and improves feature engineering in a targeted manner; error source tracing also includes sample analysis, which identifies the samples with the largest prediction errors, analyzes the characteristics of these samples, and identifies the weak links of the model.

[0086] Based on the error analysis results, the charging demand prediction model is calibrated and optimized, including bias correction, incremental learning, and ensemble learning. Specifically, bias correction refers to adding a bias correction term to the prediction output to eliminate systematic errors if the model has systematic biases; incremental learning involves adding newly collected data to the training set to incrementally train the model, update the model parameters, and adapt the model to the latest data distribution; ensemble learning involves training multiple prediction models and weightedly fusing the prediction results of multiple models to improve the overall prediction accuracy and robustness.

[0087] Regularly evaluate the performance of the optimized model. If the performance improves, deploy the new charging demand prediction model to the production environment to replace the old charging demand prediction model, thereby achieving continuous iterative optimization of the model and obtaining optimized charging demand prediction parameters.

[0088] Step 6.4: Evaluate the actual response rate of users to the rate adjustment, and use the response elasticity calculation method to obtain the corrected price response parameters; Based on the operational data warehouse output in step 6.1 and the price response model output in step 3.2, during the implementation of dynamic pricing, the actual observed user response behavior may deviate from the pre-assumed price response model, requiring adjustments to the model parameters based on actual data. The actual changes in user charging demand under different pricing adjustments are statistically analyzed to calculate the actual demand price elasticity. The actual elasticity is compared with the assumed elasticity in the model; a significant difference indicates that the model parameters need adjustment. The actual response rates for different user types are analyzed; price-sensitive users should have higher actual response rates, while time-sensitive users should have lower rates. If the actual situation differs from expectations, it indicates inaccurate user classification or elasticity parameter settings. Parameter estimation methods are used to re-estimate the price response model parameters based on actual observation data. Least squares or maximum likelihood estimation methods are used to find the parameter values ​​that best fit the model's predicted values ​​to the actual values. The actual effect of the incentive mechanism is analyzed, statistically analyzing the proportion of users participating in points rewards, priority services, and intelligent recommendations, and calculating the marginal impact of the incentive mechanism on user behavior. If the incentive effect is not significant, the incentive intensity or method needs to be adjusted. Establish user feedback collection channels and conduct surveys and interviews to gain a deeper understanding of users' subjective feelings about rate adjustments and incentive mechanisms, obtaining qualitative feedback information to supplement the deficiencies of quantitative data analysis. By combining the results of quantitative and qualitative analysis, revise the parameters of the price response model and incentive mechanism to make the model more accurately reflect user behavior patterns, improve the effectiveness of rate optimization, and obtain the revised price response parameters.

[0089] In some embodiments, since user response behavior is influenced by a combination of factors, a simple price elasticity model may not be sufficient to fully characterize the complex user decision-making process. A machine learning-based user response prediction model can be employed to learn complex patterns in user responses through a data-driven approach, thereby improving response prediction accuracy. Specifically, a user response classification model is constructed, using whether a user changes their charging behavior under a specific rate adjustment as the classification label. The rate adjustment magnitude, user characteristics, charging demand characteristics, and external environmental characteristics are used as input features. Machine learning algorithms such as random forests or gradient boosting trees are used to train the classification model. The model can learn the nonlinear relationships and interactions between features, capturing complex patterns that are difficult to characterize using traditional elasticity models. The trained classification model can predict the response probability of a specific user under a specific rate adjustment. This response probability can be used for demand prediction during rate optimization, improving optimization accuracy. Compared to traditional elasticity models, machine learning models have stronger fitting and generalization capabilities, but require sufficient training data. As system data accumulates, model performance continuously improves.

[0090] Step 7: Based on the optimized charging demand forecast parameters and the corrected price response parameters, perform data encryption and privacy protection to obtain tamper-proof transaction credentials; Step 7.1: Design a user data encryption storage scheme, using a combination of symmetric and asymmetric encryption to obtain a secure data storage mechanism; Based on the optimized model parameters and strategies output from step 6, the charging system involves a large amount of user personal information and charging behavior data, making data security and privacy protection crucial. A data encryption storage mechanism is established to encrypt and store sensitive data to prevent leakage. Sensitive data includes user identity information, vehicle information, charging records, location information, payment information, etc. A symmetric encryption algorithm is used to encrypt the large amount of data. Symmetric encryption algorithms offer fast encryption and decryption speeds, suitable for large data volume scenarios. The AES encryption algorithm is selected, with a key length of 256 bits, providing high-strength encryption protection. The symmetric encryption key itself needs secure storage and transmission; therefore, an asymmetric encryption algorithm is used to encrypt the symmetric key. The RSA algorithm is selected, with a key length of 2048 bits. Asymmetric encryption uses a public key for encryption and a private key for decryption. The private key is only stored on the authorized server to ensure key security. A key management system is established to manage the entire lifecycle of keys, including generation, storage, distribution, updating, and destruction. Encryption keys are changed regularly, and immediately replaced when a security threat is detected or a key may be leaked, reducing security risks. Encrypt the data transmission process using the TLS protocol to protect data security during network transmission and prevent eavesdropping or tampering. Establish a data anonymization mechanism to anonymize sensitive information during data analysis and presentation; for example, replace user names with anonymous identifiers and partially hide license plate numbers to protect user privacy.

[0091] Step 7.2: Establish a user authentication and access control mechanism, adopt multi-factor authentication and role-based access control, and obtain an access security control scheme; Based on the data security storage mechanism output in step 7.1, a strict user authentication mechanism is established to ensure that only legitimate users can access the system. A multi-factor authentication method is adopted, requiring users to provide both a password and a dynamic verification code upon login. The dynamic verification code is generated via SMS or mobile application, has a validity period of 5 minutes, and expires after one use. Multi-factor authentication enhances account security and prevents account theft due to password leakage. An access control mechanism is established, employing a role-based access control model, dividing system users into different roles such as ordinary users, operations personnel, and system administrators, each with different access permissions. Ordinary users can only access their own charging records and personal information, and cannot access other users' data. Operations personnel can access system operation data and statistical analysis results, but cannot access users' sensitive personal information. System administrators have the highest privileges, responsible for system configuration and maintenance, but their operations are subject to strict auditing. An access log recording mechanism is established to record all user access behavior, including access time, access content, and operation type, and the access logs are reviewed regularly to identify abnormal access behavior. Establish an abnormal access alarm mechanism. When abnormal behavior such as multiple failed login attempts, logins from different locations, or access to sensitive data outside of working hours is detected, the system will automatically trigger an alarm, notify the security administrator to investigate, and freeze the account if necessary.

[0092] Step 7.3: Use blockchain technology to construct a trusted record of charging transactions, and use distributed ledgers and smart contracts to obtain tamper-proof transaction certificates; Based on the access security control scheme output in step 7.2, a charging transaction record system is constructed using blockchain technology to enhance the credibility and traceability of charging transactions. Charging transactions involve multiple parties, including charging piles, users, operating platforms, and power grid companies. Traditional centralized recording methods suffer from single-point-of-failure risks and trust issues. Blockchain employs distributed ledger technology, distributing transaction records across multiple nodes. Any data tampering on any single node will be detected by other nodes, ensuring data immutability. After each charging transaction is completed, a transaction record is generated, containing information such as user identifier, charging pile identifier, charging start time, charging end time, charging amount, rate information, and charging cost. The transaction record is hashed to generate a digital fingerprint. The hash values ​​of multiple transactions form a block, which is connected through a hash chain to form the blockchain. Nodes in the blockchain network verify new blocks using a consensus algorithm. Only blocks approved by a majority of nodes are added to the blockchain, ensuring data consistency. Smart contracts are used to automatically execute charging transaction settlements. These contracts predefine settlement rules, and once charging is complete and the transaction record is uploaded to the blockchain, the smart contract automatically calculates the fee and triggers the payment process, eliminating the need for manual intervention and improving settlement efficiency and transparency. Blockchain records can also serve as trusted evidence in resolving transaction disputes. When disputes arise between users and operators, they can obtain immutable transaction evidence by querying the blockchain records, protecting the rights of all parties.

[0093] A charging station charging fee optimization system based on dynamic rates, such as Figure 3 As shown, the method for optimizing charging station fees based on dynamic rates, as described above, includes: The data acquisition module is used to collect multidimensional feature data and user charging data, perform preprocessing and feature extraction, and obtain standardized multidimensional feature datasets and user charging behavior features. The demand forecasting module constructs and trains a charging demand forecasting model based on a standardized multidimensional feature dataset, forecasts charging demand in real time, and obtains a charging load forecast curve. The rate optimization module constructs a price response model based on user charging behavior characteristics, combines it with the charging load prediction curve, and performs dynamic rate optimization to obtain a dynamic rate scheme. The user analysis module performs user clustering analysis based on user charging behavior characteristics, designs personalized incentive strategies, and obtains differentiated incentive schemes for different user types. The resource scheduling module sets rules for publishing rate information based on dynamic rate schemes and differentiated incentive schemes for different user types, dynamically schedules charging resources, and obtains rate publishing schemes and resource scheduling results. The optimization module, based on the rate release scheme and resource scheduling results, collects operational data and user response data in real time, and combines them with the charging load prediction curve to optimize the charging demand prediction model and the price response model, thereby obtaining optimized charging demand prediction parameters and corrected price response parameters. The security protection module, based on optimized charging demand forecasting parameters and corrected price response parameters, performs data encryption and privacy protection to obtain tamper-proof transaction credentials.

[0094] In one embodiment of the present invention, a specific example is provided: This invention has been practically applied in a charging network within a city's business district. The business district comprises three charging stations with a total of 120 charging piles, including 60 fast-charging piles and 60 slow-charging piles, serving approximately 5,000 electric vehicles within the area. This invention focuses on the application area of ​​uneven spatial and temporal distribution of charging demand in urban business districts. It optimizes the charging load distribution through a dynamic pricing mechanism, alleviating grid pressure, reducing user charging costs, and improving the quality of charging services.

[0095] After system deployment, operational data was collected over a period of three months. The following table presents an example of the acquired data.

[0096] Table 1 shows an example of charging pile operation data for a certain weekday: Table 1: Example of charging pile operation data on a certain weekday;

[0097] Table 2 shows examples of incentive participation data for different user types: Table 2: Example of incentive participation data for different user types;

[0098] By comparing data before and after the implementation of the dynamic pricing rate, the system's operation performance is good. Regarding grid load, the peak load during peak hours decreased from 20.8 MW to 19.2 MW, while the load during off-peak hours increased from 11.5 MW to 13.1 MW. The peak-to-valley load difference was significantly reduced, the load curve became smoother, and the peak-shaving and valley-filling effects were evident. Regarding charging pile utilization, the average daily utilization rate increased from 0.52 to 0.64, with improved utilization during off-peak hours, indicating a certain improvement in charging resource allocation efficiency. Regarding user charging costs, the average monthly charging cost for users participating in off-peak charging decreased, and overall user satisfaction remained at a high level. Regarding system operation, although the rate was reduced, the overall operating revenue remained stable and slightly increased due to the increase in total charging volume. The practical application results fully verify the effectiveness and practicality of the method of this invention, providing a feasible technical solution for charging pile operation and management.

[0099] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

Claims

1. A method for optimizing charging fee of a dynamic rate-based charging pile, characterized in that, The method comprises the following steps: Collecting multi-dimensional feature data and user charging data, preprocessing and feature extraction, obtaining a standardized multi-dimensional feature data set and user charging behavior characteristics; Based on the standardized multi-dimensional feature data set, a charging demand prediction model is constructed and trained to predict the charging demand in real time, and a charging load prediction curve is obtained; Based on the user charging behavior characteristics, a price response model is constructed, combined with the charging load prediction curve, dynamic rate optimization is carried out, and a dynamic rate scheme is obtained; Based on the user charging behavior characteristics, user clustering analysis is carried out, individualized incentive strategies are designed, and different user types are obtained. Differentiated incentive schemes; Based on the dynamic rate scheme and the differentiated incentive scheme of different user types, the rate information release rule is set, the charging resources are dynamically scheduled, and the rate release scheme and resource scheduling result are obtained; Based on the rate release scheme and resource scheduling result, real-time collection of operation data and user response data is carried out, combined with the charging load prediction curve, the charging demand prediction model and the price response model are optimized, and the optimized charging demand prediction parameters and the corrected price response parameters are obtained; Based on the optimized charging demand prediction parameters and the corrected price response parameters, data encryption and privacy protection are carried out, and tamper-proof transaction credentials are obtained. 2.The dynamic rate-based charging station charging fee optimization method of claim 1, wherein, The collection of multi-dimensional feature data and user charging data, preprocessing and feature extraction comprises: Real-time reading of charging pile operation parameters by the Internet of Things data acquisition module and transmission, obtaining charging pile real-time operation data stream; Through the data proxy server, the regional power grid monitoring system is connected to obtain real-time data of regional power grid load; Call meteorological open data API and traffic open data API to obtain environmental influence factor data; Extract historical charging records from the user behavior database and calculate user charging behavior characteristics; Time alignment of multi-dimensional feature data and user charging behavior characteristics, multi-dimensional feature data including charging pile real-time operation data stream, regional power grid load real-time data, environmental influence factor data; Identify abnormal data and repair abnormal values to obtain cleaned data set; Based on the cleaned data set, time series samples are constructed, time characteristics, periodic characteristics and trend characteristics are extracted and scaled to a preset interval to obtain a standardized multi-dimensional feature data set, and the standardized multi-dimensional feature data set is divided into training set, validation set and test set according to time sequence.

3. The charging pile charging fee optimization method based on dynamic rate according to claim 2, characterized in that, The extraction of historical charging records from the user behavior database comprises: Extracting user charging data from the user behavior database, including user identification, charging time, charging location, charging duration, charging power, charging cost, vehicle model and battery capacity; Statistical analysis of historical charging records of a single user to calculate user charging behavior characteristics; Normalization processing of user charging behavior characteristics. 4.The method of claim 1, wherein, The construction of the charging demand prediction model and the training, the real-time prediction of the charging demand comprises: Constructing a charging demand prediction model combining convolutional neural network and long short-term memory network; Input the training set in the standardized multi-dimensional feature data set into the charging demand prediction model for training to obtain the trained charging demand prediction model; The performance of the trained charging demand prediction model is evaluated using a test set of standardized multi-dimensional feature data, and a performance evaluation report is generated; Real-time multi-dimensional feature data is collected and input into the trained charging demand prediction model, which uses a rolling prediction method for multi-step prediction and smooths the prediction results to obtain a charging load prediction curve. 5.The dynamic rate-based charging station charging fee optimization method of claim 1, wherein, The dynamic rate optimization includes: Calculating load curve statistical features, including maximum load, minimum load, load peak-valley difference, and load rate, and taking load variance as the optimization objective function; Establishing a price response model to quantify the relationship between rate and demand, estimating different demand price elasticity coefficients for different user groups, and establishing a demand transfer matrix to describe the demand transfer relationship between time periods; Constructing a multi-objective optimization function that considers both grid load peak clipping and valley filling and user charging cost reduction, and using a weighted summation method to convert the multi-objective into a single objective; Setting rate optimization constraints to form a constraint set for the optimization problem and define the feasible solution space; Solving the multi-variable nonlinear constraint optimization problem using a particle swarm optimization algorithm, initializing the particle swarm to randomly generate initial particle positions within the feasible solution space, calculating the new speed and position of the particles, and obtaining the dynamic rate scheme. 6.The method of claim 5, wherein, The estimation of different demand price elasticity coefficients for different user groups includes: Classifying users to obtain user classification results, including price-sensitive users, time-sensitive users, and convenience-priority users; setting the absolute values of the demand price elasticity coefficients for price-sensitive users, time-sensitive users, and convenience-priority users; According to the rate adjustment amplitude and the demand price elasticity coefficient of the corresponding user type, calculate the demand change after rate adjustment; Establishing a demand transfer matrix to describe the demand transfer relationship between time periods.

7. The dynamic rate-based charging station toll optimization method of claim 1, wherein, The design of individualized incentive strategies includes: Designing an integral reward mechanism for price-sensitive users, with the number of points linked to the rate discount amplitude, setting an integral accumulation mechanism and a redemption mechanism, and establishing an integral rating system; Designing a preferential service mechanism for time-sensitive users, allowing users to pay a preferential rate to obtain charging priority, establishing a preferential charging queue and reservation mechanism, and balancing supply and demand through rate adjustment; Designing an intelligent recommendation mechanism for convenience-priority users, establishing a charging pile comprehensive scoring model, and recommending by score ranking after weighted summation of each dimension score. 8.The dynamic rate-based charging station charging fee optimization method of claim 1, wherein, The setting of rate information release rules and dynamic scheduling of charging resources includes: Establishing a rate information release rule engine, including setting rate advance release rules, setting rate lock rules, setting emergency rate adjustment rules, and setting rate recovery rules; Establishing a rate subscription and reminder service, allowing users to set personalized rate reminder conditions and push reminder services, and establishing a push frequency control mechanism; Building a charging resource intelligent scheduling model to match user charging demand with charging pile resources and obtain the optimal matching relationship between users and charging piles; establishing a decision matrix, normalizing the decision matrix, and calculating the weighted comprehensive score to determine the optimal matching charging pile; Establishing a charging reservation and resource locking mechanism to achieve precise scheduling of charging resources through time window allocation and resource locking, and obtaining the resource scheduling result. 9.The dynamic rate-based charging station charging fee optimization method of claim 1, wherein, The data encryption and privacy protection includes: A data encryption storage mechanism is established to encrypt and store user identity information, vehicle information, charging records, location information, and payment information; A user identity authentication and access control mechanism is established for multi-factor authentication and role permission management; A blockchain technology is used to build charging transaction records, which are distributed and stored. After a charging transaction is completed, a transaction record is generated and hashed, and a smart contract is used to automatically execute transaction settlement to obtain a tamper-proof transaction voucher.

10. A dynamic rate-based charging pile charging optimization system, characterized in that, A dynamic rate-based charging pile charging optimization method according to any one of claims 1-9, comprising: A data acquisition module for acquiring multi-dimensional feature data and user charging data, preprocessing and feature extraction, obtaining standardized multi-dimensional feature data sets and user charging behavior characteristics; A demand prediction module based on the standardized multi-dimensional feature data set, a charging demand prediction model is constructed and trained to predict real-time charging demand and obtain a charging load prediction curve; A rate optimization module based on user charging behavior characteristics to construct a price response model, combine the charging load prediction curve, and optimize the dynamic rate to obtain a dynamic rate scheme; A user analysis module based on user charging behavior characteristics for user clustering analysis and design of personalized incentive strategies to obtain differentiated incentive schemes for different user types; A resource scheduling module based on the dynamic rate scheme and the differentiated incentive scheme for different user types to set rate information release rules, dynamically schedule charging resources, and obtain rate release schemes and resource scheduling results; An optimization module based on the rate release scheme and resource scheduling results to collect real-time operation data and user response data, combine the charging load prediction curve, and optimize the charging demand prediction model and the price response model to obtain optimized charging demand prediction parameters and corrected price response parameters; A security protection module based on the optimized charging demand prediction parameters and the corrected price response parameters for data encryption and privacy protection to obtain a tamper-proof transaction voucher.

Images