Charging pile cluster ordered charging method based on traffic flow optimization

Abstract

The invention relates to the technical field of new energy automobile charging scheduling, in particular to a charging pile cluster ordered charging method based on traffic flow optimization, which comprises the steps of collecting traffic flow, power grid load data, charging pile operation state data and user related data of a target area; deploying a plurality of sensors for different scene combinations, and adjusting the weight of each sensor based on environmental conditions; a deep learning model is adopted to fuse multi-source data, and a traffic flow distribution thermodynamic diagram and a power grid load prediction curve in the future 1-6 hours are output; performing cross-cluster traffic flow guidance and in-station power flexible distribution based on a prediction result; and issuing a scheduling instruction, collecting user response data, and iteratively optimizing the model and the scheduling strategy. According to the invention, the technical problems of lack of traffic flow dynamic perception, insufficient cross-cluster collaboration and low user participation degree in the existing charging scheduling technology can be solved.

Description

Technical Field

[0001] This invention relates to the field of new energy vehicle charging scheduling technology, and in particular to an orderly charging method for charging pile clusters based on traffic flow optimization. Background Technology

[0002] With the rapid popularization of electric vehicles, the application of large-scale charging pile clusters is becoming increasingly widespread. However, this also faces the dual challenges of increased pressure on the power grid during peak and off-peak hours and an imbalance in the spatial and temporal allocation of resources. User charging behavior is mostly concentrated during peak electricity consumption periods, which can easily lead to local power grid overload. In popular areas, charging piles often experience queuing, while in less popular areas, charging pile clusters remain idle for a long time, resulting in low resource utilization.

[0003] Existing technologies mostly adopt single-dimensional optimization schemes such as time-of-use pricing and power allocation within stations based on vehicle power priority. However, due to the lack of static traffic flow prediction, insufficient cross-cluster coordination capabilities, and weak economic incentives for users, it is difficult to achieve global coordinated optimization of dynamic traffic flow distribution and grid load. This fails to meet the actual needs of improving system efficiency and enhancing user participation. There is an urgent need to break through the technical bottlenecks of multi-source data fusion prediction and multi-objective decision-making scheduling. Summary of the Invention

[0004] The purpose of this invention is to propose an orderly charging method for charging pile clusters based on traffic flow optimization, which solves the technical problems of lack of dynamic perception of traffic flow, insufficient cross-cluster coordination and low user participation in existing charging scheduling technologies. It enables accurate prediction of charging demand and cross-regional traffic flow guidance, optimizes power allocation within the station, stimulates users to actively respond, and ultimately achieves the goals of reducing peak grid load, shortening user waiting time and improving equipment utilization.

[0005] To achieve this objective, the present invention adopts the following technical solution: A method for orderly charging of charging pile clusters based on traffic flow optimization includes the following steps: S1. Collect traffic flow, power grid load data, charging pile operation status data, and user-related data in the target area; S2. Deploy multiple sensors in combination for different scenarios and adjust the weights of each sensor based on environmental conditions; S3. Employ a deep learning model to fuse multi-source data and output a heat map of traffic flow distribution and a power grid load prediction curve for the next 1-6 hours. S4. Based on the prediction results, conduct cross-cluster traffic flow guidance and flexible power allocation within the station; S5. Issue scheduling instructions and collect user response data to iteratively optimize the model and scheduling strategy.

[0006] Preferably, in S1, the data collection is achieved through traffic monitoring equipment, power grid load sensors, charging pile status monitoring terminals, and user mobile APP, wherein the traffic monitoring equipment collects traffic flow data with a time granularity of one data point every 15 minutes.

[0007] Preferably, in S2, the different scenarios involve deploying multiple sensors as follows: When the scenario is an intersection, a combination of video sensors, microwave sensors, and geomagnetic sensors is deployed. When the scenario is a highway, a combination of microwave sensors, RFID sensors and ultrasonic sensors should be deployed. When the scenario is a parking lot, deploy a combination of geomagnetic sensors and video sensors; The sensor weight adjustment is based on a Bayesian network and is dynamically determined according to environmental conditions such as light intensity and rainfall level.

[0008] Preferably, in S3, the deep learning model is a dual-channel temporal deep learning model based on an attention mechanism, including a main channel, an auxiliary channel, and an attention layer; The main channel employs a long short-term memory network or a temporal convolutional network to learn long-term dependency patterns in temporal data; the auxiliary channel processes non-strict temporal features such as historical charging behavior and weather; and the attention layer dynamically calculates the importance weights of each input feature for different prediction periods.

[0009] Preferably, the input data for the deep learning model includes traffic flow data, historical charging behavior data, and weather data; The traffic flow data includes the average traffic density, average driving speed, and congestion index of key roads in the region; the historical charging behavior data includes the historical total charging volume, number of charging orders, average charging time, and typical charging curves for each cluster; and the weather data includes temperature, humidity, precipitation type and probability, wind speed, light intensity, and extreme weather indicators.

[0010] Preferably, the deep learning model adopts an update strategy that combines online learning with periodic retraining, performing incremental learning during the early morning hours when business is at a low point, and performing full retraining weekly or monthly using all data from the past three months.

[0011] Preferably, in S4, the cross-cluster traffic guidance specifically includes the following steps: Calculate the real-time load rate of each charging pile cluster. and the global average load rate of the cluster within the region ,in, For the first Real-time charging power of each cluster For the cluster's rated capacity, The total number of clusters within the region; Dynamic discount coupons are generated for clusters with loads below the regional average. The optimal charging route, which includes travel time and economic cost, is calculated based on real-time traffic data and pushed to the user's mobile app. With V2G technology enabled, vehicles supply power to the grid in reverse. Static peak hours are defined as 07:00-10:00 and 17:00-21:00, while static off-peak hours are all other times excluding the static peak hours. The initial default value for the compensation coefficient is set to: static peak hours. r peak =1.35, static off-peak hours r o ff =1.15, where the compensation coefficient is determined by a mapping function. Automatic adjustment; By improving LightGBM, the charging demand of each charging station in time slice t is predicted to obtain D(t), and the prediction error distribution is established using kernel density estimation. A scarcity index is constructed using short-term load forecasting and error estimation. ,in, For the available grid capacity of the cluster, The risk adjustment factor is 1.64-1.96. Normalization process is performed to obtain ,pass Determine if a dynamic peak is triggered.

[0012] Preferably, the upper limit of the instantaneous active power of the vehicle's reverse power supply. ≤5kW, with a default SOC lower limit of 20%, vehicles are prohibited from supplying power to the grid in reverse when below this threshold; certified bidirectional metering devices are used, with a metering resolution of no less than 1 minute and an energy accuracy of no less than 0.01kWh. Metering data includes the pile ID and timestamp and is digitally signed, and data records are kept for at least 3 years; settlement adopts the default next-day batch settlement or optional small-amount real-time settlement mode. In the event of an anomaly or communication failure, compensation is temporarily frozen and the settlement result is corrected after verification.

[0013] Preferably, in S4, the flexible allocation of power within the station includes the following specific steps: According to the formula via the battery management system Calculate battery health, where, For the battery at all times Real-time available capacity, This refers to the initial rated capacity of the battery. Battery grades are classified according to State of Health (SOH) and corresponding charging modes: When SOH ≥ 80%, fast charging mode is activated, with a maximum charging power of [value missing]. , ; When 60%≤SOH<80%, the conventional charging mode is used, and the power is... , ; When SOH < 60%, current-limited slow charging is implemented, with the lower limit of charging power being [value missing]. , ; Introducing power allocation factor Based on the real-time available power of the cluster Allocate charging power according to SOH priority to ensure ≤ Where M represents the number of vehicles waiting to be charged at the station. For the first The car at any time The charging power.

[0014] Preferably, in S5, the issuance of scheduling instructions and collection of user response data are implemented through a distributed system consisting of a cloud server, a charging pile controller, and a user mobile APP. The cloud server sends scheduling instructions to the charging pile controller via the MQTT protocol, and the controller parses and executes the instructions before feeding back the status. The user's mobile app synchronizes information with the cloud server via a WebSocket long connection, and refreshes the interface in real time based on JSON data with animated prompts. The user's mobile APP has multiple data collection points to collect user operation behavior and feedback data. The data is cleaned and preprocessed before being stored. Machine learning algorithms are used to analyze and mine the data, and the dynamic prediction model and scheduling strategy are updated periodically.

[0015] One of the above technical solutions has the following beneficial effects: (1) Breaking the limitations of single-dimensional optimization such as time-of-use pricing and single-station power allocation in existing technologies, we construct a full-process, multi-dimensional technical system of "data acquisition - sensor collaboration - dynamic prediction - scheduling decision - iterative optimization" to achieve coordinated optimization of dynamic distribution of traffic flow and power grid load from a global perspective, effectively solving the dual problems of increased peak and valley pressure of power grid load and imbalance of resource temporal and spatial allocation.

[0016] (2) The closed-loop iterative optimization mechanism enables the system to respond in real time to the dynamic changes in traffic flow, grid load and user behavior, significantly improving the system’s flexibility and adaptability, ensuring the stability and reliability of charging services, and promoting the evolution of vehicle-charging-grid from local single-point scheduling to global dynamic collaborative optimization.

[0017] (3) Provide a unified implementation framework for subsequent detailed technical solutions, ensure the organic connection and coordinated cooperation of various technical modules such as sensor collaboration, dynamic prediction, and scheduling decision-making, maximize the overall technical effect, and enhance the feasibility and promotion value of the system. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating an orderly charging method for a cluster of charging piles based on traffic flow optimization. Figure 2 This is a schematic diagram of traffic flow data collection in a certain area in a charging pile cluster orderly charging method based on traffic flow optimization; Figure 3 This is a traffic dynamic prediction map of a certain area in a charging pile cluster orderly charging method based on traffic flow optimization; Figure 4 This is a schematic diagram of the charging pile cluster-microgrid structure in an orderly charging method based on traffic flow optimization. Figure 5 This is a schematic diagram of the charging power allocation strategy based on SOH in an orderly charging method for charging pile clusters based on traffic flow optimization. Detailed Implementation

[0019] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0020] A method for orderly charging of charging pile clusters based on traffic flow optimization includes the following steps: S1. Collect traffic flow, power grid load data, charging pile operation status data, and user-related data in the target area; S2. Deploy multiple sensors in combination for different scenarios and adjust the weights of each sensor based on environmental conditions; S3. Employ a deep learning model to fuse multi-source data and output a heat map of traffic flow distribution and a power grid load prediction curve for the next 1-6 hours. S4. Based on the prediction results, conduct cross-cluster traffic flow guidance and flexible power allocation within the station; S5. Issue scheduling instructions and collect user response data to iteratively optimize the model and scheduling strategy.

[0021] This technical solution constructs a complete technical framework for the orderly charging of charging pile clusters based on traffic flow optimization. It forms a closed-loop collaborative optimization system through five core steps to achieve global dynamic coordination between vehicles, charging piles, and the network: S1 mainly involves data acquisition steps: integrating multiple sources of equipment such as traffic monitoring equipment, power grid load sensors, charging pile status monitoring terminals, and user mobile APPs to comprehensively collect traffic flow, power grid load data, charging pile operation status data, and user-related data in the target area, providing basic data support for subsequent analysis. S2 mainly involves the sensor collaborative deployment steps: for different scenarios such as intersections, highways, and parking lots, multiple sensors are deployed in combination, and the weights of each sensor are dynamically adjusted based on environmental conditions to ensure the accuracy and adaptability of data collection. S3 primarily involves dynamic forecasting: A deep learning model is used to fuse multi-source data, deeply mining the complex relationship between traffic flow and grid load. This outputs a traffic flow heatmap and grid load forecast curve for the next 1-6 hours, providing a scientific basis for scheduling decisions. The traffic flow heatmap, presented as a Geographic Information System (GIS) map, visualizes the road traffic density and the expected number of vehicles heading towards charging stations in each area over the next 1-6 hours. The grid load forecast curve is a time-series curve, predicting the total active power demand (kW) of the entire charging station cluster and its substations over the next 1-6 hours. S4 mainly involves scheduling decision-making steps: based on the prediction results, dual scheduling decisions are made. On the one hand, cross-cluster traffic flow optimization is achieved through economic incentives and path guidance. On the other hand, flexible power allocation within the station is carried out in combination with battery health status to achieve efficient resource utilization. S5 mainly involves iterative optimization steps: sending scheduling instructions to the charging pile controller and updating the user APP interface simultaneously, while collecting user response data, continuously iterating and optimizing the model and scheduling strategy to ensure that the system always adapts to the dynamically changing operating environment.

[0022] To further explain, in S1, the data collection is achieved through traffic monitoring equipment, power grid load sensors, charging pile status monitoring terminals, and user mobile APP. The traffic monitoring equipment collects traffic flow data with a time granularity of one data point every 15 minutes.

[0023] The specific implementation methods and key technical parameters for data collection are further clarified, covering all dimensions of information related to vehicles, charging piles, the network, and users through a multi-device collaborative collection mode. Specifically, traffic monitoring equipment (including intersection cameras and real-time traffic data from navigation apps) is responsible for collecting traffic flow data, power grid load sensors collect power grid load data, charging pile status monitoring terminals collect charging pile operating status data, and user mobile apps collect relevant user data. Given the unique characteristics of traffic flow data, its time granularity is set to one data point every 15 minutes. This setting not only captures the dynamic changes in traffic flow in a timely manner, ensuring data timeliness, but also avoids data redundancy and excessive processing pressure caused by overly fine data granularity. This provides high-quality, timely input data for subsequent dynamic prediction models, ensuring the accuracy of prediction results. Figure 2 The diagram shows a sample of traffic flow data collected in a certain area over a period of nearly 24 hours on a certain day.

[0024] The above multi-device collaborative acquisition mode ensures the comprehensiveness and integrity of the data source, solving the problems of low prediction accuracy and one-sided scheduling decisions caused by single data sources in the existing technology. It provides complete data support for building a dynamic collaborative model of vehicle flow and power grid, enabling the model to more comprehensively explore the correlation between vehicle flow and power grid load.

[0025] To further explain, in S2, the different scenarios involve deploying multiple sensors as follows: When the scenario is an intersection, a combination of video sensors, microwave sensors, and geomagnetic sensors is deployed. When the scenario is a highway, a combination of microwave sensors, RFID sensors and ultrasonic sensors should be deployed. When the scenario is a parking lot, deploy a combination of geomagnetic sensors and video sensors; The sensor weight adjustment is based on a Bayesian network and is dynamically determined according to environmental conditions such as light intensity and rainfall level.

[0026] Based on the environmental characteristics and data acquisition requirements of different application scenarios, a targeted sensor combination scheme is designed and dynamic weight adjustment is implemented: In intersection scenarios, a combination of video, microwave, and geomagnetic sensors is used. Video sensors capture vehicle characteristics and lane-changing behavior, microwave sensors detect vehicle speed and position, and geomagnetic sensors verify vehicle presence to avoid misjudging stationary vehicles. In highway scenarios, a combination of microwave, RFID and ultrasonic sensors is used. Microwave sensors monitor traffic speed in real time, RFID sensors identify ETC vehicles and record their routes, and ultrasonic sensors detect illegal parking in emergency lanes. In parking lot scenarios, a combination of geomagnetic and video sensors is used. The geomagnetic sensor detects the occupancy status of parking spaces, while the video sensor identifies license plates and generates parking records.

[0027] The aforementioned scenario-based sensor combination scheme fully leverages the technical advantages of each sensor and compensates for the performance shortcomings of a single sensor in specific scenarios. For example, the geomagnetic sensor solves the problem of video sensors misjudging stationary vehicles, and the RFID sensor enables accurate identification of vehicle paths in highway scenarios, significantly improving the accuracy and reliability of data collection and providing high-quality data support for subsequent dynamic prediction and scheduling decisions.

[0028] Meanwhile, based on Bayesian networks, dynamic weight fusion of sensors is achieved. Utilizing the probabilistic reasoning capability of Bayes' theorem, environmental conditions such as light intensity and rainfall level are used as evidence variables to calculate the reliability (probability) of different detection technologies under the current environment, thereby determining the weight of each sensor. For example, during a sunny day, the video sensor has a weight of 70%, the microwave sensor 20%, and the geomagnetic sensor 10%; during a heavy rain at night, the microwave sensor has a weight of 50%, the geomagnetic sensor 30%, and the RFID sensor 20%. This solves the problem of decreased data acquisition accuracy in complex environments with a single sensor solution, ensuring the stability and consistency of data acquisition and improving the system's adaptability to complex environments.

[0029] To further explain, in S3, the deep learning model is a dual-channel temporal deep learning model based on an attention mechanism, including a main channel, an auxiliary channel, and an attention layer; The main channel employs a long short-term memory network or a temporal convolutional network to learn long-term dependency patterns in temporal data; the auxiliary channel processes non-strict temporal features such as historical charging behavior and weather; and the attention layer dynamically calculates the importance weights of each input feature for different prediction periods.

[0030] A dual-channel temporal deep learning model based on an attention mechanism serves as the core prediction engine. It constructs an architecture that processes the main channel and auxiliary channel in parallel, while introducing an attention layer to optimize feature weight allocation.

[0031] The main channel employs a Long Short-Term Memory (LSTM) network or a Temporal Convolutional Network (TCN) to focus on learning long-term dependency patterns in historical traffic flow, average vehicle speed, and other time-series data, accurately capturing the temporal evolution characteristics of traffic flow and grid load. The auxiliary channel specifically handles features that are not strictly time-series but have strong correlations, such as historical charging behavior and weather, supplementing key influencing factors not covered by time-series data and achieving deep fusion of multi-dimensional features. An attention layer is introduced at the top of the model to dynamically calculate the importance weights of each input feature for different prediction periods. For example, during the upcoming peak period, the model focuses on core features such as the current congestion index, precipitation probability, and historical charging volume for the same period, enabling the model to focus on key influencing factors, improve prediction accuracy, and ultimately output a heatmap of traffic flow distribution and a predicted grid load curve for the next 1-6 hours. Figure 3 The image shown is a traffic dynamic forecast map for a certain area.

[0032] In summary, the dual-channel architecture enables parallel processing and deep fusion of time-series data and non-time-series features. It fully leverages long-term dependencies within time-series data while effectively integrating key external features such as historical charging behavior and weather conditions. This addresses the predictive bias issues caused by single-dimensional data processing in existing technologies, improving the generalization ability and accuracy of the prediction model. Furthermore, the introduction of the attention mechanism allows the model to dynamically adjust the weights of each input feature based on the needs of different prediction periods, focusing on core influencing factors. Especially during peak periods, it can more accurately predict traffic flow distribution and grid load, providing a scientific basis for scheduling decisions and ensuring the effectiveness of scheduling strategies.

[0033] To further explain, the input data for the deep learning model includes traffic flow data, historical charging behavior data, and weather data; The traffic flow data includes the average traffic density, average driving speed, and congestion index of key roads in the region; the historical charging behavior data includes the historical total charging volume, number of charging orders, average charging time, and typical charging curves for each cluster; and the weather data includes temperature, humidity, precipitation type and probability, wind speed, light intensity, and extreme weather indicators.

[0034] Further clarify the composition of input data for deep learning models and the specific characteristics of each data type. Through comprehensive and detailed input feature design, ensure that the model can fully learn the complex nonlinear relationship between traffic flow and power grid load.

[0035] The input data encompasses three core dimensions: traffic flow data, historical charging behavior data, and weather data. Traffic flow data originates from transportation department sensors and navigation platform APIs, including average vehicle density (vehicles / km), average speed (km / h), and congestion index on key roads in the region, directly reflecting traffic flow status. Historical charging behavior data comes from the charging pile cluster's backend database, including aggregated data such as total historical charging volume (kWh), number of charging orders, and average charging duration for each cluster at an hourly granularity, as well as pattern data such as typical charging curves for weekdays / weekends, holidays, and seasonality, enabling the discovery of patterns in user charging behavior. Weather data comes from the meteorological bureau's API, including temperature (°C), humidity (%), precipitation type and probability (%), wind speed (m / s), and light intensity. Extreme weather (such as heavy rain and high temperatures) is used as a special event marker, covering external environmental factors affecting traffic flow and charging demand. These comprehensive and detailed input features provide rich information support for the model, ensuring the accuracy of the prediction results.

[0036] To further explain, the deep learning model adopts an update strategy that combines online learning with periodic retraining. Incremental learning is performed during the early morning hours when business is at a low point, and complete retraining is performed weekly or monthly using all data from the past three months.

[0037] A model update strategy of "online learning + periodic retraining" was designed for this deep learning model. Update times are rationally scheduled based on business operation patterns to ensure long-term stability of the model's predictive performance. During the daily off-peak business hours (02:00-04:00), new data generated the previous day is automatically collected, including traffic flow data, charging behavior data, grid load data, and weather data. This data is used for incremental learning of the model, fine-tuning model parameters to enable it to quickly adapt to the latest traffic flow trends, changes in user charging behavior, and grid load fluctuations. Weekly or monthly, a complete retraining of the model is performed using all data from a relatively long period (e.g., 3 months). This comprehensively optimizes the model's basic weights, avoiding model drift caused by long-term incremental learning and ensuring stable long-term predictive performance. Simultaneously, the model uses root mean square error (RMSE) and mean absolute percentage error (MAPE) as core evaluation metrics. After training and validation with large-scale historical data, in peak-hour predictions on typical workdays, the MAPE for the grid load prediction of the next 1-2 hours is ≤8%, and the MAPE for the prediction of the next 3-6 hours is ≤15%, ensuring the model's predictive accuracy.

[0038] In summary, the daily incremental update strategy can promptly absorb effective information from the latest data, enabling the model to quickly adapt to dynamic changes in traffic flow, user behavior, and grid load. This solves the problem of declining prediction accuracy over time caused by untimely model updates in existing technologies, ensuring the model's short-term prediction performance and ensuring that scheduling decisions can respond promptly to the latest operational status. Meanwhile, weekly / monthly periodic retraining optimizes the model's base weights, effectively suppressing model drift and ensuring long-term model stability and prediction accuracy. This avoids the accumulation of model parameter biases caused by long-term reliance on incremental learning, extending the model's effective lifespan and reducing model maintenance costs. Furthermore, updating the model during off-peak hours avoids impacting charging services and data processing, ensuring normal system operation and achieving a synergistic balance between model optimization and business services, thus improving the overall system efficiency.

[0039] To further explain, in S4, the cross-cluster traffic guidance specifically includes the following steps: Calculate the real-time load rate of each charging pile cluster. and the global average load rate of the cluster within the region ,in, For the first Real-time charging power of each cluster For the cluster's rated capacity, The total number of clusters within the region; Dynamic discount coupons are generated for clusters with loads below the regional average. The optimal charging route, which includes travel time and economic cost, is calculated based on real-time traffic data and pushed to the user's mobile app. With V2G technology enabled, vehicles supply power to the grid in reverse. Static peak hours are defined as 07:00-10:00 and 17:00-21:00, while static off-peak hours are all other times excluding the static peak hours. The initial default value for the compensation coefficient is set to: static peak hours. r peak =1.35, static off-peak hours r o ff =1.15, where the compensation coefficient is determined by a mapping function. Automatic adjustment; By improving LightGBM, the charging demand of each charging station in time slice t is predicted to obtain D(t), and the prediction error distribution is established using kernel density estimation. A scarcity index is constructed using short-term load forecasting and error estimation. ,in, For the available grid capacity of the cluster, The risk adjustment factor is 1.64-1.96. Normalization process is performed to obtain ,pass Determine if a dynamic peak is triggered.

[0040] The complete implementation process of cross-cluster traffic flow guidance is further elaborated, and global optimal matching between traffic flow and grid load is achieved through load factor quantification, economic incentives, V2G technology application and dynamic peak determination.

[0041] First, the real-time load rate of each charging pile cluster is calculated using a real-time load rate quantification model. and the global average load rate of the cluster within the region The system clarifies the load status of each cluster; generates dynamic discount coupons (e.g., 20% reduction in charging fees in low-load areas) for clusters with loads below the regional average; calculates the optimal charging route, including travel time and economic costs, using real-time traffic data and pushes it to the user's app, incentivizing users to actively divert traffic; enables V2G technology, allowing vehicles to supply power to the grid; defines static peak hours as 07:00-10:00 and 17:00-21:00, and static off-peak hours as the remaining time periods; and sets a default compensation coefficient for peak hours. r peak =1.35, off-peak hours r o ff =1.15; Simultaneously, the improved LightGBM is used to predict the charging demand of each charging station in time slice t to obtain D(t), and kernel density estimation (KDE) is used to establish the prediction error distribution. A scarcity index is constructed using short-term load forecasting and error estimation. Normalizing S(t) yields ,pass Determine if a dynamic peak is triggered; the compensation coefficient is obtained through a mapping function. Automatic adjustment enables economic incentive optimization under the dual constraints of static time window and dynamic scarcity.

[0042] In summary, dynamic discount coupons based on load factor and the optimal charging route guidance mechanism encourage users to charge in low-load areas through economic incentives, effectively alleviating the imbalance in the spatial and temporal allocation of resources, such as queuing in popular areas and idle charging piles in less popular areas. This significantly improves the overall utilization rate of the charging pile cluster and shortens user waiting time. Furthermore, the combination of V2G technology and a dual peak-valley determination mechanism (static + dynamic) provides a basic incentive framework through static time period division and accurately captures real-time changes in charging demand through dynamic scarcity indicators, automatically adjusting the compensation coefficient. This achieves two-way interaction of "peak-hour electricity sales and off-peak electricity storage," significantly reducing the peak-valley load difference in the power grid and lowering the peak load (experimental data shows a reduction of 15%~25%), thus enhancing grid stability. Simultaneously, the differentiated compensation coefficient design and dynamic adjustment mechanism greatly enhance user participation. The basic compensation coefficient of 135% during peak hours and 115% during off-peak hours, along with the dynamic adjustment strategy, stimulates users' initiative in actively responding to peak shaving and valley filling. Actual measurements show an increase in user active participation rate of over 30%, further alleviating grid pressure and achieving a win-win situation for both the grid and users.

[0043] To further clarify, the upper limit of the instantaneous active power of the vehicle's reverse power supply... ≤5kW, with a default SOC lower limit of 20%, vehicles are prohibited from supplying power to the grid in reverse when below this threshold; certified bidirectional metering devices are used, with a metering resolution of no less than 1 minute and an energy accuracy of no less than 0.01kWh. Metering data includes the pile ID and timestamp and is digitally signed, and data records are kept for at least 3 years; settlement adopts the default next-day batch settlement or optional small-amount real-time settlement mode. In the event of an anomaly or communication failure, compensation is temporarily frozen and the settlement result is corrected after verification.

[0044] A comprehensive safeguard plan has been developed for the safe operation and compliant settlement of V2G technology, clearly defining reverse power supply constraints, metering standards, settlement models, and anomaly handling mechanisms. To protect the vehicle-mounted battery and on-site power grid equipment, the upper limit of instantaneous active power for a single vehicle to supply power to the grid in reverse is specified. ≤5kW (configurable from 0-10kW), this single-vehicle power constraint, together with the pile end and cluster concurrent power limit, constitutes the grid connection safety constraint; the default value of the SOC lower limit is set to 20%, and vehicles are prohibited from supplying power to the grid in reverse when below this threshold to avoid damage to the battery due to excessive discharge; certified bidirectional metering devices are used, with a metering resolution of not less than 1 minute and an energy accuracy of not less than 0.01kWh. The metering data must include the pile ID and timestamp and be digitally signed to prevent tampering. Data records must be kept for at least 3 years to ensure the authenticity, accuracy, and traceability of the metering data; the settlement mode adopts two methods: default next-day batch settlement (exporting the metered energy from 00:00 to 24:00 on the settlement day and completing the settlement and accounting on the next day) and optional small-amount real-time settlement (requiring higher data granularity, online signature, and additional service fee strategy) to meet the needs of different users; in the event of an anomaly or communication failure, compensation is temporarily frozen and the settlement result is corrected after verification to ensure the auditability and compliance of the settlement.

[0045] To further explain, in S4, the flexible allocation of power within the station includes the following specific steps: According to the formula via the battery management system Calculate battery health, where, For the battery at all times Real-time available capacity, This refers to the initial rated capacity of the battery. Battery grades are classified according to State of Health (SOH) and corresponding charging modes: When SOH ≥ 80%, fast charging mode is activated, with a maximum charging power of [value missing]. , ; When 60%≤SOH<80%, the conventional charging mode is used, and the power is... , ; When SOH < 60%, current-limited slow charging is implemented, with the lower limit of charging power being [value missing]. , ; Introducing power allocation factor Based on the real-time available power of the cluster Allocate charging power according to SOH priority to ensure ≤ Where M represents the number of vehicles waiting to be charged at the station. For the first The car at any time The charging power.

[0046] like Figure 4As shown, flexible power allocation within the station is achieved based on battery health (SOH), taking into account battery protection, charging efficiency, and grid capacity constraints.

[0047] First, the battery management system follows the formula. Real-time calculation of battery health; classifying batteries into three levels based on SOH value and matching corresponding charging modes: When SOH ≥ 80% (battery condition is good), fast charging mode is activated, with a maximum charging power of [value missing]. ; When 60% ≤ SOH < 80% (medium battery status), use the standard charging mode with a power of [power value missing]. ; When SOH < 60% (poor battery condition), current-limited slow charging is implemented, with the lower limit of charging power being [missing value]. Meanwhile, considering grid capacity constraints, based on the real-time available power of the cluster P9 i d(t), introducing the power allocation coefficient The charging power is allocated to vehicles according to SOH priority, first meeting the needs of fast-charging vehicles, then proportionally allocating power to the regular charging group, and finally ensuring the minimum power for the slow-charging group. ≤ .

[0048] In summary, the differentiated charging mode design based on State of Health (SOH) dynamically adjusts the charging power according to the battery's health status, avoiding damage caused by high-power fast charging of batteries in poor health. This significantly extends battery life (estimated to increase by 20%~30%), reduces user battery maintenance costs and equipment replacement frequency, and improves the user charging experience. Furthermore, the power allocation strategy based on SOH priority, combined with grid capacity constraints, ensures that vehicles with good battery health can be charged quickly, improving charging efficiency, while ensuring that the total charging power does not exceed the grid's available capacity, avoiding local grid overload, enhancing grid stability, and achieving synergistic optimization of battery health protection, charging efficiency improvement, and grid load balancing. Simultaneously, the dynamic adjustment mechanism of the power allocation coefficient allows the system to flexibly adapt to changes in grid available power and the battery status of the vehicles being charged, optimizing the efficiency of charging power allocation, improving the utilization rate of charging piles, and ensuring the rational allocation of limited grid resources.

[0049] To further explain, in S5, the issuance of scheduling instructions and collection of user response data are achieved through a distributed system consisting of a cloud server, a charging pile controller, and a user mobile APP. The cloud server sends scheduling instructions to the charging pile controller via the MQTT protocol, and the controller parses and executes the instructions before feeding back the status. The user's mobile app synchronizes information with the cloud server via a WebSocket long connection, and refreshes the interface in real time based on JSON data with animated prompts. The user's mobile APP has multiple data collection points to collect user operation behavior and feedback data. The data is cleaned and preprocessed before being stored. Machine learning algorithms are used to analyze and mine the data, and the dynamic prediction model and scheduling strategy are updated periodically.

[0050] like Figure 5 As shown, a distributed execution feedback system based on a cloud server, a charging pile controller, and a user's mobile APP enables secure transmission of scheduling instructions, real-time information synchronization, and closed-loop optimization of the model.

[0051] The cloud server generates scheduling instructions based on real-time monitoring data and user reservation information, and transmits them to the charging pile controller via MQTT protocol with encryption to ensure the security and reliability of instruction transmission. After receiving the instructions, the charging pile controller parses and executes them, and feeds back the execution status to the cloud server, forming a closed loop of instruction execution. The user's mobile APP and the cloud server achieve real-time information synchronization through a WebSocket long connection. The APP refreshes the interface in real time based on JSON data and provides animated prompts to display key information such as discount information, optimal charging route, and charging status to the user. The APP sets up multiple data collection points to comprehensively collect user operation behavior (such as whether to accept scheduling and select charging duration) and feedback data. After cleaning and preprocessing, the data is stored in MySQL or MongoDB databases. Machine learning algorithms are used to analyze and mine the collected data to uncover user behavior patterns and demand preferences. The dynamic prediction model and scheduling strategy are regularly iterated and updated to enable the system to continuously adapt to changes in user needs and operating environment.

[0052] Furthermore, the functional units in the various embodiments of the present invention can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0053] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for orderly charging of charging pile clusters based on traffic flow optimization, characterized in that, Includes the following steps: S1. Collect traffic flow, power grid load data, charging pile operation status data, and user-related data in the target area; S2. Deploy multiple sensors in combination for different scenarios and adjust the weights of each sensor based on environmental conditions; S3. Employ a deep learning model to fuse multi-source data and output a heat map of traffic flow distribution and a power grid load prediction curve for the next 1-6 hours. S4. Based on the prediction results, conduct cross-cluster traffic flow guidance and flexible power allocation within the station; S5. Issue scheduling instructions and collect user response data to iteratively optimize the model and scheduling strategy.

2. The orderly charging method for a cluster of charging piles based on traffic flow optimization according to claim 1, characterized in that, In S1, the data collection is achieved through traffic monitoring equipment, power grid load sensors, charging pile status monitoring terminals, and user mobile APP. The traffic monitoring equipment collects traffic flow data with a time granularity of 15 minutes per data point.

3. The orderly charging method for a cluster of charging piles based on traffic flow optimization according to claim 1, characterized in that, In S2, the different scenarios involve deploying multiple sensors as follows: When the scenario is an intersection, a combination of video sensors, microwave sensors, and geomagnetic sensors is deployed. When the scenario is a highway, a combination of microwave sensors, RFID sensors and ultrasonic sensors should be deployed. When the scenario is a parking lot, deploy a combination of geomagnetic sensors and video sensors; The sensor weight adjustment is based on a Bayesian network and is dynamically determined according to environmental conditions such as light intensity and rainfall level.

4. The orderly charging method for a cluster of charging piles based on traffic flow optimization according to claim 1, characterized in that, In S3, the deep learning model is a dual-channel temporal deep learning model based on an attention mechanism, including a main channel, an auxiliary channel, and an attention layer; The main channel employs a long short-term memory network or a temporal convolutional network to learn long-term dependency patterns in temporal data; the auxiliary channel processes non-strict temporal features such as historical charging behavior and weather; and the attention layer dynamically calculates the importance weights of each input feature for different prediction periods.

5. The orderly charging method for a cluster of charging piles based on traffic flow optimization according to claim 4, characterized in that, The input data for the deep learning model includes traffic flow data, historical charging behavior data, and weather data; The traffic flow data includes the average traffic density, average driving speed, and congestion index of key roads in the region; the historical charging behavior data includes the historical total charging volume, number of charging orders, average charging time, and typical charging curves for each cluster; and the weather data includes temperature, humidity, precipitation type and probability, wind speed, light intensity, and extreme weather indicators.

6. The orderly charging method for a cluster of charging piles based on traffic flow optimization according to claim 4, characterized in that, The deep learning model adopts an update strategy that combines online learning with periodic retraining. Incremental learning is performed during the early morning hours when business is slow, and complete retraining is performed weekly or monthly using all data from the past three months.

7. The orderly charging method for a cluster of charging piles based on traffic flow optimization according to claim 1, characterized in that, In S4, the cross-cluster traffic guidance specifically includes the following steps: Calculate the real-time load rate of each charging pile cluster. and the global average load rate of the cluster within the region ,in, For the first Real-time charging power of each cluster For the cluster's rated capacity, The total number of clusters within the region; Dynamic discount coupons are generated for clusters with loads below the regional average. The optimal charging route, which includes travel time and economic cost, is calculated based on real-time traffic data and pushed to the user's mobile app. With V2G technology enabled, vehicles supply power to the grid in reverse. Static peak hours are defined as 07:00-10:00 and 17:00-21:00, while static off-peak hours are all other times excluding the static peak hours. The initial default value for the compensation coefficient is set to: static peak hours. r peak =1.35, static off-peak hours r o ff =1.15, where the compensation coefficient is determined by a mapping function. Automatic adjustment; By improving LightGBM, the charging demand of each charging station in time slice t is predicted to obtain D(t), and the prediction error distribution is established using kernel density estimation. A scarcity index is constructed using short-term load forecasting and error estimation. ,in, For the available grid capacity of the cluster, The risk adjustment factor is 1.64-1.

96. Normalization process is performed to obtain ,pass Determine if a dynamic peak is triggered.

8. The orderly charging method for a cluster of charging piles based on traffic flow optimization according to claim 7, characterized in that, The upper limit of instantaneous active power of the vehicle's reverse power supply. ≤5kW, with a default SOC lower limit of 20%, vehicles are prohibited from supplying power to the grid in reverse when below this threshold; certified bidirectional metering devices are used, with a metering resolution of no less than 1 minute and an energy accuracy of no less than 0.01kWh. Metering data includes the pile ID and timestamp and is digitally signed, and data records are kept for at least 3 years; settlement adopts the default next-day batch settlement or optional small-amount real-time settlement mode. In the event of an anomaly or communication failure, compensation is temporarily frozen and the settlement result is corrected after verification.

9. The orderly charging method for a cluster of charging piles based on traffic flow optimization according to claim 1, characterized in that, In S4, the flexible power allocation within the station includes the following specific steps: According to the formula via the battery management system Calculate battery health, where, For the battery at all times Real-time available capacity, This refers to the initial rated capacity of the battery. Battery grades are classified according to State of Health (SOH) and corresponding charging modes: When SOH ≥ 80%, fast charging mode is activated, with a maximum charging power of [value missing]. , ; When 60%≤SOH<80%, the conventional charging mode is used, and the power is... , ; When SOH < 60%, current-limited slow charging is implemented, with the lower limit of charging power being [value missing]. , ; Introducing power allocation factor Based on the real-time available power of the cluster Allocate charging power according to SOH priority to ensure ≤ Where M represents the number of vehicles waiting to be charged at the station. For the first The car at any time The charging power.

10. The orderly charging method for a cluster of charging piles based on traffic flow optimization according to claim 1, characterized in that, In S5, the issuance of scheduling instructions and collection of user response data are achieved through a distributed system consisting of a cloud server, a charging pile controller, and a user mobile APP. The cloud server sends scheduling instructions to the charging pile controller via the MQTT protocol, and the controller parses and executes the instructions before feeding back the status. The user's mobile app synchronizes information with the cloud server via a WebSocket long connection, and refreshes the interface in real time based on JSON data with animated prompts. The user's mobile APP has multiple data collection points to collect user operation behavior and feedback data. The data is cleaned and preprocessed before being stored. Machine learning algorithms are used to analyze and mine the data, and the dynamic prediction model and scheduling strategy are updated periodically.

Images