A cloud platform-based charging pile remote regulation and control method and system
By constructing a two-dimensional model of user behavior and power grid status, and combining multi-timescale optimization decision-making and edge adaptive execution, the problems of delay and inaccuracy in remote control of charging piles are solved, and deep collaboration and security between the charging process and the power grid are achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHANGRAO FANCE INTELLIGENT TECHNOLOGY CO LTD
- Filing Date
- 2026-05-06
- Publication Date
- 2026-06-05
Smart Images
Figure CN122143720A_ABST
Abstract
Description
Technical Field
[0001] This invention proposes a remote control method and system for charging piles based on a cloud platform, belonging to the field of new energy infrastructure and smart grid technology. Background Technology
[0002] With the widespread adoption of electric vehicles, the number of charging stations, as a key infrastructure for their energy replenishment, is increasing daily. However, current mainstream charging station cloud platforms, such as TELD, Star Charge, and State Grid eCharge, have many limitations in their remote control functions.
[0003] In terms of control operations, most are basic remote start / stop, only supporting remote start or stop charging via an app, or forced power cut-off via the platform. These are single-point operations lacking coordination. Regarding power limiting, only simple coarse-grained peak shaving is performed based on preset thresholds, without comprehensively considering user preferences, battery status, electricity price signals, and grid topology. Control rules are also mostly offline and statically configured; rules such as off-peak charging at night require manual setting and cannot dynamically respond to real-time grid conditions. Furthermore, all calculations are centralized in the cloud, resulting in high latency in cloud-based decision-making and a second-level delay in sending control commands to the charging piles, making it difficult to cope with rapid fluctuations in the distribution network.
[0004] In addition, most manufacturers' promotions remain at a general level such as "intelligent scheduling" and "orderly charging," and no one has deeply integrated user behavior profiling, power distribution network flow calculation, multi-time scale optimization, and edge adaptive execution into a unified cloud-edge collaborative architecture. Summary of the Invention
[0005] This invention provides a method and system for remote control of charging piles based on a cloud platform, in order to solve the problems mentioned in the background art above:
[0006] This invention proposes a remote control method for charging piles based on a cloud platform, the method comprising:
[0007] S1. Collect historical charging data of charging pile users and historical power grid operation data of the area to generate user charging raw dataset and power grid raw operation dataset; and use data mining algorithms to construct a two-dimensional model of user behavior and power grid status to generate two-dimensional model data.
[0008] S2. Based on the dual-dimensional model data, combined with the real-time acquired current power grid operating status data and user current charging demand data, perform multi-timescale optimization decision processing to obtain long-term control strategy data; and perform fusion processing to generate comprehensive control strategy data.
[0009] S3. Send control commands to the edge nodes of the charging pile through the cloud platform; after receiving the control commands, the edge nodes of the charging pile combine the local real-time sensing data of the charging pile operation status and the real-time data of the surrounding power grid to perform adaptive execution processing and generate edge execution feedback data.
[0010] S4. Continuously collect charging process data and real-time grid operation data fed back from the edge nodes of charging piles to generate charging process monitoring datasets and real-time grid operation monitoring datasets; dynamically update the two-dimensional model of user behavior and grid status to generate updated two-dimensional model data; optimize and adjust the comprehensive control strategy data to generate optimized comprehensive control strategy data.
[0011] S5. Perform charging safety assessment processing on the charging pile to generate charging safety assessment data; classify the charging safety assessment data into risk levels to obtain data of different risk levels; perform risk warning processing to generate remote control risk warning data for the charging pile; and dynamically adjust the charging process of the charging pile based on the remote control risk warning data for the charging pile.
[0012] The cloud-based remote control system for charging piles proposed in this invention includes:
[0013] One or more processors;
[0014] Memory, used to store one or more programs;
[0015] Wherein, when the one or more programs are executed by the one or more processors, the one or more processors are made to implement the method described in any one of the above.
[0016] The beneficial effects of this invention are as follows: By constructing a two-dimensional model of user behavior and power grid status, it can accurately capture user charging habits and real-time power grid operation characteristics, providing detailed and comprehensive data support for the regulation of charging piles. This effectively avoids the inaccurate regulation caused by the single data dimension in traditional methods, improving the scientific and rational nature of charging regulation decisions. Utilizing multi-timescale optimization decision-making, and combining data from different time dimensions to generate comprehensive regulation strategies, it can meet users' short-term charging needs while also considering the medium-term operation plan and long-term stable development of the power grid. This achieves deep synergy between the charging process and power grid operation, reducing the impact on the power grid caused by unreasonable distribution of charging load and ensuring the stable operation of the power grid. The charging pile edge nodes adaptively execute the comprehensive regulation strategy, flexibly adjusting the charging process based on local real-time data, reducing the impact on cloud-based aggregation. The delayed decision-making process enables rapid response to real-time changes in the power grid, avoiding untimely control issues caused by command transmission delays and enhancing the real-time performance and flexibility of charging control. Continuous data collection and dynamic updates to the two-dimensional model optimize the comprehensive control strategy, forming a self-learning and self-improving control loop. This loop continuously adjusts control methods based on changes in user behavior and power grid conditions, reducing control failures caused by model and strategy rigidity and improving the adaptability and long-term effectiveness of remote charging pile control. Safety assessments and risk warnings based on the optimized comprehensive control strategy can identify potential charging safety hazards in advance and take timely dynamic adjustment measures, reducing the probability of safety accidents during charging and preventing greater losses due to untimely handling of safety risks, thus enhancing the safety and reliability of the charging process. Attached Figure Description
[0017] Figure 1 This is a diagram illustrating the steps of the method described in this invention;
[0018] Figure 2 This is a block diagram of the system architecture described in this invention;
[0019] Figure 3 This is a diagram of the cloud-edge collaborative regulation mechanism described in this invention;
[0020] Figure 4 This is the closed-loop diagram of safety assessment and risk warning described in this invention. Detailed Implementation
[0021] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.
[0022] One embodiment of the present invention, such as Figure 1 and Figure 2 As shown, a method for remote control of charging piles based on a cloud platform includes:
[0023] S1. Collect historical charging data of charging pile users and historical power grid operation data of the area to generate user charging raw dataset and power grid raw operation dataset; Based on user charging raw dataset and power grid raw operation dataset, use data mining algorithms to construct a two-dimensional model of user behavior and power grid status, and generate two-dimensional model data.
[0024] S2. Based on the two-dimensional model data, combined with the real-time acquired current power grid operating status data and user current charging demand data, multi-timescale optimization decision processing is performed to obtain preliminary control strategy data for the short time scale, medium-term control strategy data for the medium time scale, and long-term control strategy data for the long time scale. The preliminary control strategy data, medium-term control strategy data, and long-term control strategy data are then fused to generate comprehensive control strategy data.
[0025] S3. Based on the comprehensive control strategy data, control commands are sent to the charging pile edge nodes through the cloud platform. After receiving the control commands, the charging pile edge nodes combine the local real-time sensing data of the charging pile's operating status and the real-time data of the surrounding power grid to perform adaptive execution processing and generate edge execution feedback data. The edge execution feedback data is verified to ensure that the control commands are executed accurately.
[0026] S4. Continuously collect charging process data and real-time power grid operation data fed back by the edge nodes of charging piles through the cloud platform to generate charging process monitoring dataset and power grid real-time operation monitoring dataset; dynamically update the two-dimensional model of user behavior and power grid status based on the charging process monitoring dataset and power grid real-time operation monitoring dataset to generate updated two-dimensional model data; optimize and adjust the comprehensive control strategy data using the updated two-dimensional model data to generate optimized comprehensive control strategy data.
[0027] S5. Based on the optimized comprehensive control strategy data, perform charging safety assessment processing for charging piles to generate charging safety assessment data; classify the charging safety assessment data into risk levels to obtain data of different risk levels; based on the data of different risk levels, perform risk warning processing to generate risk warning data for remote control of charging piles; and dynamically adjust the charging process of charging piles according to the risk warning data for remote control of charging piles.
[0028] The working principle and effects of the above technical solution are as follows: By combining a two-dimensional model with multi-timescale optimization decision-making, the regulation of charging piles is made more aligned with the actual needs of users and the operating status of the power grid, significantly improving the accuracy of regulation. This ensures both the timeliness and convenience of charging for users and avoids peak overload of the power grid, achieving a dynamic balance between charging demand and power grid capacity. Dynamic model updates and iterative optimization of strategies enhance the adaptability to changes in user behavior and power grid fluctuations, reducing regulation deviations caused by fixed strategies and effectively reducing power grid operating energy consumption and ineffective operating costs of charging piles. The adaptive execution and feedback verification mechanism of edge nodes improves the accuracy of command execution, avoids equipment failures or charging interruptions caused by regulation errors, and ensures the stability of the charging process. Safety assessment and graded early warning processing proactively avoid charging safety risks and reduce the probability of safety accidents, making charging more reassuring for users and strengthening the defense for stable power grid operation, achieving safe, economical, and efficient coordinated regulation of charging.
[0029] In one embodiment of the present invention, S1 includes:
[0030] S11. Based on the charging pile regulation requirements, determine the user's historical charging data collection dimensions and the regional power grid's historical operation data collection dimensions. The historical charging data collection dimensions include user ID, single charging duration, charging amount, charging time period, charging frequency, and charging pile number. The regional power grid's historical operation data collection dimensions include power grid voltage, current, real-time load, power supply capacity, fault records, and time-limited power fluctuations. Define the data collection time range, which is the past 12 months, covering peak-valley electricity price cycles and seasonal electricity consumption differences, and generate a data collection dimension list.
[0031] S12. Extract user charging history records through the local storage module of the charging pile, obtain regional power grid operation logs through the power grid dispatch system interface, and supplement charging reservation and cancellation data synchronously through the user charging APP. Use a distributed data acquisition framework (such as Flink) to realize real-time aggregation of multi-source data. After removing duplicate collection items, generate user charging raw dataset and power grid raw operation dataset respectively.
[0032] S13. Preprocess the user charging raw dataset, including cleaning (filling in missing charging power data and removing abnormal ultra-short duration charging records) and standardization (unifying the charging time unit to minutes and the power unit to kWh); denoise the original power grid operation dataset (filtering instantaneous power grid fluctuation data) and perform spatiotemporal alignment (partitioning and matching the power grid data according to the region where the charging pile is located) to generate a preprocessed user and power grid related dataset.
[0033] S14. Based on the preprocessed user and power grid related dataset, a combination of data mining algorithms is used (association rule algorithm to mine the relationship between user charging time and power grid load, BP neural network to build user charging demand prediction sub-model, LSTM algorithm to build power grid load prediction sub-model) to form a two-dimensional model of user behavior and power grid status; the model accuracy is verified using a test set (accounting for 20% of the total data) (for example, the user demand prediction error is required to be ≤8% and the power grid load prediction error is required to be ≤5%), and after correcting the model parameters, two-dimensional model data is generated.
[0034] The working principle and effects of the above technical solution are as follows: It accurately determines the data collection dimensions for user charging and grid operation, combining nearly 12 months of collection time to cover peak-valley electricity price cycles and seasonal differences, making basic data collection more aligned with actual control scenarios and significantly improving the targeting of data collection. By aggregating data from multiple sources and removing duplicates, it avoids data omissions or redundancy, enhancing the integrity of the original dataset. The preprocessing stages, including cleaning, standardization, noise reduction, and spatiotemporal alignment, reduce outliers and biases in the data, significantly improving data quality and providing a reliable foundation for modeling. The algorithm combination constructs a two-dimensional model, which is then validated and corrected using a test set. This accurately uncovers the correlation between charging behavior and grid load while controlling prediction errors within a reasonable range, improving model reliability and preventing deviations in subsequent control due to model inaccuracies, thus providing solid support for overall control.
[0035] In one embodiment of the present invention, S14 includes:
[0036] Based on the preprocessed user and power grid related dataset, the positioning of the two major sub-models, user charging demand and power grid load prediction, is determined. The parameters of the association rules (minimum support of 10%), BP neural network (3 hidden layers), and LSTM (time step 24) algorithm are configured to generate the sub-model algorithm configuration scheme.
[0037] Based on the generated sub-model algorithm configuration scheme, the association rules are used to mine the relationship between charging time period and power grid load. The user demand prediction sub-model is trained by BP neural network and the power grid load prediction sub-model is trained by LSTM, generating the initial draft of two core sub-models.
[0038] The weighted fusion method (45% for user sub-model and 55% for power grid sub-model) is used to integrate the sub-model outputs, eliminate prediction conflicts, and generate a draft of the two-dimensional model.
[0039] The model is validated using a 20% test set. If the user demand prediction error is greater than 8% or the power grid load prediction error is greater than 5%, the parameters are adjusted (e.g., the LSTM learning rate is reduced from 0.01 to 0.005) until the error meets the standard, and a valid two-dimensional model is generated. The model structure and parameters are then solidified, packaged into a callable module, and two-dimensional model data is generated.
[0040] The working principle and effects of the above technical solution are as follows: Precise configuration of core parameters for each algorithm provides a clear execution basis for sub-model training, avoiding the low training efficiency caused by blind parameter tuning and significantly improving the targeting of sub-model construction. For two core scenarios—user needs and grid load—different algorithms are used to train sub-models separately. Association rule mining explores the correlation between the two, and BP neural networks and LSTM accurately predict their respective trends, making the single-model function more focused and enhancing the accuracy of sub-model predictions. A weighted fusion method is used to integrate the sub-model outputs. By reasonably allocating weights to balance the two types of prediction results, prediction conflicts caused by a single sub-model dominating are effectively eliminated, making the two-dimensional model more aligned with actual control needs. The test set verification and parameter iteration adjustment mechanism strictly controls prediction errors within the standard, significantly improving model reliability and preventing deviations in subsequent control due to model inaccuracies. Finally, the structure is solidified and encapsulated into a callable module, reducing adaptation costs during subsequent use, enhancing the model's practicality, and providing stable and reliable core support for the entire remote control process.
[0041] One embodiment of the present invention, such as Figure 3 As shown, S2 includes:
[0042] S21. Collect current power grid operating status data and user current charging demand data through the cloud platform real-time data interface. The current operating status data includes real-time total load, voltage stability of each area, and backup power supply capacity. The user current charging demand data includes the number of immediate charging requests, scheduled charging time periods, and charging power demand. Compare the real-time data with historical features in the two-dimensional model data, mark the data differences (e.g., the current power grid load is 12% higher than the same period in history), and generate a real-time and historical fusion dataset.
[0043] S22. Based on the needs of charging pile control scenarios, divide the time scale into: short time scale (15 minutes to 2 hours, to deal with instantaneous load fluctuations in the power grid), medium time scale (2 to 12 hours, to schedule the allocation of charging pile resources within the day), and long time scale (12 to 24 hours, to plan the matching of charging load and power grid supply for the next day); and set optimization objectives for each scale (short-term objective: maintain the power grid voltage stable within ±2% range; medium-term objective: improve the utilization rate of charging piles to over 85%; long-term objective: reduce the peak-valley load difference of the power grid by 30%), and generate time scale optimization objective schemes.
[0044] S23. Based on the real-time and historical fusion datasets and the time-scale optimization target scheme, corresponding algorithm calculation strategies are adopted: for short time scales, particle swarm optimization algorithm is used to generate preliminary control strategy data for adjusting the power of charging piles according to the instantaneous load of the power grid (e.g., when the load exceeds the threshold, the power of a single pile is reduced to below 5kW); for medium time scales, linear programming algorithm is used to generate medium-term control strategy data for allocating charging pile charging time periods according to regions (e.g., charging piles in commercial areas prioritize charging demand from 10:00 to 14:00); for long time scales, genetic algorithm is used to generate long-term control strategy data for coordinating the start and stop of charging piles with power grid supply (e.g., starting backup charging piles 3 hours before the morning peak of the next day).
[0045] S24. Establish strategy fusion rules (based on long-term strategy as the basic framework, medium-term strategy to adjust resource allocation details, and short-term strategy to correct sudden deviations). Use a weighted allocation method (40% for long-term, 35% for medium-term, and 25% for short-term) to integrate the three types of strategy data and eliminate conflict items (for example, when the charging pile start time in the long-term plan conflicts with the medium-term time allocation, the grid power supply capacity is adjusted as the priority). Simulate the strategy execution effect through a grid simulation platform (to verify whether it meets safety and economic indicators). After correcting unreasonable items, generate comprehensive control strategy data.
[0046] The working principle and effects of the above technical solution are as follows: Real-time and historical data are integrated and differences are labeled, providing comprehensive data support for strategy formulation and avoiding the one-sidedness of relying solely on real-time or historical data, thus improving the reliability of the data foundation. Dividing the data into short, medium, and long time scales and setting differentiated objectives allows for precise matching of different scenario needs in regulation. This includes short-term response to fluctuations, medium-term resource scheduling, and long-term planning and matching, enhancing the targeted nature of regulation. Each time scale is paired with an adaptation algorithm: particle swarm optimization adapts for rapid short-term adjustments, linear programming optimizes intraday resource allocation, and genetic algorithms support long-term overall planning, making strategies at each scale more aligned with scenario characteristics, improving strategy accuracy, and reducing the disconnect between strategies and actual needs. During strategy fusion, strategies are integrated according to weights and conflicts are eliminated, followed by simulation verification and correction to ensure that the comprehensive strategy takes into account both long-term planning frameworks and short-term sudden adjustments, avoiding execution chaos caused by conflicts between strategies at different scales. This design not only ensures grid voltage stability and reduces peak-valley load differences but also improves charging pile utilization, providing reliable strategy support for subsequent precise regulation.
[0047] In one embodiment of the present invention, S3 includes:
[0048] S31, the cloud platform parses the comprehensive control strategy data into machine-executable instructions (e.g., charging pile ID: 001, execution time: 08:00-10:00, charging power limit: 6kW), encapsulates them in JSON format and transmits them through an SSL / TLS encrypted channel, and distributes them to the corresponding charging pile edge nodes in groups according to the region where the charging piles are located (each edge node manages 10-20 charging piles), while recording the instruction distribution time, node ID, instruction content, and generating instruction distribution logs;
[0049] S32. After receiving the command, the charging pile edge node starts the local sensor module (collecting the charging pile's real-time temperature, charging current, and charging interface status) and the surrounding power grid data acquisition unit (acquiring the real-time load and voltage values of the power grid in the node's coverage area). It compares the locally collected charging pile operating status data and the surrounding power grid real-time data with the parameters in the issued command (for example, comparing the command power limit of 6kW with the local power grid's current allowable power of 5.8kW), removes abnormal items with data deviations exceeding 10%, and generates the edge node data fusion result.
[0050] S33. Based on the data fusion results of edge nodes, determine whether the command needs to be adjusted (e.g., when the local power grid allows power to be lower than the command limit, automatically lower the power limit to 5.8kW), use fuzzy control algorithm to generate specific execution operations (e.g., adjust the output of the charging pile PWM module, start the overload protection mechanism), collect the changes in the charging pile operating parameters in real time after the operation is executed (e.g., the current stabilizes at 26A after adjustment), integrate the execution time, adjustment parameters, and actual operating status to generate edge execution feedback data;
[0051] S34. The cloud platform receives edge execution feedback data and compares it with the expected results in the comprehensive control strategy data (e.g., expected charging power of 6kW, actual execution of 5.8kW, verifying whether the error is within the allowable range of ±5%). If the error exceeds the range, the platform analyzes the cause of the deviation (e.g., a sudden increase in grid load) and issues a correction instruction (e.g., further reducing the power to 5.5kW). If the error meets the requirements, the platform generates a verification pass report and stores the feedback data and verification results together to ensure accurate execution of the control instructions.
[0052] The working principle and effects of the above technical solution are as follows: The cloud platform parses the strategy into machine instructions and encrypts and distributes them. Combined with regional grouping and log recording, this avoids the risk of data leakage during instruction transmission and allows for clear traceability of instruction distribution, significantly improving the security and traceability of instruction transmission. The grouping mode of managing 10-20 devices per edge node also reduces redundancy overhead in instruction distribution and improves distribution efficiency. After receiving instructions, the edge nodes collect local operating and grid data, compare instruction parameters, and eliminate anomalies, making the execution basis more aligned with the actual situation on site. This avoids the problem of simply copying cloud instructions while ignoring local operating conditions, enhancing the reliability of data support. Instructions are adaptively adjusted based on the fusion results, such as reducing the charging pile power according to the grid's allowable power, and then using fuzzy control to generate execution operations, improving the flexibility of instruction execution and effectively avoiding equipment failures caused by overload. The cloud platform's feedback verification mechanism compares the actual execution with the expected results to correct deviations, ensuring that instruction execution errors are controlled within a reasonable range and reducing the risk of grid fluctuations caused by execution deviations. This closed-loop design of issuing, executing, providing feedback, and correcting not only ensures that control instructions are accurately implemented, but also makes the execution process more adaptable to dynamic changes on site, thus providing solid support at the execution level for the overall control effect.
[0053] In one embodiment of the present invention, step S31 includes:
[0054] Based on comprehensive control strategy data, core parameters of charging piles are extracted, including ID, execution period, and power limit. These parameters are converted into machine-recognizable instruction codes (e.g., ID001|T0800-1000|P6kW) to generate a draft instruction.
[0055] The initial draft of the instructions is standardized and packaged in JSON format, and then transmitted and encrypted using the SSL / TLS protocol to form an encrypted instruction data packet, ensuring the security of data transmission.
[0056] The charging piles are grouped according to their administrative region and power grid zone. Encrypted instruction data packets are then sent in batches to the corresponding charging pile edge nodes (each node is associated with 10-20 charging piles), and a sending and execution record is generated.
[0057] Integrate the information in the execution log (instruction content, issuance time, target node ID, transmission status, etc.), archive it by timestamp + region index, and generate instruction issuance log.
[0058] The working principle and effects of the above technical solution are as follows: Core parameters are extracted and converted into machine-readable code, allowing instructions to be directly interpreted by the device, avoiding execution failures caused by inconsistent parameter formats and improving the accuracy of instruction generation. JSON encapsulation ensures a standardized instruction format, and combined with SSL / TLS encrypted transmission, effectively preventing the risk of data leakage or tampering, enhancing the security of instruction transmission. Grouping by administrative region and power grid partition, combined with a batch distribution mode managing 10-20 devices per node, eliminates the tediousness of distributing instructions to each device individually, significantly improving distribution efficiency and preventing the problem of instructions being mistakenly sent to nodes in other regions. Integrating distribution information to generate logs based on timestamps and regional indexes allows for rapid location of key information such as the distribution time and target node of specific instructions should execution anomalies occur, reducing the time cost of troubleshooting and avoiding the embarrassment of having no evidence to trace, providing reliable support for the entire process of instruction execution control.
[0059] In one embodiment of the present invention, step S4 includes:
[0060] S41. The cloud platform continuously receives charging process data (including the amount of electricity completed in a single charge, the number of charging interruptions, and charging pile fault codes) uploaded by each charging pile edge node via the WebSocket real-time transmission protocol, as well as real-time power grid operation data (including regional load change curves, power supply energy consumption data, and voltage fluctuation records) pushed by the power grid dispatching system. It uses a time-series database (such as InfluxDB) to store data indexed by timestamp + device ID, automatically cleans up redundant data (retaining high-frequency data from the past 3 months and low-frequency data from the past 12 months), and generates charging process monitoring datasets and power grid real-time operation monitoring datasets.
[0061] S42. Extract user behavior features (e.g., peak shift during charging periods, changes in single charging power, and adjustments to user charging priorities) from the charging process monitoring dataset; extract grid status features (e.g., extended peak load duration, fluctuations in power supply energy consumption, and seasonal energy consumption differences) from the real-time grid operation monitoring dataset; analyze feature change trends using the sliding window method (e.g., the peak charging time for users in commercial areas has shifted from 18:00 to 17:30 in the past month); generate a data feature trend analysis report; and mark the deviations from the two-dimensional model prediction results (e.g., the model predicts the peak load at 19:00, but it actually occurs at 18:30).
[0062] S43. Based on the data feature trend analysis report, update the two-dimensional model using an incremental learning algorithm: For the user behavior sub-model, supplement new charging period feature data and adjust the demand forecast weight (e.g., increase the demand forecast weight for the 17:30-18:30 period by 15%); for the power grid status sub-model, incorporate new load trend data and optimize the load forecast formula; test the updated model with the latest monitoring data (last 7 days), requiring the prediction accuracy to be improved by ≥5% compared to the original model (e.g., from 85% to 91%), and generate the updated two-dimensional model data after the target is met.
[0063] S44. Input the updated two-dimensional model data into the multi-timescale optimization framework and adjust the strategy parameters at each scale: In the short-term strategy, lower the grid load threshold from 80% to 78% (to cope with the load growth trend); in the medium-term strategy, reallocate the charging time slots in commercial areas (increase the charging quota for the 17:30-18:30 time slot); in the long-term strategy, optimize the next day's power supply plan (start the grid reserve capacity 1 hour in advance); verify the economic efficiency (e.g., energy consumption reduction of 6%) and fairness (user charging demand satisfaction rate improvement of 9%) of the optimized strategy through a simulation platform, and generate the optimized comprehensive control strategy data.
[0064] The working principle and effects of the above technical solution are as follows: WebSocket real-time transmission enables timely convergence of charging process and grid operation data. Combined with a time-series database indexed by timestamp and device ID, and automatic cleanup of redundant data, this ensures the timeliness and integrity of the dataset while preventing invalid data from consuming storage resources, thus improving data management efficiency. After extracting user behavior and grid status characteristics, a sliding window method is used to analyze trends and label deviations. This accurately captures changes such as charging peak migration and load fluctuations, avoiding prediction gaps caused by long-term use of old features and enhancing the perception of actual operational changes. When updating the model using the incremental learning algorithm, new features are added and weights are adjusted in a targeted manner, avoiding full retraining. This improves model iteration efficiency and ensures accuracy improvement through 7 days of data testing, reducing prediction bias. The updated model is then incorporated into the optimization framework to adjust the strategy, and simulations verify its economic efficiency and fairness. This allows the strategy to adapt to the latest operational trends, avoiding the problem of fixed strategies being insufficient to cope with changes. This reduces grid energy consumption and improves the satisfaction rate of user charging needs, providing core support for dynamic adaptation for continuous and precise regulation.
[0065] In one embodiment of the present invention, S42 includes:
[0066] S421. Based on the charging process monitoring dataset and the real-time power grid operation monitoring dataset, determine the specific extraction indicators for user behavior characteristics (peak charging time, single charge, etc.) and power grid status characteristics (peak load duration, energy consumption fluctuation, etc.), and generate a list of feature extraction indicators.
[0067] S422. Based on the indicator list, extract user behavior feature data from the charging process monitoring dataset and extract power grid status feature data from the real-time power grid operation monitoring dataset, remove invalid feature items (such as duplicate time period records), and generate an initial feature dataset.
[0068] S423. Use the sliding window method (window duration set to 7 days) to perform trend calculation on the initial feature dataset, analyze the feature change patterns (e.g., peak charging migration in commercial areas, load fluctuation amplitude changes), and generate feature trend data;
[0069] S424. Compare the characteristic trend data with the two-dimensional model prediction results, mark the deviation items (such as the difference between the predicted and actual peak load time), integrate the trend analysis results and deviation information, archive them according to the structure of feature type + trend description + deviation description, and generate a data feature trend analysis report.
[0070] The working principle and effects of the above technical solution are as follows: First, the feature extraction indicators of user behavior and power grid status are clearly defined and listed, providing a clear direction for subsequent extraction work and avoiding the omission of key features or the redundancy of invalid features caused by blindly selecting indicators, thus improving the targeting of feature extraction. Based on the list, features are extracted from the dataset and invalid items such as duplicate records are removed, reducing the interference of impurity data on subsequent analysis and significantly improving the quality of the initial feature dataset. The 7-day sliding window method is used to analyze trends, which can dynamically capture the patterns of charging peak migration and load fluctuation amplitude changes, avoiding gradual feature changes that are difficult to detect in static analysis and enhancing the perception of the operational situation. The trend data is compared with the model prediction results to mark the deviations, and then integrated into a report according to the standardized structure, making the model deviation points (such as the time difference between load peak prediction and actual load) clearly visible, avoiding the problem of lack of accurate basis for subsequent model updates, and providing reliable analytical support for model iteration and optimization.
[0071] In one embodiment of the present invention, S423 includes:
[0072] Based on the time span and data density of the initial feature dataset, determine the core parameters of the sliding window (window duration of 7 days and step size of 1 day), clarify the trend calculation indicators (such as feature mean, peak occurrence time, and fluctuation variance), and generate a window configuration scheme.
[0073] Based on the window configuration scheme, the initial feature dataset is slidably segmented in chronological order to obtain continuous window feature subsets (e.g., days 1-7, days 2-8, etc.). Invalid subsets with less than 80% data volume within the window are removed to generate windowed feature subsets.
[0074] For each window feature subset, calculate the trend index of user behavior features (e.g., average charging peak time) and power grid state features (e.g., load fluctuation variance), record the change magnitude of features within each window, and generate window trend calculation results;
[0075] All window trend calculation results are concatenated in chronological order to analyze the continuous change pattern of features (e.g., charging peaks advance week by week), the calculation results of abnormal fluctuation windows are removed, and the data is integrated into structured data to generate feature trend data.
[0076] The working principle and effects of the above technical solution are as follows: By combining the span and density of the initial feature dataset to determine window parameters and trend indicators, a clear configuration scheme is formed, avoiding analytical biases caused by blindly setting window duration and step size, and improving the targeting of trend analysis. After sliding the data according to the configuration, invalid subsets with less than 80% data volume are actively removed, reducing the interference of incomplete data on the analysis results and enhancing the reliability of the windowed feature subsets. Trend indicators are accurately calculated and the magnitude of change is recorded for each subset, clearly presenting the patterns of charging peak mean, load fluctuation variance, etc., within a single window, improving the accuracy of capturing local trends. When concatenating all window results, abnormal fluctuation items are removed, clearly extracting continuous change patterns such as the weekly advance of charging peaks, avoiding trend misjudgments caused by isolated window analysis or abnormal data, and providing accurate and realistic trend basis for subsequent two-dimensional model updates.
[0077] One embodiment of the present invention, such as Figure 4 As shown, S5 includes:
[0078] S51. Based on the optimized comprehensive control strategy data, determine the core dimensions of charging safety assessment: charging pile equipment safety (including temperature threshold, current overload count, insulation resistance value, and charging interface aging degree), power grid operation safety (including voltage deviation range, frequency stability, short-circuit current tolerance value, and regional load overload risk), and user electricity safety (including charging interface temperature, leakage protection response time, and charging interruption warning timeliness). Set safety thresholds for each indicator (e.g., charging pile operating temperature ≤ 60℃, voltage deviation ≤ ±5%, leakage protection response time ≤ 0.1s), and use the analytic hierarchy process (AHP) to determine the indicator weights (equipment safety 40%, power grid safety 35%, user safety 25%), generating a safety assessment indicator system.
[0079] S52. The cloud platform collects real-time operating data of charging pile equipment (temperature, current, insulation resistance), power grid safety monitoring data (voltage, frequency, short-circuit current), and user power safety data (interface temperature, leakage protection status). It compares the data with the safety assessment index system and uses the fuzzy comprehensive evaluation method to calculate the score of each index (0-100 points, with 80 points or above considered safe). Then, it calculates the comprehensive safety score based on the index weights and generates charging safety assessment data (for example, the comprehensive safety score of charging piles in a certain area is 72 points).
[0080] S53. Set risk level classification standards: A comprehensive safety score ≥80 is low risk (no safety hazards, normal operation), 60-79 is medium risk (local hazards exist, adjustment is required), and <60 is high risk (serious hazards exist, emergency handling is required); determine the current risk level based on charging safety assessment data (e.g., 72 points corresponds to medium risk), and generate early warning information for different levels: push routine safety reminders to the operator for low risk; push local parameter adjustment suggestions (e.g., reduce the power of a charging pile to 4kW) to the edge node and the operator for medium risk; push emergency shutdown warnings to the edge node, the operator, and the user's APP for high risk, and generate remote control risk warning data for charging piles;
[0081] S54. Dynamic adjustments based on risk warning data: In low-risk scenarios, maintain the optimized comprehensive control strategy; in medium-risk scenarios, edge nodes adjust charging pile operating parameters according to warning suggestions (e.g., reduce power, extend heat dissipation interval), and the cloud platform simultaneously optimizes the allocation of regional power grid resources (e.g., increase the reserve capacity of the regional power grid); in high-risk scenarios, edge nodes immediately execute shutdown commands, the cloud platform coordinates nearby idle charging piles to meet user demand, and notifies maintenance personnel to conduct on-site repairs; after adjustment, continuously monitor safety assessment data (updated every 5 minutes) until the risk level drops to low risk.
[0082] The working principle and effects of the above technical solution are as follows: A three-dimensional assessment system is established from equipment and power grid to user electricity consumption. Coupled with clearly defined thresholds and weight allocations, this ensures comprehensive safety assessment without blind spots, avoiding the problem of overlooking potential hazards in single-dimensional assessments and significantly improving the comprehensiveness of the assessment. Real-time collection of multiple types of safety data and calculation of scores using fuzzy comprehensive evaluation transforms abstract safety states into intuitive scores, enhancing the accuracy of the assessment. Three levels of risk are categorized according to the scores, with differentiated early warnings—low-risk alerts, medium-risk suggestions for adjustments, and high-risk emergency shutdowns—making early warnings more targeted and reducing the possibility of missed safety hazards. A dynamic adjustment mechanism implements precise measures based on risk level: adjusting parameters for medium-risk and shutting down and coordinating alternative charging piles for high-risk situations. This not only promptly curbs the escalation of risks and avoids equipment failures or power grid accidents but also minimizes user charging interruptions, balancing safety and user experience. Continuous monitoring and updating of scores until the risk decreases creates a closed-loop safety management system.
[0083] One embodiment of the present invention provides a cloud-based remote control system for charging piles, comprising:
[0084] One or more processors;
[0085] Memory, used to store one or more programs;
[0086] Wherein, when the one or more programs are executed by the one or more processors, the one or more processors are made to implement the method described in any one of the above.
[0087] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.
Claims
1. A method for remote control of charging piles based on a cloud platform, characterized in that, The method includes: S1. Collect historical charging data of charging pile users and historical power grid operation data of the area to generate user charging raw dataset and power grid raw operation dataset; and use data mining algorithms to construct a two-dimensional model of user behavior and power grid status to generate two-dimensional model data. S2. Based on the dual-dimensional model data, combined with the real-time acquired current power grid operating status data and user current charging demand data, perform multi-timescale optimization decision processing to obtain long-term control strategy data; and perform fusion processing to generate comprehensive control strategy data. S3. Send control commands to the edge nodes of the charging pile through the cloud platform; after receiving the control commands, the edge nodes of the charging pile combine the local real-time sensing data of the charging pile operation status and the real-time data of the surrounding power grid to perform adaptive execution processing and generate edge execution feedback data. S4. Continuously collect charging process data and real-time grid operation data fed back from the edge nodes of charging piles to generate charging process monitoring datasets and real-time grid operation monitoring datasets; dynamically update the two-dimensional model of user behavior and grid status to generate updated two-dimensional model data; optimize and adjust the comprehensive control strategy data to generate optimized comprehensive control strategy data. S5. Perform charging safety assessment processing on the charging pile to generate charging safety assessment data; classify the charging safety assessment data into risk levels to obtain data of different risk levels; perform risk warning processing to generate remote control risk warning data for the charging pile; and dynamically adjust the charging process of the charging pile based on the remote control risk warning data for the charging pile.
2. The remote control method for charging piles based on a cloud platform according to claim 1, characterized in that, S1 includes: S11. Based on the charging pile regulation requirements, determine the user's historical charging data collection dimensions and the regional power grid's historical operation data collection dimensions, define the data collection time range, and generate a data collection dimension list. S12. Extract user charging history records through the local storage module of the charging pile, obtain regional power grid operation logs through the power grid dispatching system interface, and supplement charging reservation and cancellation data synchronously through the user charging APP. Use a distributed data acquisition framework to realize real-time aggregation of multi-source data and generate user charging raw dataset and power grid raw operation dataset respectively. S13. Preprocess the original user charging dataset to generate a preprocessed user and power grid related dataset. S14. Based on the preprocessed user and power grid related dataset, a combination of data mining algorithms is used to fuse and form a two-dimensional model of user behavior and power grid status. The accuracy of the model is verified using a test set, and the model parameters are corrected to generate two-dimensional model data.
3. The remote control method for charging piles based on a cloud platform according to claim 1, characterized in that, S2 includes: S21. Through the real-time data interface of the cloud platform, collect the current operating status data of the power grid and the current charging demand data of users, compare the real-time data with the historical features in the two-dimensional model data, mark the data differences, and generate a real-time and historical fusion dataset. S22. Based on the needs of charging pile control scenarios, divide the time scale into short time scale, medium time scale, and long time scale; and set optimization targets for each scale to generate time scale optimization target schemes. S23. Based on the real-time and historical fusion dataset and time scale, optimize the target scheme and adopt the corresponding algorithm calculation strategy; S24. Establish strategy fusion rules, integrate the three types of strategy data using the weight allocation method, simulate the strategy execution effect through the power grid simulation platform, correct unreasonable items, and generate comprehensive control strategy data.
4. The remote control method for charging piles based on a cloud platform according to claim 3, characterized in that, The corresponding algorithm calculation strategy includes using a particle swarm optimization algorithm for short time scales to generate preliminary control strategy data for adjusting the power of charging piles according to the instantaneous load of the power grid; and using a linear programming algorithm for medium time scales to generate medium-term control strategy data for allocating charging time periods of charging piles by region. Genetic algorithms are used over long time scales to generate long-term control strategy data for coordinating the start-up and shutdown of charging piles with grid power supply.
5. The remote control method for charging piles based on a cloud platform according to claim 1, characterized in that, The S3 includes: S31. The cloud platform parses the comprehensive control strategy data into machine-executable instructions, encapsulates them in JSON format and transmits them through an SSL / TLS encrypted channel, and distributes them to the corresponding charging pile edge nodes in groups according to the area where the charging pile is located. At the same time, it records the instruction distribution time, node ID, instruction content and generates instruction distribution logs. S32. After receiving the instruction, the charging pile edge node starts the local sensor module and the surrounding power grid data acquisition unit, compares the locally acquired charging pile operation status data, the surrounding power grid real-time data with the parameters in the issued instruction, and generates the edge node data fusion result. S33. Based on the data fusion results of edge nodes, determine whether the instruction needs to be adjusted, use a fuzzy control algorithm to generate specific execution operations, collect changes in charging pile operating parameters in real time after the operation is executed, integrate execution time, adjustment parameters, and actual operating status, and generate edge execution feedback data. S34. The cloud platform receives edge execution feedback data and compares it with the expected results in the comprehensive control strategy data: if the error exceeds the range, it analyzes the cause of the deviation and issues an instruction to correct the plan; if the error meets the requirements, it generates a verification qualified report and stores the feedback data and verification results together.
6. The remote control method for charging piles based on a cloud platform according to claim 1, characterized in that, The S4 includes: S41. The cloud platform continuously receives charging process data uploaded by edge nodes of each charging pile and real-time power grid operation data pushed by the power grid dispatching system through the WebSocket real-time transmission protocol; it uses a time-series database to store data indexed by timestamp + device ID, and generates charging process monitoring dataset and power grid real-time operation monitoring dataset. S42. Extract user behavior features from the charging process monitoring dataset and extract power grid status features from the real-time power grid operation monitoring dataset; use the sliding window method to analyze the feature change trend, generate a data feature trend analysis report, and mark the deviation items from the two-dimensional model prediction results; S43. Based on the data feature trend analysis report, use the incremental learning algorithm to update the two-dimensional model and generate the updated two-dimensional model data. S44. Input the updated two-dimensional model data into the multi-timescale optimization framework, adjust the strategy parameters at each scale, and generate optimized comprehensive control strategy data.
7. The remote control method for charging piles based on a cloud platform according to claim 6, characterized in that, S42 includes: S421. Based on the charging process monitoring dataset and the real-time power grid operation monitoring dataset, determine the specific extraction indicators for user behavior characteristics and power grid status characteristics, and generate a list of feature extraction indicators. S422. Based on the indicator list, extract user behavior feature data from the charging process monitoring dataset and extract power grid status feature data from the real-time power grid operation monitoring dataset. Eliminate invalid feature items and generate an initial feature dataset. S423. Use the sliding window method to calculate the trend of the initial feature dataset, analyze the feature change pattern, and generate feature trend data; S424. Compare the feature trend data with the prediction results of the two-dimensional model, mark the deviation items, integrate the trend analysis results and deviation information, archive them according to the structure of feature type + trend description + deviation explanation, and generate a data feature trend analysis report.
8. The remote control method for charging piles based on a cloud platform according to claim 7, characterized in that, S423 includes: Based on the time span and data density of the initial feature dataset, the core parameters of the sliding window are determined, the trend calculation indicators are clarified, and the window configuration scheme is generated. Based on the window configuration scheme, the initial feature dataset is slid-cut in chronological order to obtain continuous window feature subsets. Invalid subsets with less than 80% data volume within the window are removed to generate windowed feature subsets. For each window feature subset, calculate the trend index of user behavior features and power grid status features, record the change magnitude of features within each window, and generate window trend calculation results; All window trend calculation results are concatenated in chronological order to analyze the continuous change pattern of features, remove the calculation results of abnormal fluctuation windows, integrate them into structured data, and generate feature trend data.
9. The remote control method for charging piles based on a cloud platform according to claim 1, characterized in that, The S5 includes: S51. Based on the optimized comprehensive control strategy data, determine the core dimensions of charging safety assessment and generate a safety assessment index system. S52. The cloud platform collects real-time data on the operation of charging pile equipment, power grid safety monitoring data, and user electricity safety data. It compares the data with the safety assessment index system, uses the fuzzy comprehensive evaluation method to calculate the score of each index, and then calculates the comprehensive safety score based on the index weight to generate charging safety assessment data. S53. Set risk level classification standards and generate risk warning data for remote control of charging piles; S54. Make dynamic adjustments based on risk warning data; after adjustment, continuously monitor safety assessment data until the risk level drops to low risk.
10. A cloud-based remote control system for charging piles, including: One or more processors; Memory, used to store one or more programs; Wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the method of any one of claims 1 to 9.