An electric vehicle busy charging station prediction point table data verification method and system

Abstract

The application provides a method and system for predicting point table data of a busy charging station of an electric vehicle, and relates to the technical field of charging station data verification. The method comprises the following steps: obtaining real-time charging data of a first preset time length and historical charging data of a second preset time length before the current time of each charging station, constructing a training sample by using the historical charging data, training a preset prediction model by using the training sample, inputting the real-time charging data into the trained model to obtain predicted charging data, filling a date matrix by using the real-time charging data and the predicted charging data, filling an information table by using information of each charging station, establishing an association between the date matrix and the information table to generate point table data, verifying the point table data according to a preset rule, and generating a charging equipment resource planning strategy in combination with the verified data and the historical charging data. The application improves the traceability or reliability of charging data in the charging station of the electric vehicle.

Description

Technical Field

[0001] This application relates to the technical field of charging station data verification, and in particular to a method and system for verifying data from a predicted list of busy charging stations for electric vehicles. Background Technology

[0002] With the rapid development of the electric vehicle industry, the charging network, as the core infrastructure supporting the operation of electric vehicles, directly affects the efficiency of transportation and the stability of energy supply through its planning and operation management. As a crucial node at the end of the power system, the operating status, service capacity, and temporal changes in the busyness of charging stations constitute key dynamic data for power system operation and management. Accurately understanding the busy status of charging stations and predicting their future load trends is of significant practical importance for optimizing the charging network layout, rationally allocating power resources, effectively avoiding long queues for users, and improving overall operational efficiency.

[0003] Chinese invention patent CN121119348A, published on December 12, 2025, discloses a method for analyzing the usage frequency of charging piles based on charging frequency data. This method obtains raw data on the number of charging times of multiple charging piles at different times and dates, constructs a charging behavior dataset with time dimension annotation, introduces a sliding window algorithm and periodic spectrum analysis method to extract frequency features, constructs a multi-scale charging pile usage frequency feature matrix, uses density peak clustering algorithm to classify the usage frequency matrix, identifies the usage preferences and activity characteristics of charging piles in various typical time scenarios, and finally predicts the usage of charging piles in future time periods and generates a frequency utilization report based on a lightweight temporal neural network combined with regression prediction and anomaly detection algorithms.

[0004] However, this method uses density peak clustering to classify charging piles, and its output is a statistical conclusion at the group level. Individual differences among different charging piles within the same category in key time dimensions such as the start and end times and duration of peak hours are compressed or lost. This makes it impossible for this method to generate accurate predictions of the busy status of individual charging stations down to the specific time point, and operators find it difficult to obtain refined information on whether a specific charging station is busy during a specific time period. Summary of the Invention

[0005] To improve the traceability and reliability of charging data in electric vehicle charging stations, this application provides a method and system for verifying data from a predicted list of busy charging stations for electric vehicles.

[0006] Firstly, this application provides a method for verifying data from a predicted list of busy charging stations for electric vehicles, employing the following technical solution: A method for verifying data from a predicted list of busy charging stations for electric vehicles includes the following steps: Real-time charging data and historical charging data for a first preset duration before the current time of each charging station are obtained respectively, and training samples are constructed based on the historical charging data of each charging station. The preset prediction model is trained using training samples to obtain the trained prediction model. All real-time charging data is input into the trained prediction model, and the predicted charging data is output. The date matrix is ​​filled with all real-time charging data and predicted charging data, and the information table is filled with information from each charging station. The relationship between the information table and the date matrix is ​​established to obtain the point table data of each charging station. The charging station data is verified according to preset rules to obtain verified charging station data. Based on the verified charging station data and historical charging data, a charging equipment resource planning strategy is generated.

[0007] This application simultaneously acquires real-time charging data for a first preset duration and historical charging data for a second preset duration, enabling it to capture both short-term real-time trends of charging stations and uncover long-term historical charging patterns. Training samples are constructed using historical charging data to train the prediction model, and then real-time charging data is input into the trained model to output predicted charging data, thus improving the accuracy and robustness of predictions for busy charging stations.

[0008] This application populates all real-time charging data and predicted charging data into a date matrix, and populates the information of each charging station into an information table. Then, it establishes the relationship between the information table and the date matrix, and finally forms the point table data of each charging station. This can structurally integrate the scattered charging data and static information of charging stations, forming a unified, standardized, and traceable point table data system, which is convenient for subsequent data verification and analysis.

[0009] This application verifies the charging station data according to preset rules to obtain verified charging station data. This step can automatically identify and correct outliers, contradictory data, and missing data in the charging station data, effectively filtering data noise caused by fluctuations in real-time charging data or prediction deviations, making the charging station data used for decision-making highly accurate and reliable. Based on the verified charging station data and historical charging data, this application generates a charging equipment resource planning strategy. Because the charging station data has undergone intelligent extrapolation and rule verification processing by the prediction model, the generated resource planning strategy can accurately match the future charging demand of each charging station, realize the rational allocation of charging equipment (such as the number of charging piles, power configuration, etc.), and minimize resource idleness or supply shortages, thereby improving the operational efficiency of busy charging stations and the user charging experience.

[0010] Optionally, after obtaining the verified point table data, the following may also be included: According to the preset busy status rules, busy periods are filtered from the verified point table data, and status tags are added to each charging station according to the busy periods. The status tags include: busy tags and non-busy tags. Status warning signals are generated and sent based on the status labels of each charging station.

[0011] This application filters the peak hours for each charging station based on preset peak hour rules. Since the point-of-charge data has undergone dual quality assurance through predictive model extrapolation and rule verification, the data itself possesses high accuracy and reliability. Therefore, the peak hours identified based on this data have higher credibility and a lower false positive rate. Compared to directly determining peak hours based on raw real-time charging data, the scheme adopted in this application effectively avoids the problem of false peak hour identification caused by data noise and outliers, further improving the reliability of the peak hour identification results.

[0012] Optionally, it also includes: Based on the verified point table data, generate the status change trend of each charging station. Based on the status change trend, identify candidate charging stations whose current status label is non-busy, but whose status change trend points to a busy state. Obtain the estimated time from the i-th busy charging station to each candidate charging station, and calculate the transition probability of each candidate charging station changing from a non-busy state to a busy state after the estimated time based on the state change trend, and send information about candidate charging stations whose transition probability is less than a preset probability threshold.

[0013] This application generates the status change trend of each charging station based on the verified point table data. Instead of relying solely on the current busy / non-busy label for static judgment, it further explores the status evolution direction of each charging station in the future, and can capture its busy trend before the charging station enters a busy state.

[0014] This application calculates the transition probability of each candidate charging station from a non-busy state to a busy state after an estimated time based on the state change trend, and uses a preset probability threshold as the screening criterion to send only the information of candidate charging stations with a transition probability less than the threshold, thereby reducing information redundancy.

[0015] Optionally, the transition probability is calculated as follows: Based on the actual busy transition frequency under similar state change trends in the training samples, a baseline probability is obtained. The baseline probability is then dynamically corrected using the time difference between the current moment and the predicted busy moment, as well as the rate of change of the state change trend. The corrected baseline probability is then used as the transition probability.

[0016] This application first analyzes training samples to extract the actual frequency of busy transitions under similar state change trends, and calculates an initial baseline probability based on this. Building upon this, the application further introduces a time dimension and trend change characteristics, using the time difference between the current moment and the predicted busy moment, as well as the rate of change of the state change trend, to adjust the baseline probability in real time, thereby obtaining a more accurate transition probability.

[0017] Optionally, after adding status tags to each charging station according to peak hours, the following may also be included: Based on the information of the charging station, the charging station is divided into high-speed stations and non-high-speed stations, and based on the status label, the charging station is divided into busy stations and non-busy stations. Determine whether the i-th busy depot is a highway depot. If so, search for non-busy depots along the legal driving direction of the route where the i-th busy depot is located, including the road segment and adjacent road segments. Obtain the search results and select the non-busy depot closest to the i-th busy depot as the recommended diversion depot. If not, obtain the non-busy depots within the area to which the i-th busy depot belongs, and denote them as target depots. Calculate the road network distance between the i-th busy depot and each target depot, and select the target depot with the smallest road network distance as the recommended diversion depot.

[0018] This solution first classifies charging stations based on their geographical attributes (highway / non-highway) and operational status (busy / not busy), constructing a refined station selection mechanism. For the i-th busy station, if it is located on a highway, considering the irreversibility and continuity of vehicle travel, this application strictly searches for non-busy stations along the legally permitted travel direction within its location and adjacent road segments, and establishes the closest one as the recommended diversion station to minimize ineffective detours; if it is located on a non-highway, it locks the non-busy target stations within its area, and selects the minimum value by calculating the road network distance as the recommended diversion station, thereby truly reflecting the urban road traffic cost.

[0019] This application achieves precise matching and efficient recommendation of alternative resources by differentiating the diversion logic under different scenarios, which significantly reduces users' waiting time for recharging and effectively improves users' travel experience and satisfaction.

[0020] Optionally, it also includes: traversing all busy stations and generating a recommendation mapping table, wherein the recommendation mapping table stores information about recommended diversion stations and the distance from each busy station to the recommended diversion station.

[0021] By adopting the above technical solution, this application can achieve unified management of the distributed data and reduce real-time computing overhead.

[0022] Optionally, when the i-th busy depot is a non-high-speed depot, it also includes: The busy stations are matched with the target stations in terms of time series. Based on the matching results, station pairs that are complementary in terms of busy and non-busy periods are obtained and recorded as time complementary candidate pairs. Add the target stations that are time-complementary candidate pairs with the i-th busy station to the recommendation mapping table.

[0023] This application, through time-series matching, overcomes the limitations of simple spatial distance by utilizing the differences in tidal characteristics among different charging stations to precisely guide peak-hour charging loads to idle stations during off-peak hours. This application not only achieves the redistribution and peak-shaving of charging loads over time, effectively alleviating grid pressure in local areas, but also maximizes the utilization rate of idle resources, thereby improving the overall operational efficiency of the charging network while ensuring users' charging needs are met.

[0024] Optionally, the historical charging data within the second preset duration includes charging data from the same period within the first preset duration, or charging data from a portion of the first preset duration.

[0025] This application uses forced temporal overlap, meaning that the training samples include historical segments that are completely aligned with the current time, to make the patterns learned by the model highly consistent with the actual operating patterns at the current time. This reduces the distribution shift problem caused by temporal misalignment and improves prediction accuracy and stability.

[0026] Secondly, this application provides a data verification system for predicting busy charging station locations for electric vehicles, employing the following technical solution: A system for verifying data from a predicted list of busy charging stations for electric vehicles includes: a processor and a memory communicatively connected to the processor; The memory is provided with a computer-readable storage medium, and a computer program is stored on the computer-readable storage medium. When the processor executes a computer program stored on the computer-readable storage medium, it implements the method as described in the first aspect.

[0027] In summary, this application includes at least one of the following beneficial technical effects: 1. This application simultaneously acquires real-time charging data for a first preset duration and historical charging data for a second preset duration, enabling it to capture both the current short-term real-time trends of charging stations and uncover long-term historical charging patterns. Training samples are constructed using historical charging data to train the prediction model, and then real-time charging data is input into the trained model to output predicted charging data, thus improving the accuracy and robustness of predictions for busy charging stations.

[0028] 2. This application fills all real-time charging data and predicted charging data into a date matrix, and fills the information of each charging station into an information table. Then, it establishes the relationship between the information table and the date matrix, and finally forms the point table data of each charging station. This can structurally integrate the scattered charging data and static information of charging stations, forming a unified, standardized and traceable point table data system, which is convenient for subsequent data verification and analysis.

[0029] 3. This application verifies the charging station data according to preset rules to obtain verified charging station data. This step can automatically identify and correct outliers, contradictory data, and missing data in the charging station data, effectively filtering data noise caused by fluctuations in real-time charging data or prediction deviations, making the charging station data used for decision-making highly accurate and reliable. Based on the verified charging station data and historical charging data, this application generates a charging equipment resource planning strategy. Because the charging station data has undergone intelligent extrapolation and rule verification processing by the prediction model, the generated resource planning strategy can accurately match the future charging needs of each charging station, realize the rational allocation of charging equipment (such as the number of charging piles, power configuration, etc.), and minimize resource idleness or supply shortages, thereby improving the operational efficiency of busy charging stations and the user charging experience. Attached Figure Description

[0030] Figure 1 This is a flowchart of Embodiment 1 of this application; Figure 2 This is a flowchart of Embodiment 2 of this application. Detailed Implementation

[0031] The following combination Figure 1 and Figure 2 This application will be further described in detail below. To facilitate understanding of the technical solutions of this application, the following defines and explains specific terms appearing in various embodiments of this application: Single-gun order count: refers to the number of charging transactions successfully completed by a single charging pile (single gun) within a unit of time. In this solution, both real-time charging data and historical charging data of the charging station use the single-gun order count as the basic data unit.

[0032] The first preset duration refers to a continuous time window preceding the current moment, used for collecting real-time charging data. The first preset duration can be dynamically set according to the type of charging station. In this application, 1 hour is preferred for high-frequency charging stations, and 24 hours is preferred for low-frequency charging stations.

[0033] The second preset duration refers to the historical data backtracking time span used to construct training samples. In this application, the second preset duration is preferably 7 days, 30 days, or 90 days to cover at least one complete operating cycle.

[0034] Date Matrix: A two-dimensional data structure with time as the row index and the number of orders per charging station as the fill data. It is used to uniformly store the real-time order number and predicted order number of each charging station at different time points. The corresponding date matrix can be retrieved by the number of each charging station.

[0035] Information table: A relational data table with the charging station number as the primary key, used to store the static attribute information of each charging station.

[0036] Point-to-point data: refers to a structured dataset obtained by associating and fusing dynamic status data in a date matrix with static attribute data in an information table.

[0037] Charging station utilization rate: This is calculated by combining the number of single-gun orders with the total number of charging stations in the charging station, and is used to characterize the busyness of the charging station.

[0038] Busy threshold: The critical value used to determine whether a charging station is in a busy state. In this solution, multiple thresholds are set according to the usage rate of charging piles.

[0039] Status tag: A discretized identifier of the charging station's busy status at a specific point in time, with a value of "busy tag" (denoted as 1) or "non-busy tag" (denoted as 0).

[0040] State change trend: describes the direction and rate of change of the charging station's busy state over time, obtained by calculating the first derivative of the charging pile utilization rate.

[0041] Candidate charging station: refers to a charging station that is currently not busy, but whose status change trend is positive and exceeds a preset threshold.

[0042] Busy / Unbusy: Classification of charging stations based on the current status label.

[0043] High-speed charging stations / non-high-speed charging stations: Classification based on the geographical location of the charging station: Charging stations located within highway service areas are called high-speed charging stations, otherwise they are called non-high-speed charging stations.

[0044] Recommended alternative charging stations: For a busy charging station, alternative charging stations are selected according to preset rules.

[0045] Road network distance: refers to the actual distance traveled from the starting point to the destination along a feasible path in a real road network.

[0046] Recommendation Mapping Table: A data structure that stores the correspondence between all busy stations and their recommended diversion stations. The recommendation mapping table also stores target stations that are time-complementary candidate pairs with the i-th busy station.

[0047] Time-complementary candidate pairs: refers to a pair of charging stations that have a complementary relationship in the time distribution of busy and non-busy periods.

[0048] Example 1: This example discloses a method for verifying data from a predicted list of busy charging stations for electric vehicles. (Refer to...) Figure 1 The method includes: S11 data acquisition and processing, S12 constructing training samples and model training, S13 generating point table data, and S14 verification and generation of charging equipment resource planning strategy. First, real-time charging data for a first preset duration and historical charging data for a second preset duration prior to the current time for each charging station are acquired. Training samples are constructed using the historical charging data to train a preset prediction model. Then, real-time charging data is input into the trained model to obtain predicted charging data. Next, the date matrix is ​​filled with real-time charging data and predicted charging data, and an information table is filled with information about each charging station. A relationship between the two is established to generate point table data. Finally, the point table data is verified according to preset rules, and the verified data is combined with historical charging data to generate a charging equipment resource planning strategy. The execution process of each step in this embodiment is as follows: S11 Data Acquisition and Processing: This involves acquiring real-time charging data for a first preset time period prior to the current moment and historical charging data for a second preset time period for each charging station. Both the real-time and historical charging data use the number of single-gun orders for each charging pile in the charging station as the core data indicator. The number of single-gun orders refers to the number of charging transactions successfully completed by a single charging pile within a preset time period, and is a fundamental indicator for measuring the busyness of a charging station. For example, if a charging station has 10 charging piles, and within a 15-minute time granularity, if each charging pile completes one charging transaction, the total number of single-gun orders for that charging station during that time period is 10; if only 5 charging piles each complete one transaction, the total number of orders is 5.

[0049] The first preset duration is preferably set to 1 hour, 2 hours, or 24 hours, and can be adjusted according to the actual operating characteristics and predicted demand of the charging station. For example, for high-frequency charging stations in urban centers, the first preset duration can be set to 1 hour to obtain sufficiently fine-grained real-time status information; for low-frequency charging stations in suburban areas or highway service areas, the first preset duration can be set to 24 hours to smooth out random fluctuations within a single day.

[0050] The second preset duration is preferably set to 7 days, 30 days, or 90 days to cover at least one complete operating cycle. Preferably, the historical charging data package within the second preset duration contains charging data from the same time period within the first preset duration, or charging data from specific time periods within the first preset duration. For example, if the first preset duration is one hour before the current time, the historical charging data within the second preset duration should include charging data from the same time period (i.e., the same time window each day) for each of the past 7 days, as well as charging data for 30 minutes before and after that time period each day, to enhance the temporal correlation of the training samples.

[0051] The real-time charging data includes the number of single-gun orders for each charging station within a preset time (e.g., 15 minutes).

[0052] Historical charging data includes the number of single-gun orders for each charging station in the past multiple time periods (intervals of preset time granularity, such as every 15 minutes).

[0053] The construction of training samples in this embodiment includes: S111 data cleaning and standardization, S112 time window slicing, S113 feature extraction and label annotation, and S114 adding labels.

[0054] S111 data cleaning and standardization involves handling missing values ​​and detecting outliers in the acquired historical charging data. For missing values, linear interpolation or forward imputation is used to complete them. For outliers (such as the number of orders per charging station exceeding the physical limit of the charging station within a unit of time, for example, the number of orders per charging station cannot exceed 1 within 15 minutes), the 3σ principle is used to identify and remove them, or the average of adjacent time points is used for replacement.

[0055] The S112 time window slicing method uses a sliding window approach to slice historical charging data. It sets the window width (e.g., 6 time points, each 15 minutes apart) and the sliding step size (e.g., 1 time point), generating multiple window data subsets along the time axis. Each window data subset contains a sequence of single-gun order counts at consecutive time points.

[0056] S113 Feature Extraction and Labeling: Multidimensional features are extracted from each window's data subset and used as input to the LightGBM regression model. These multidimensional features include temporal features, periodic encoded features, and lag features, which are then integrated into an input vector.

[0057] The time-series features include: the mean, maximum, minimum, and standard deviation of the number of single-gun orders in the current window; the first-order difference feature of the number of single-gun orders in the current window; the total number of orders in the current window; and the fluctuation range of the number of orders in the current window (the difference between the maximum and minimum values).

[0058] Periodic encoding characteristics: The current time is within the hour of the day (0-23). ​​Sine-cosine encoding is used to process the current time to minimize periodic jumps in the value.

[0059]

[0060] in, Pi To represent a complete circular angle, this embodiment uses sine-cosine encoding to map the 24-hour cycle to... Within the range; The current hour; The value is the hour of the current time after sine coding. The value obtained by cosine encoding the hour of the current moment; It is a trigonometric sine function; It is a trigonometric cosine function.

[0061]

[0062]

[0063] in, Pi To represent a complete circular angle, this embodiment uses sine-cosine encoding to map the weekdays from Monday to Sunday. Within the range; The week number within a week at the current time; This is the value after sinusoidal encoding at the current moment; The value after cosine encoding at the current moment; It is a trigonometric sine function; It is a trigonometric cosine function.

[0064] Workday indicator: 1 indicates a workday, 0 indicates a weekend / holiday; Holiday indicators: 1 indicates a holiday, 0 indicates a non-holiday; Lagging characteristics: charging pile utilization rates at 1, 2, 3, and 6 time points in the past (i.e., lagging by 1, 2, 3, and 6 steps); number of single-gun orders at the same time in the past 24 hours (daily lag characteristic); number of single-gun orders at the same time in the past 168 hours (weekly lag characteristic).

[0065] S114 adds labels, using the number of single-gun orders at the next time point of each window's data subset as the prediction target label, forming the training samples required for supervised learning. Each training sample is represented as (input vector, label value), where the label value is the number of single-gun orders at the next time point.

[0066] S12 constructs training samples and trains the model, using the training samples to train the preset prediction model.

[0067] In this embodiment, the prediction model adopts the LightGBM regression model. LightGBM (Light Gradient Boosting Machine) is an efficient ensemble learning algorithm based on gradient boosting decision trees. Its core advantage lies in the use of one-sided gradient sampling technology and mutually exclusive feature binding technology, which significantly improves training speed while ensuring prediction accuracy.

[0068] In this embodiment, the parameter configuration of the LightGBM regression model is shown in the table below.

[0069] ; In other embodiments, the parameter configuration of the LightGBM regression model can be set according to requirements, and the prediction model can also adopt deep learning models such as LSTM and BI-LSTM.

[0070] The training process of the prediction model includes S121 dataset partitioning, S122 model training, S123 feature importance evaluation, S124 prediction, and S125 post-processing.

[0071] The S121 dataset was partitioned, with the training samples divided into a training set, a validation set, and a test set in a 7:1.5:1.5 ratio. The training set was used for model parameter learning, the validation set for hyperparameter tuning and early stopping detection, and the test set for model performance evaluation.

[0072] For the S122 model training, this embodiment uses a five-fold cross-validation method. Specifically, the training data is randomly divided into five non-overlapping subsets. Four subsets are taken as training data and one subset is taken as validation data. This process is repeated five times, and the average of the five validation results is taken as the model performance evaluation index.

[0073] S123 Feature Importance Evaluation: After training, the LightGBM model can output the importance ranking of each feature.

[0074] Feature importance is measured by calculating the number of times each feature is used for splits across all decision trees (or the total gain from splits). Based on feature importance analysis, redundant features with importance below a preset threshold (e.g., 0.01) are removed, simplifying the model and improving inference speed.

[0075] S124 prediction: After training, acquire all real-time charging data (i.e., the number of single-gun orders at each charging station within the first preset time period before the current time), and process the real-time single-gun order data within the first preset time period before the current time according to the same feature engineering process as the training samples (including sliding window slicing, feature extraction, periodic encoding, etc.) to form a prediction input vector with the same feature dimensions as the training samples. Input the prediction input vector into the trained LightGBM regression model.

[0076] After inputting the prediction vector into the model, the model outputs the predicted number of single-gun orders for a future period. To achieve multi-step prediction (e.g., predicting the next 4 hours, one point every 15 minutes, for a total of 16 time points), this embodiment adopts a rolling prediction strategy: Step 1 Prediction: Use the current window data to predict the number of single-gun orders at the next time point; Step 2 prediction: Using the predicted value from step 1 as part of the input, slide the window to predict the next time point; This process continues until the number of single-gun orders at all target time points is predicted.

[0077] The predicted charging data includes: the predicted number of single-gun orders at various future time points.

[0078] S125 post-processing smooths the prediction results. In this embodiment, the exponential moving average method is used for smoothing, and the calculation formula is as follows:

[0079] in, The smoothed predicted value at time t; The smoothing coefficient is 0.3 in this embodiment. The predicted value at time t; This is the smoothed predicted value at time t-1.

[0080] S13 generates point table data, filling the date matrix with all real-time single-charger order count data and predicted single-charger order count data, filling the information table with information from each charging station, and establishing the association between the information table and the date matrix to obtain the point table data for each charging station. This step includes S131 constructing the date matrix, S132 constructing the information table, and S133 establishing the association.

[0081] S131 constructs a date matrix, which is constructed as follows: Suppose there are M charging stations, the time discretization granularity is Δt (preferably 15 minutes), and the prediction time window length is T (e.g., 96 time points in the next 24 hours). Then the size of the date matrix is ​​(T+N)×M, where N is the number of time points corresponding to the real-time data acquired within the first preset time period.

[0082] In this embodiment, the real-time single-gun order count of each charging station within a first preset time period is filled into the corresponding position of the date matrix. Then, the predicted single-gun order count of each charging station within the prediction time window, as predicted by the LightGBM regression model, is filled into the corresponding position of the date matrix.

[0083] S132 constructs an information table, and the information for each charging station includes, but is not limited to: Charging station identification: charging station number, name, and area; Location: Road section, highway (for charging stations in highway service areas), and accessible distance; Equipment parameters: number of charging piles, number of charging guns.

[0084] S133 establishes a relationship by using the primary key charging station number to establish a relationship between the information table and the date matrix. In this embodiment, the complete point table data is obtained by associating the charging station number in the information table with the corresponding date matrix.

[0085] S14 verifies and generates a charging equipment resource planning strategy. The point table data is verified according to preset rules to obtain verified point table data. The preset rules include data consistency verification rules and logical rationality verification rules. This step includes: S141 data consistency verification, S142 logical rationality verification, and S143 generating a resource planning strategy.

[0086] S141 Data Consistency Verification: Traverse each data point in the point table to check for any anomalies, including: the rate of change in charging pile utilization between adjacent time points exceeds a preset threshold (e.g., 30%), and the change is unexplained; contradictory data of high utilization and low order count appearing simultaneously at the same charging station during the same time period; and the predicted order count and the real-time order count jumping at the same time point (e.g., the real-time order count is 4, and the predicted order count is 9).

[0087] The formula for calculating the utilization rate of charging piles is as follows: Charging pile utilization rate = Predicted number of orders per charging gun ÷ (Total number of charging piles × Maximum service capacity coefficient of a single charging pile within the time window) × 100% Among them, the maximum service capacity coefficient of a single pile refers to the maximum number of orders that a single gun can complete within a unit of time granularity.

[0088] In this step, the following method is used to handle data points with abnormal conditions (i.e., abnormal data points): Based on the spatiotemporal characteristics of abnormal data points, this embodiment classifies abnormal situations into three categories: sensor failure, data loss, and real-world emergencies.

[0089] Sensor failures refer to situations where the number of orders per gun is 0 or the physical upper limit (e.g., a single gun can complete a maximum of 1 transaction within 15 minutes) for at least 3 consecutive time points, and there are no synchronization anomalies in the data of charging stations within the adjacent preset distance range. Data loss refers to the absence of data from 2 to 5 consecutive time points, and the rate of change of data before and after the missing period is lower than a preset rate of change threshold (e.g., 30%). A true sudden event refers to a situation where at least two adjacent charging stations experience a simultaneous increase in usage, and the residual between the trend and the output of the prediction model is lower than a preset residual threshold (e.g., 5 instances).

[0090] For sensor fault-related anomalies, this embodiment takes the abnormal site as the center, calculates the reciprocal of the road network distance of each charging station in its spatial neighborhood, uses the reciprocal of the road network distance of each charging station in the spatial neighborhood of the abnormal site as the fusion weight, and uses a weighted average algorithm to calculate the correction value of the abnormal site based on the number of single-gun orders of each charging station in the spatial neighborhood and its fusion weight.

[0091] For data packet loss anomalies, this embodiment uses the nearest normal time point before and after the missing segment as the endpoint to perform linear interpolation to complete the abnormal data points.

[0092] For real sudden anomalies, this embodiment retains the original values ​​without modification and labels these data points with real sudden anomaly tags.

[0093] In other embodiments, the following methods can also be used to correct detected abnormal data points: If it is an isolated outlier, replace it with the average of the two adjacent time points; If the segment is a continuous anomalous segment, linear interpolation should be used for reconstruction. For jumps between the predicted order count and the real-time order count, a weighted smoothing method is used, and the calculation formula is as follows: Corrected order count = β × Real-time single-gun order count + (1-β) × Predicted single-gun order count In this embodiment, β=0.3.

[0094] S142 performs a logic rationality check, verifying that the predicted number of orders per charging station is within its physical service capacity. For example, the predicted number of orders per charging station should not exceed the product of the number of charging piles and the maximum service capacity coefficient per charging pile. For predicted values ​​that exceed a reasonable range, they are truncated to the boundary value.

[0095] S143 generates a resource planning strategy based on the verified charging station data and historical single-gun order data. In this embodiment, the charging station resource planning strategy is as follows: The number of single-gun orders in the verified charging station data is counted, and periods when the number of single-gun orders exceeds a preset busy threshold are identified and recorded as busy periods. If a charging station experiences single-gun order numbers exceeding the preset busy threshold for more than 5 consecutive days and the duration exceeds 6 hours, a suggestion to increase equipment allocation is generated. For example, it is suggested to increase the number of charging piles at the charging station, with the increase calculated based on the ratio of peak utilization to the existing number of charging piles.

[0096] Where A is the number of additional units; B is the maximum number of single-gun orders; C is the number of existing charging piles; and k is the gain coefficient (e.g., 0.5). This is an operation to find the minimum value; This is for floor function.

[0097] Example 2: Refer to Figure 2 The difference between this embodiment and Embodiment 1 is that this embodiment further includes: S21 Warning: In this embodiment, by statistically analyzing the number of single-gun orders in the data table after verification of each charging station, the time period in which the number of single-gun orders exceeds the preset busy threshold (such as 1.2) is identified and recorded as a busy time period.

[0098] Based on the selected busy periods, a status label is added to each charging station at each time point. The status label includes a busy label (denoted as 1) or a non-busy label (denoted as 0). Specifically, if the charging station belongs to any busy period at time t, then the status label L(t) of the charging station at time t is equal to 1; otherwise, the status label L(t) of the charging station at time t is equal to 0.

[0099] Based on the status tags of each charging station, a status warning signal is generated and sent. The warning rules are as follows: When the current status label L(t) of a charging station is equal to 0 (not busy), but the predicted status label will become 1 (busy) in the next 15 minutes, an early warning signal of impending busyness is generated. When the current status label L(t) of a charging station is equal to 1 (busy), and the busy state is expected to last for more than 30 minutes, a warning signal for continuous busyness is generated. When the status label L(t) of a charging station is equal to 0 for three consecutive time points (each time point is 15 minutes apart), but the charging pile utilization rate shows a continuous upward trend, a trend warning signal is generated.

[0100] Warning signals can be sent in the following ways: by pushing pop-up messages to the operations management platform and by sending SMS notifications to the mobile terminals of designated operations and maintenance personnel.

[0101] S22 Status Identification: Based on the verified point table data, the status change trend of each charging station is generated. The status change trend is characterized by calculating the rate of change of charging pile utilization over time.

[0102] For charging station j, the calculation model for its state change trend is as follows:

[0103] in, The numerical value represents the trend of the state change of charging station j at time t; For charging station j in The utilization rate of charging stations at any given time; For charging station j in The utilization rate of charging stations at any given time; For time intervals.

[0104] when When this time, it indicates an upward trend in busy status; when When the busy state is decreasing, it indicates that the busy state is trending downward; when When the time is right, it indicates a stable state.

[0105] Based on the state change trend, identify charging stations whose current state label is non-busy (i.e., L(t)=0), but whose state change trend points to a busy state. ,in, In this embodiment, the preset trend threshold is used. The value is 0.2% per minute (i.e., it increases by 3 percentage points every 15 minutes), and it is recorded as the set of candidate charging stations.

[0106] The estimated travel time from the i-th busy depot to each candidate charging station is obtained. The method for calculating the estimated travel time is determined based on the location of the busy depot and the candidate charging stations: For high-speed depots, this embodiment calculates the estimated travel time based on the road network distance between the i-th busy depot and each candidate charging station; for non-high-speed depots, this embodiment calculates the estimated travel time based on the road network distance between the i-th busy depot and each candidate charging station.

[0107] Based on the stated state change trend, the transition probability of each candidate charging station changing from a non-busy state to a busy state after the estimated duration is calculated. The calculation process for the transition probability includes S221 calculating the baseline probability and S222 making a correction.

[0108] S221 calculates the baseline probability. In the historical training samples, historical time periods with similar trends to the current state are retrieved. This embodiment uses a dynamic time warping algorithm to calculate time series similarity, with a similarity threshold set to 0.85. Historical time periods with a similarity greater than 0.85 are considered similar historical time periods. The frequency of transitions from a non-busy state to a busy state within the estimated duration is then statistically analyzed among all similar historical time periods to obtain the baseline probability. The formula for calculating the baseline probability is as follows:

[0109] S222 Correction: The baseline probability is dynamically corrected by using the time difference between the current time and the predicted busy time and the rate of change of the state change trend, and the corrected baseline probability is used as the transition probability.

[0110] The revised baseline probability calculation model is shown below:

[0111] in, This is the corrected baseline probability; This is the time difference weighting coefficient, with a value range of 0.01-0.05; This represents the time difference between the current moment and the predicted busy moment. This is the acceleration penalty coefficient, with a value ranging from 0.1 to 0.5; The rate of change of the state change trend is the first derivative of the state change trend. This is the baseline probability.

[0112] Preferably, if the corrected baseline probability exceeds the range of [0,1], this embodiment further applies boundary constraints to the corrected baseline probability, limiting it to the range of [0,1]. The boundary constraint method is as follows:

[0113] in, The baseline probability after boundary constraint processing; To obtain the maximum value; This is the corrected baseline probability; This is done to obtain the minimum value.

[0114] Finally, the corrected baseline probability or the baseline probability after constraint processing is used as the transition probability.

[0115] S23 sends information, sets a preset probability threshold (preferably 0.65), and sends information on candidate charging stations with a transition probability less than the preset probability threshold. Candidate charging stations with a transition probability less than the preset probability threshold are more likely to remain in a non-busy state after the estimated time, and are suitable as diversion targets.

[0116] The information sent includes: candidate charging station number, address, reachability distance, hop probability value, and navigation route from the currently busy charging station to the station.

[0117] Example 3: This example differs from Example 2 in that, after adding status tags to each charging station according to peak hours, it also includes: Based on the information of the charging stations, charging stations are divided into high-speed stations and non-high-speed stations. The distinction between high-speed and non-high-speed stations is based on the high-speed station identifier in the information table. Based on the status label, charging stations are divided into busy stations (current status label L(t)=1) and non-busy stations (current status label L(t)=0).

[0118] Iterate through the high-speed station identifiers of all busy stations and determine whether the i-th busy station is a high-speed station.

[0119] If the i-th busy depot is a highway depot, then along the legal driving direction of the route containing the i-th busy depot and its adjacent road segments, non-busy depots are searched, as follows: Rule 1: The road segment where the current busy depot is located (e.g., a 20-kilometer radius before and after the current busy depot), and the adjacent road segments directly connected to this road segment; Rule 2: Search only in the legal direction of travel, do not search in the opposite lane or the reverse direction; Rule 3: Non-busy stations (status label L(t)=0).

[0120] The searched non-busy stations are aggregated into a set. Where K is the number of non-busy stations found; This is the i-th non-busy station.

[0121] If K>0, then the non-busy depot that is closest to the i-th busy depot in the search results will be the recommended diversion depot for the i-th busy depot. The distance between the search results and the i-th busy depot is calculated using the actual mileage of the path between the i-th busy depot and the recommended diversion depot along the highway, rather than the straight-line distance.

[0122] If K=0, expand the search range and repeat the above search process until a recommended diversion station can be output and then stop the search.

[0123] If the i-th busy depot is not a high-speed depot, then obtain the non-busy depots within the area to which the i-th busy depot belongs, and denote this as the target depot set. The area can be determined based on administrative divisions, or within a radius of 5 kilometers centered on the i-th busy depot.

[0124] For each target station in the target station set, calculate the road network distance between the i-th busy station and each target station. The road network distance adopts the actual shortest path length in the urban road network. In this embodiment, the road network distance is calculated by Dijkstra's algorithm or by calling the map API.

[0125] The target depot with the shortest road network distance is selected as the recommended diversion depot for the i-th busy depot.

[0126] Preferably, if multiple target stations are at the same distance from busy stations (within ±50 meters), the remaining service capacity (number of available charging piles) of each target station at the same distance is further compared, and the station with the largest remaining service capacity is selected as the final recommended diversion station.

[0127] All busy charging stations are traversed to generate a recommendation mapping table. This table stores information about each busy charging station and its corresponding recommended diversion station, as well as the distance from each busy station to the recommended diversion station. The information for each recommended diversion station includes the station name, geographical coordinates, number of charging piles, and current status tag.

[0128] In other embodiments, if the i-th busy depot is not a high-speed depot, the method further includes: The busy depots are matched with the target depots (i.e., the non-busy depots in the same area) in terms of time series data. The matching process is as follows: From the verified point table data, extract the status label sequences of busy and unbusy stations at each time point in the past 7 days, with the time discretization granularity set to 15 minutes.

[0129] For the station pair (D, E), the complementarity calculation formula is as follows:

[0130] in, The complementarity of the station pair (D, E); Represents the set at time points The number of times when charging station D is busy and charging station E is idle. Represents the set at time points The total number of busy times at charging station D in the middle; This is the set of all time points within a statistical time period; Represents a set of time points a point in time .

[0131] Preferably, since there may be time shifts during peak hours, the time series of charging station E is shifted forward or backward by a certain number of time points (e.g., ±30 minutes) using the sliding matching method, and the complementarity is recalculated. The maximum value is taken as the final complementarity.

[0132] Based on the matching results (i.e., pairs of stations with different status labels), obtain pairs of stations that are complementary in terms of busy and non-busy periods (i.e., time-displaced complementary relationships), and denot them as time-complementary candidate pairs. Add the target stations that are time-complementary candidate pairs with the i-th busy station to the recommendation mapping table.

[0133] To facilitate user selection, in this embodiment, a target site is added to the recommendation mapping table by selecting one from the time-complementary candidate pairs. The selection criteria include: Screening criterion 1: The complementarity with the i-th busy station is the largest and ≥0.75; Screening criterion 2: Select the charging station with the smallest road network distance between the i-th busy station and the target station that is a time-complementary candidate pair with it.

[0134] Example 4: This example discloses a data verification system for predicting busy charging stations for electric vehicles. The system includes a processor and a memory communicatively connected to the processor. The memory is provided with a computer-readable storage medium, and a computer program is stored on the computer-readable storage medium. When the processor executes a computer program stored on the computer-readable storage medium, it implements the method.

[0135] The above are all preferred embodiments of this application and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A method for verifying data from a predicted list of busy charging stations for electric vehicles, characterized in that, include: Real-time charging data and historical charging data for a first preset duration before the current time of each charging station are obtained respectively, and training samples are constructed based on the historical charging data of each charging station. The preset prediction model is trained using training samples to obtain the trained prediction model. All real-time charging data is input into the trained prediction model, and the predicted charging data is output. The date matrix is ​​populated with all real-time charging data and predicted charging data. The date matrix is ​​a two-dimensional data structure with time as the row index and the number of orders per charging station as the filling data. The information table is populated with information of each charging station. The information table is a relational data table with the charging station number as the primary key and stores the static attribute information of each charging station. The association between the information table and the date matrix is ​​established, and the dynamic status data in the date matrix is ​​associated and merged with the static attribute data in the information table to obtain the point table data of each charging station. The charging station data is verified according to preset rules to obtain verified charging station data. Based on the verified charging station data and historical charging data, a charging equipment resource planning strategy is generated. After obtaining the verified point table data, the following is also included: According to the preset busy status rules, busy periods are filtered from the verified point table data, and status tags are added to each charging station according to the busy periods. The status tags include: busy tags and non-busy tags. Generate and send status warning signals based on the status tags of each charging station; Based on the verified point table data, generate the status change trend of each charging station. Based on the status change trend, identify candidate charging stations whose current status label is non-busy, but whose status change trend points to a busy state. Get the estimated time from the i-th busy station to each candidate charging station, and calculate the transition probability of each candidate charging station from a non-busy state to a busy state after the estimated time according to the state change trend, and send information of candidate charging stations whose transition probability is less than a preset probability threshold. The transition probability is calculated as follows: Based on the actual busy transition frequency under similar state change trends in the training samples, a baseline probability is obtained. The baseline probability is then dynamically corrected using the time difference between the current moment and the predicted busy moment, as well as the rate of change of the state change trend. The corrected baseline probability is then used as the transition probability.

2. The method for verifying the predicted charging station data of electric vehicles according to claim 1, characterized in that, After adding status labels to each charging station according to peak hours, the following is also included: Based on the information of the charging station, the charging station is divided into high-speed stations and non-high-speed stations, and based on the status label, the charging station is divided into busy stations and non-busy stations. Determine whether the i-th busy depot is a highway depot. If so, search for non-busy depots along the legal driving direction of the route where the i-th busy depot is located, including the road segment and adjacent road segments. Obtain the search results and select the non-busy depot closest to the i-th busy depot as the recommended diversion depot. If not, obtain the non-busy depots within the area to which the i-th busy depot belongs, and denote them as target depots. Calculate the road network distance between the i-th busy depot and each target depot, and select the target depot with the smallest road network distance as the recommended diversion depot.

3. The method for verifying the data of the predicted busy charging station table for electric vehicles according to claim 2, characterized in that, Also includes: Traverse all busy stations and generate a recommendation mapping table. The recommendation mapping table stores information about recommended diversion stations and the distance from each busy station to the recommended diversion station.

4. The method for verifying the data of the predicted busy charging station table for electric vehicles according to claim 3, characterized in that, When the i-th busy depot is a non-high-speed depot, it also includes: The busy stations are matched with the target stations in terms of time series. Based on the matching results, station pairs that are complementary in terms of busy and non-busy periods are obtained and recorded as time complementary candidate pairs. Add the target stations that are time-complementary candidate pairs with the i-th busy station to the recommendation mapping table.

5. The method for verifying the data of the predicted busy charging station table for electric vehicles according to claim 1, characterized in that, The historical charging data within the second preset time period includes charging data from the same period within the first preset time period, or charging data from a portion of the first preset time period.

6. A data verification system for predicting busy charging station locations for electric vehicles, characterized in that, include: A processor, and a memory communicatively connected to the processor; The memory is provided with a computer-readable storage medium, and a computer program is stored on the computer-readable storage medium. When the processor executes a computer program stored on the computer-readable storage medium, it implements the method as described in any one of claims 1-5.

Images