A distributed robust control method suitable for multi-peak and impact load characteristics of super charging station

Abstract

The application discloses a distributed robust control method and system suitable for multi-peak and impact load characteristics of super charging stations, and belongs to the technical field of super charging station control of electric power systems. The method comprises the following steps: collecting multi-source data by using a differentiated sampling strategy and preprocessing; decomposing the load into trend, rhythm and impact components based on variational mode decomposition, realizing load type determination and power grid state perception; performing distributed robust optimization by using an improved distributed alternating direction multiplier method, dynamically adjusting the penalty factor according to the load type and the power grid state, and integrating the min-max robust optimization to inhibit impact noise; constructing a three-level scheduling execution architecture of "station level-pile level-equipment level"; establishing a three-level safety protection mechanism of early warning, intervention and emergency and linking a PI adaptive dynamic correction. The method further comprises multi-super charging station collaborative scheduling based on an OCPP protocol and load prediction based on XGBoost, BiLSTM and an attention mechanism. The application significantly improves the operation stability of super charging stations and the collaborative efficiency of equipment.

Description

Technical Field

[0001] This invention belongs to the field of power system supercharging station control technology, specifically relating to a distributed robust control method adapted to the multi-peak and impact load characteristics of supercharging stations. Background Technology

[0002] With the increasing prevalence of electric vehicles (EVs) and the development of V2G (Vehicle-to-Grid) technology, multiple charging stations connected to the same regional power distribution network can coordinate and utilize on-site online bidirectional charging and discharging equipment and transformers to feed on-demand battery power back to the power distribution network, thereby participating in grid regulation. This technology aims to improve the power distribution network's capacity to absorb distributed energy resources, alleviate peak load pressure, and optimize the benefits for electric vehicle users.

[0003] However, with the popularization of large-scale ultra-fast charging technology, the load in supercharging stations exhibits significant "multi-peak + impact" composite characteristics: multi-peak loads are concentrated between 7-9 am (before commuting), 6-8 pm (after commuting), and midnight, with multiple supercharging piles and fast charging piles charging intensively during these periods, resulting in a load density of over 1500kW; instantaneous impact loads originate from scenarios such as the start-up of 600kW-level supercharging piles and the simultaneous start-up and shutdown of multiple piles, with millisecond-level power mutations reaching 600kW, causing sudden rises and falls in bus voltage and conflicts in equipment power distribution, seriously affecting the lifespan of charging equipment and the safe operation of the power grid.

[0004] To address the challenges of grid regulation posed by the charging load of supercharging stations, existing technologies have conducted extensive research and proposed various regulation schemes, which can be broadly categorized into three technical directions: general regulation at the distribution network level, single-station V2G control, and load smoothing and safety protection. Among these, Chinese patent CN121367202A achieves flexible regulation of ultrafast charging load through multi-source data fusion and time-series load forecasting technology, effectively improving load forecasting accuracy; Chinese patent CN120222485A employs a hierarchical collaborative optimization architecture and distributed robust control method to achieve distribution network access without capacity expansion and multi-time-scale optimization, reducing grid transformation costs; and Chinese patent CN121238640A achieves differentiated regulation of charging load by constructing a vehicle-grid integrated safety assessment model and a vehicle state perception system, improving the collaborative adaptability between vehicles and the grid. In addition, some existing technologies focus on passive energy storage compensation, which smooths out load fluctuations at supercharging stations by configuring energy storage systems, but do not consider the matching of energy storage capacity with multi-peak loads; some technologies use centralized control algorithms, which calculate and issue power adjustment commands through the distribution network dispatch center to achieve load control of supercharging stations; and some studies have proposed a single threshold safety protection mechanism, which triggers protective actions such as power limiting of supercharging piles by setting single thresholds such as transformer load rate and bus voltage.

[0005] The aforementioned existing technologies provide useful references for the field in terms of multi-source data fusion, hierarchical optimization, load smoothing, and safety assessment. However, whether it is general control at the distribution network level, single-station V2G control, or a single load smoothing or safety protection scheme, none of them can fully adapt to the unique multi-peak rhythm and instantaneous impact composite load characteristics within supercharging stations. They have not built a full-link, refined control system for these characteristics and still have many technical shortcomings. (1) Insufficient targeting of data acquisition and processing: Although existing technologies emphasize the fusion of multi-source data, their sampling strategies are often globally uniform and fail to perform differentiated high-frequency sampling for millisecond-level changes in impact loads and minute-level changes in multi-peak loads. At the same time, conventional abnormal data processing methods are prone to misjudging and removing peak data brought about by impact loads as noise, resulting in the loss of key features and failing to provide a high-fidelity data foundation for subsequent precise regulation.

[0006] (2) Poor adaptability of load feature extraction and optimization algorithms: Existing technologies mostly use wavelet transform, LSTM and other models for load forecasting, or optimize and schedule the load as a whole, failing to effectively separate and model the two physically completely different load components, "multi-peak rhythm" and "instantaneous impact". This makes the optimization algorithm parameters fixed and unable to be adaptively adjusted according to the dynamic changes of load type and power grid status. When facing impact loads, its robustness and response speed are difficult to meet the requirements.

[0007] (3) Lack of coordination between scheduling execution and security protection: Although existing technologies have proposed the concept of hierarchical or graded control, their hierarchical division (such as township-village-user) or risk classification (low-medium-high) does not match the on-site time scale (second-level / millisecond-level) control requirements focused on in this application. More importantly, existing technologies have failed to deeply couple the security protection mechanism with real-time dynamic correction, making it difficult to form a multi-level, adaptive security protection closed loop from early warning and intervention to emergency handling when the equipment faces risks such as impact and overload.

[0008] In summary, how to construct a refined closed-loop control scheme that integrates data acquisition, feature extraction, optimization decision-making, instruction execution, and safety assurance for the complex load characteristics of multi-peak rhythms and instantaneous impacts within supercharging stations is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0009] In view of this, the present invention aims to solve the technical problems of poor adaptability, slow response and insufficient robustness of the existing supercharging station load regulation method for multi-peak and impact load characteristics, and to provide a distributed robust control method and system adapted to the multi-peak and impact load characteristics of supercharging stations.

[0010] To achieve the above objectives, the present invention provides the following technical solution: A distributed robust control method adapted to the multi-peak and impact load characteristics of supercharging stations includes the following steps: S1. Dedicated collection and preprocessing of multi-source heterogeneous data within supercharging stations: For multi-peak load periods and instantaneous impact scenarios, a differentiated sampling strategy is adopted to collect data from core equipment and cross-platforms within the station; the collected data is safely transmitted, standardized, layered filtered and noise suppressed, and abnormal data is removed and missing data is filled through preprocessing and post-verification dual-layer identification; S2. Hierarchical Extraction and Status Awareness of Station Load Characteristics: Based on the variational mode decomposition algorithm, the total load within the station is decomposed into trend components, rhythmic components, and impulse components. The objective function of variational mode decomposition is:

[0011] in, To decompose the component set and center frequency set Find the minimum value. For time partial derivatives, For the Dirac function, For the first Temporal representation of each decomposed component , The first Each decomposed component and its center frequency The imaginary unit, It is the frequency shift factor; S3. Improved distributed robust control optimization of dADMM: Construct a multi-objective function that balances voltage and / or frequency stability with operating costs, and solve it using an improved distributed alternating direction multiplier method. The multi-objective function is as follows:

[0012] in, For voltage and frequency weights, Weighting operating costs and photovoltaic power generation revenue, These are the actual voltage and rated voltage of the power grid, respectively. These are the actual frequency and rated frequency of the power grid, respectively. The total operating cost of the supercharging station Photovoltaic grid integration rate; The penalty factor of the algorithm is dynamically adjusted according to the load type and grid status, and min-max robust optimization is incorporated to suppress impulse noise.

[0013] Furthermore, in S1, the differentiated sampling strategy specifically includes: using a millisecond-level sampling period for supercharging pile impact data and a second-level sampling period for regular operation data; and using sampling periods adapted to the changing characteristics of energy storage systems, V2G equipment, station busbars, and environmental data respectively.

[0014] Furthermore, in step S1, abnormal data is removed during the preprocessing stage. Outliers are identified, and normalized residuals are calculated during the post-validation phase. The data was then removed from the residuals that exceeded the limit, and the missing data was filled in based on the LSTM model and the operating rules of similar equipment.

[0015] Furthermore, in step S2, the load type determination rules are as follows: when the periodic correlation is ≥0.8 and the overlap between the peak period and the historical multi-peak period is ≥90%, it is determined to be a multi-peak rhythm load; when the load change amplitude is ≥200kW, the duration is ≤10s, and it matches the start-stop record of the supercharging pile, it is determined to be an instantaneous impact load; when the load fluctuation amplitude is ≤5%, the duration is ≥30s, and there are no obvious periodic and change characteristics, it is determined to be a stable load.

[0016] Furthermore, in S2, the distribution network status sensing includes calculating the voltage deviation rate:

[0017] Harmonic distortion rate:

[0018] Three-phase imbalance:

[0019] in, The first The actual voltage at the monitoring point, the first Rated voltage at each monitoring point These are the effective values ​​of the nth harmonic voltage and the fundamental voltage, respectively. These are the voltage measurements for phases A, B, and C, respectively. This represents the average effective value of the three-phase voltage. The input feature set is based on a random forest and a Transformer model, and the output state result is also based on that.

[0020] Furthermore, in S3, the dynamic adjustment of the penalty factor includes: adjusting the virtual droop coefficient according to the load type and grid status.

[0021] in, This is a virtual droop coefficient. This is the rated power of the supercharging station. This represents the maximum allowable frequency deviation of the system.

[0022] And a min-max objective function is constructed to suppress impact noise and measurement error:

[0023] in, To monitor the system state Find the infimum. For non-zero disturbances Find the supremum. for System state variables at any given time. for Time-based system control variables This is the weight matrix. This is the noise suppression coefficient. for Constant impact noise and measurement error.

[0024] Furthermore, the present invention also includes: S4. Three-level scheduling execution: "Station level - Pile level - Equipment level": The site-level global optimization is executed every 15 minutes, aiming to minimize operating costs, and outputs the maximum available power of the supercharging pile, the energy storage charging and discharging plan, and the photovoltaic consumption ratio.

[0025] in, For charging loss costs, for Total charging power of the supercharging station at all times. For energy storage dispatch costs, for The charging and discharging power of the energy storage system at all times. For the revenue from photovoltaic power grid integration, for Photovoltaic power output at all times Benefits of reverse power supply for V2G for V2G reverse power supply power at any time; The pile-level real-time control is executed once every 0.1 seconds, generating a single pile power adjustment command based on the impact component and voltage and frequency fluctuations.

[0026] in, For the first One supercharging station The power adjustment command value at any given time. for The total power adjustment of the station-level global optimization output at any given time. For the first One supercharging station Time priority weight, This is the frequency adjustment coefficient. For power grid frequency deviation, This is the voltage regulation coefficient. This refers to the deviation of the power grid voltage. During equipment operation, the energy storage system synchronously discharges to compensate for voltage drops during peak periods, V2G absorbs surplus power during off-peak periods and supplies power in reverse during peak periods, and the supercharging pile supports 30% to 100% rated power adjustment.

[0027] S5. Security Protection and Dynamic Correction: The security protection adopts a three-level response strategy. Warning-level protection is triggered when the transformer load rate is ≥70% or ≤30%, the voltage deviation is ≥3%, and the frequency deviation is ≥±0.015Hz. The adjustable margin allocation ratio is reduced, and the power limit of the overcharging pile is reduced by 10%-20%. Intervention-level protection is triggered when a charging pile fault is detected, the vehicle's SOC is greater than 80%, or the harmonic distortion rate is greater than 5%, to isolate the faulty unit and lock the adjustment authority. Emergency protection is triggered when the transformer load rate is >80% or <30% and a short circuit / ground fault occurs, performing power freeze and GOOSE millisecond-level tripping. The dynamic correction employs PI adaptive control, and the output formula is:

[0028] in, for The power execution command value at any given time. for The algorithm continuously optimizes the output power value. This is the proportionality coefficient. for Power deviation value at any time The integral coefficient is... This is the integral term of the power deviation, which occurs when the load fluctuates drastically. The system is stable. .

[0029] S6. Multi-Supercharging Station Cooperative Scheduling: When the peak-valley difference of the distribution network load is ≥1000kW or the adjustable margin of a single station is less than 50kW, multi-station coordination is initiated. Data exchange is performed through the OCPP protocol, and the globally optimal regulation capacity is calculated based on the improved dADMM algorithm.

[0030] in, Forecast load for the next period. The average daily load This represents the total adjustable capacity of all stations. This represents the available energy storage capacity.

[0031] S7. Load prediction based on supercharging station cluster: Variational mode decomposition is used to decompose the cluster load into trend, rhythm and impact components. XGBoost, BiLSTM+attention mechanism and improved BiLSTM model are used for prediction respectively. The prediction results are integrated through dynamic weight fusion mechanism and a "prediction-feedback-correction" closed loop mechanism is established. When the prediction error exceeds 5%, the model parameters are corrected.

[0032] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) Improve the accuracy of data collection and processing: By differentiating the sampling and double-layer abnormal data identification for the characteristics of "multi-peak + impact", the millisecond-level characteristics of impact load and the minute-level trend of multi-peak load are accurately captured, providing a high-quality data foundation for subsequent regulation.

[0033] (2) Enhance the adaptability and robustness of the optimization algorithm: By deconstructing the load into trend, rhythm and impact components through VMD, and dynamically adjusting the parameters of the dADMM algorithm based on this, the optimization strategy can adaptively cope with the periodic changes of multi-peak loads and the sudden interference of impact loads, which significantly improves the control accuracy and algorithm robustness.

[0034] (3) Achieving deep synergy between regulation and protection: Through the deep coupling of three-level time scale scheduling ("station level-pile level-equipment level") and three-level safety protection ("early warning-intervention-emergency"), supplemented by PI adaptive dynamic correction, a closed-loop regulation system that can meet global optimization, respond quickly to emergencies, and has high security has been constructed, effectively ensuring the safe and stable operation of supercharging stations and distribution networks.

[0035] (4) Improve multi-station coordination efficiency: By using the hybrid protocol system of IEC 61850 and OCPP, the system takes into account both millisecond-level real-time control within the station and standardized coordination between stations, realizing adjustable capacity aggregation and global load balancing of multiple supercharging station clusters, and significantly improving the energy utilization efficiency of the regional distribution network.

[0036] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art based on the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0037] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein: Figure 1 This is a block diagram of the overall architecture of the method of the present invention; Figure 2 This is a flowchart illustrating the control and protection process of the method of the present invention; Figure 3 This is the node interaction graph of the improved dADMM algorithm of this invention. Detailed Implementation

[0038] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0039] The accompanying drawings are for illustrative purposes only and are schematic diagrams, not actual pictures, and should not be construed as limiting the invention. To better illustrate the embodiments of the invention, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0040] Please see Figures 1-3 ,in: Figure 1 The diagram shows the overall architecture of the method of this invention. The system consists of a multi-source data acquisition layer, a load feature analysis layer, a distributed optimization layer, a three-level scheduling execution layer, a security protection layer, a multi-station collaboration layer, and a feedback layer. Each layer achieves closed-loop linkage through data transmission links, jointly supporting the precise control of the supercharging station load.

[0041] Figure 2 The flowchart for the regulation and protection of the method of the present invention details the complete process from multi-supercharging station collaborative trigger judgment, single-station dual-dimensional margin assessment, multi-source data acquisition and preprocessing, load feature hierarchical extraction, power grid status perception, improved dADMM distributed robust optimization, safety threshold trigger detection to PI adaptive dynamic correction.

[0042] Figure 3 This is a node interaction graph for an improved dADMM algorithm, illustrating the interaction mechanism where each device node independently solves its local subproblem and achieves distributed collaborative optimization by exchanging Lagrange multipliers with neighboring nodes.

[0043] The present invention will be further explained below with reference to various embodiments.

[0044] Example 1 provides a distributed robust control method adapted to the multi-peak and impulsive load characteristics of supercharging stations. This method addresses the combined load characteristics of multi-peak rhythms and instantaneous impacts at supercharging stations by employing deep fusion of multi-source data, hierarchical feature extraction, distributed robust optimization, and three-level collaborative scheduling to achieve active load balancing in the distribution network. Figure 1 As shown, the method includes the following core steps: S1. Dedicated collection and preprocessing of multi-source heterogeneous data within the supercharging station.

[0045] This step focuses on multi-peak load periods, instantaneous impact scenarios, and cross-platform data interaction scenarios. Follow these steps: (1) Collect multi-source data in all dimensions.

[0046] Supercharging / fast charging piles collect charging power, current, voltage, and operating status, with an impact data sampling period of 20ms and a normal operating data sampling period of 15s; energy storage systems collect charging and discharging power, state of charge, and charging and discharging efficiency; V2G devices collect bidirectional power and access status; the station bus collects voltage, current, and frequency, with a sampling period of 6s; environmental data collects temperature and humidity, with a sampling period of 30min; cross-platform data is collected from the power distribution automation system, electricity consumption information collection system, and charging operation platform, respectively.

[0047] (2) Secure data transmission and standardized processing.

[0048] Cross-platform transmission employs high-compression lossless compression, data signature, identity authentication, and data auditing mechanisms; intra-station transmission is based on the IEC 61850 protocol, and the communication latency is guaranteed to be ≤80ms through the GOOSE message mechanism; linear transformation is used to unify the dimensions, and the power measurement format of different devices is unified through the line power conversion formula.

[0049] (3) Optimize the processing of measurement data.

[0050] PMU data is filtered using recursive averaging; SCADA data retains the original measurements; AMI data is augmented with linear interpolation to complete the trend; and impact data is filtered using improved wavelet filtering to suppress noise to within 3%. Based on the station bus data, the spatiotemporal deviation of cross-equipment and cross-platform data is corrected to within 0.5s using an interpolation synchronization algorithm. Abnormal data is removed through a two-layer identification process of preprocessing and post-verification, and missing data is filled in based on an LSTM model to ensure data integrity ≥98%.

[0051] S2. Layered extraction and status perception of load characteristics within the station.

[0052] Perform this operation after data preprocessing is complete, following these steps: (1) Multidimensional decomposition of load sequence.

[0053] The total load within the station is decomposed into physically meaningful components based on the Variational Mode Decomposition (VMD) algorithm, as shown in the following formula:

[0054] in, To decompose the component set and center frequency set Find the minimum value. For time partial derivatives, For the Dirac function, For the first Temporal representation of each decomposed component , The first Each decomposed component and its center frequency The imaginary unit, It is the frequency shift factor; After decomposition, the following components are obtained: trend component, reflecting the base load level; rhythm component, showing multi-peak periodic changes, with peak value, duration and regional adaptation parameters marked; and impact component, from which the sudden change time, amplitude and time-frequency characteristics are extracted.

[0055] The load type determination rule is as follows: ① The periodic correlation is ≥0.8, and the overlap between the peak period and the historical multi-peak period is ≥90%. After adjusting the peak period based on the regional type, it is determined to be a multi-peak rhythm load. ② Load surge amplitude ≥ 200kW, duration ≤ 10s, and match the start-stop record of the supercharging pile, and confirmed by extracting time-frequency features through fast KS transformation, are determined to be instantaneous impact loads; ③ A load with a fluctuation amplitude of ≤5% and a duration of ≥30s, and without obvious periodic or abrupt changes, is considered a stable load.

[0056] (2) Construction of full-dimensional feature set.

[0057] It integrates load decomposition characteristics, equipment status characteristics, and power grid status characteristics, covering core features such as pile type power level, regional type, multi-peak period parameters, impact characteristics, voltage deviation rate, harmonic distortion rate, and fault identification.

[0058] The formula for calculating the voltage deviation rate is as follows:

[0059] The formula for calculating harmonic distortion rate is:

[0060] in, The first Actual voltage and rated voltage at each monitoring point These are the effective values ​​of the nth harmonic voltage and the fundamental voltage, respectively.

[0061] (3) Distribution network status perception.

[0062] Calculate voltage quality indicators such as voltage deviation, fluctuation, unbalance, and stability margin, as well as reliability indicators such as heavy overload, line outage, short circuit, and grounding. The formula for calculating three-phase unbalance is:

[0063] in, The voltage measurements for phases A, B, and C are respectively. This represents the average effective value of the three-phase voltage.

[0064] Based on the input feature set of the Random Forest and Transformer models, the output state results are generated by using the Guangming large model to supplement the lack of fault data. After 50 iterations, the accuracy is ≥95%.

[0065] Among them, when the fault identification accuracy is ≥95%, it is judged as a fault state; the decomposed components must meet the following requirements: the peak value ratio of the rhythm component is ≥60% and the consistency between the impact component and the equipment operation record is ≥95%; otherwise, they must be decomposed again; based on power conservation, the deviation between the total equipment load and the bus load must be ≤2%; otherwise, abnormal data will be removed.

[0066] S3. Improved distributed robust control optimization of dADMM.

[0067] After feature extraction and state awareness are completed, perform the following operations: (1) Modeling of distributed optimization problems.

[0068] The objective function takes into account power balance, voltage and frequency stability, and operating costs. The comprehensive objective function is as follows:

[0069]

[0070] in, The voltage and frequency weights are dynamically adjusted based on the real-time status of the power grid. When the voltage deviation exceeds 3%, When the weight is increased to 0.6, and the frequency deviation exceeds ±0.015Hz... The weighting has been increased to 0.6. ， Weighting operating costs and photovoltaic grid integration revenue, peak and off-peak periods Weighting 0.5, photovoltaic-storage-charging scenario Weighting percentage: 0.4 The weight matrix is ​​a quadratic term. To regulate the quadratic term of the variable, To regulate the linear term of the variable, The instantaneous frequency and rated angular frequency of the power grid. This is the squared term of the angular frequency deviation.

[0071] The constraint system includes equipment operation constraints, timing response constraints, and safe operation constraints.

[0072] Equipment operation constraints require vehicle SOC ≤ 80%, transformer load rate ∈ [30%, 80%], charging pile power adjustment range ∈ [30% of rated power, 100% of rated power], and compatibility with different specifications of supercharging piles and fast charging piles such as 600kW / 250kW / 160kW; Timing response constraints specify that the communication delay within the station is ≤80ms, the control response time is <1 second, the real-time control cycle at the pile level is 0.1 seconds, and the global optimization cycle at the station level is 15 minutes. Safety operation constraints stipulate that voltage deviation ≤ 5%, frequency deviation ≤ ±0.02Hz, harmonic distortion rate ≤ 5%, and power execution error ≤ 5%.

[0073] The QU droop strategy adjusts reactive power output to maintain stable bus voltage at the supercharging station based on the deviation between the actual bus voltage and the reference voltage. The formula is as follows:

[0074] in, This is the reactive power output value. , These are the upper limit of maximum reactive power and the lower limit of minimum reactive power, respectively. This is the reactive power-voltage droop factor. This is the voltage reference value. This is the actual bus voltage.

[0075] (2) Dynamic parameter adjustment and robust optimization.

[0076] The penalty factor is dynamically adjusted based on the load type and grid status. The virtual droop coefficient is calculated from the rated power of the charging pile and the maximum allowable frequency deviation of the system, using the following formula:

[0077] in, This is a virtual droop coefficient. This is the rated power of the supercharging station. This represents the maximum permissible frequency deviation for the system. Frequency variations are correlated with active power regulation to suppress frequency deviation.

[0078] The min-max objective function is constructed to suppress impact noise and measurement error, and the formula is as follows:

[0079] in, To monitor the system state Find the infimum. For non-zero disturbances Find the supremum. for System state variables at any given time. for Time-based system control variables This is the weight matrix. This is the noise suppression coefficient. for Constant impact noise and measurement error.

[0080] The SOGI-PLL phase-locked loop technology accurately acquires voltage, frequency, and phase. The relevant formula for SOGI is:

[0081] in, for time Shaft output voltage, for time Shaft output voltage, The SOGI cutoff frequency, The sampling period is These are the initial conditions.

[0082] The improved PLL model formula is:

[0083] in, The imaginary unit, The input signal angular frequency, This is the PLL scaling factor. The time constant of the PLL damping is... For PLL integral coefficients, This is the PLL damping coefficient.

[0084] (3) Distributed solution and mode switching.

[0085] Each device node independently computes its local subproblem:

[0086] in, For the first device nodes The optimized variable values ​​for the next iteration. For the first Optimization variables for each device node For the number of iterations, For the first Lagrange functions of nodes, For the first Nodes The Lagrange multipliers of the next iteration, for The global consensus variable for the next iteration. This is a penalty factor.

[0087] Only boundary information and Lagrange multipliers are exchanged. When the frequency deteriorates and the deterioration rate exceeds the critical value, virtual positive inertial control is supplemented. When the frequency recovers and the recovery rate is below the critical value, virtual negative inertial control is supplemented. The relevant formulas for mode switching are:

[0088] in, This represents the total power adjustment for virtual inertial control. This is the virtual damping power adjustment. This is the virtual positive inertia power adjustment quantity. This is the virtual negative inertia power adjustment amount. This is the positive inertia control switching coefficient. For negative inertia control switching coefficient; The algorithm terminates when the power allocation error between two consecutive iterations is ≤0.1% or when the number of iterations reaches 50.

[0089] S4, three-level scheduling execution: "station level - pile level - equipment level".

[0090] This step is performed after the optimization algorithm outputs the optimal solution. The specific steps are as follows: (1) Global optimization at the station level.

[0091] Execute every 15 minutes, with the goal of minimizing operating costs. The formula is:

[0092] in, For charging loss costs, for Total charging power of the supercharging station at all times. For energy storage dispatch costs, for The charging and discharging power of the energy storage system at all times. For the revenue from photovoltaic power grid integration, for Photovoltaic power output at all times Benefits of reverse power supply for V2G for V2G reverse power supply at any time.

[0093] The optimized output includes the maximum available power of supercharging piles and fast charging piles, energy storage charging and discharging plans, and photovoltaic consumption ratio. The benchmark value for active regulation on the pile side is issued to the Chongqing demonstration station, and the decomposition plan for power regulation on the line side is issued to the Wuhan demonstration station to ensure that the transformer load rate is ≤80% and the total regulation capacity of the supercharging station is ≥10000kW.

[0094] (2) Real-time control of pile level.

[0095] Executed every 0.1 seconds, a single-pile power adjustment command is generated based on the impact component and voltage and frequency fluctuations. The formula is as follows:

[0096] in, For the first One supercharging station The power adjustment command value at any given time. for The total power adjustment of the station-level global optimization output at any given time. For the first One supercharging station Time priority weight, This is the frequency adjustment coefficient. For power grid frequency deviation, This is the voltage regulation coefficient. This refers to the deviation of the power grid voltage.

[0097] (3) Precise execution at the equipment level.

[0098] The energy storage system synchronously discharges to compensate for voltage drops during peak periods, and quickly smooths out fluctuations in photovoltaic output. V2G absorbs surplus power during off-peak hours and provides reverse power supply support during peak hours or faults. The supercharging station dynamically adjusts its power according to instructions, supporting power adjustment from 30% to 100% of the rated power.

[0099] The equipment's execution data is fed back to the optimization module in real time, which corrects the parameters for the next round, forming a closed-loop control.

[0100] S5. Safety protection and dynamic correction.

[0101] This step is repeated throughout the entire equipment operation process: (1) Three-level security protection mechanism.

[0102] The warning-level protection is triggered when the transformer load rate is ≥75% or ≤35%, the voltage deviation is ≥3%, and the frequency deviation is ≥±0.015Hz. It reduces the adjustable margin allocation ratio (the reduction coefficient is between 0.7 and 0.9), reduces the upper limit of the supercharging pile power by 10%-20%, and activates the energy storage backup capacity in advance.

[0103] Intervention-level protection isolates faulty charging piles, locks the adjustment authority of vehicles with SOC > 80%, and is triggered when the harmonic distortion rate is ≥ 5%. It isolates the faulty unit, locks the adjustment authority of the corresponding vehicle, and starts the reactive power compensation module.

[0104] Emergency protection is triggered when the transformer load rate is >80% or <30%, a short circuit or ground fault occurs, or the voltage suddenly rises or falls. It triggers a power freeze command, causing the faulty equipment to shut down urgently and switch to local autonomous mode. It achieves millisecond-level tripping through the GOOSE message mechanism.

[0105] (2) PI adaptive dynamic correction.

[0106] Millisecond-level correction of power output is achieved through a PI controller:

[0107] in, for The power execution command value at any given time. The algorithm continuously optimizes the output power value. for Power deviation value at any time, proportional coefficient when load fluctuates drastically Integral coefficients when the system is stable When the power execution error is greater than 5% or the voltage or frequency deviation exceeds the target value, dynamic correction is activated. When the load fluctuates drastically, the PI proportional coefficient is increased, and when the system is stable, the integral coefficient is decreased.

[0108] (3) Determine the termination condition.

[0109] When the actual load of the distribution network deviates from the target load by ≤2% for 10 minutes, or when a charging station completes charging or the equipment status is abnormal, the current dispatch cycle is terminated and the next iteration is started.

[0110] S6, multi-scenario adaptation and demonstration implementation verification.

[0111] This step is applied to the implementation phase of the entire regulation process: (1) Scene parameter adaptation.

[0112] Based on the regional type, identify the peak periods and adjust the frequency and voltage regulation coefficients to ensure that the control is adapted to the load characteristics of different regions; adapt solutions for supercharging stations of different sizes.

[0113] For example, peak hours in residential areas are 7-8 AM and 7-9 PM, in commercial areas it's 12-2 PM and 6-8 PM, and at the airport it's 6-8 AM and 8-10 PM. To address the differences between the Chongqing and Wuhan demonstration models, differentiated parameters were set: Chongqing's pile-side control dead zone accuracy is 0.005Hz, and the response time is <1 second; Wuhan's station-side servo terminal uses the OCPP protocol to communicate with the vehicle network platform, with an information latency ≤1 minute. Model training used: a random forest + Transformer model with a training set:validation set:test set ratio of 7:2:1, 50 training iterations, and a learning rate of 0.001; the XGBoost + BiLSTM + attention mechanism model had 128 hidden layer neurons, with an initial attention weight coefficient of 0.5.

[0114] (2) Simulation and field verification.

[0115] An IEEE 14-node model incorporating supercharging stations, photovoltaics, energy storage, and V2G was built on MATLAB, verifying voltage estimation error ≤0.04%, frequency deviation ≤±0.02Hz, and power regulation error ≤1.03MW. Field tests were conducted at the Chongqing Luneng Xingcheng 600kW supercharging station, lasting 72 hours and covering peak hours (7-9 AM, 6-8 PM, and midnight) as well as a scenario involving the simultaneous activation of three supercharging piles (total impact power 880kW). Test results showed a voltage deviation rate of 1.8%, an impact response time of 0.3s, an inter-pile power conflict rate of 0.2%, and a 15% reduction in operating costs. In the integrated photovoltaic-energy storage-charging verification at the Wuhan South Taizihu station, the photovoltaic absorption rate increased to 92%, and the energy storage charging and discharging efficiency remained above 95%.

[0116] (3) Iterative optimization.

[0117] Based on simulation and field verification results, the dADMM parameters, scheduling weights, and PI coefficients were adjusted to ensure that all assessment indicators, such as fault identification accuracy ≥95%, load forecast accuracy ≥95%, and control response time <1 second, were fully achieved.

[0118] S7. Distribution network adaptive control system architecture based on supercharging station cluster collaboration.

[0119] Deploy and operate according to the following architecture: (1) Distribution Network Dispatch Center: As the global coordination unit, it is responsible for formulating load regulation benchmarks and issuing global dispatch instructions. The distribution network dispatch center periodically receives real-time load data and historical operating curves of the regional distribution network, and, in conjunction with the weather, holidays and supercharging station operation data of the day, sets time thresholds for peak and off-peak periods of the distribution network load, and clarifies core parameters such as the upper and lower limits of transformer load rate, the adjustable power range of supercharging piles, the SOC critical value and the impact load response threshold. At the same time, the dispatch center decomposes the macro load regulation target into the adjustable capacity range and regulation priority of each supercharging station, and issues it to the follow-up control terminal of each station in the form of dispatch task packages.

[0120] The dispatching strategy adopts a flexible weight adjustment mechanism: when the peak-valley difference of the distribution network load is too large (≥1000kW), the adjustment weight will be biased towards the load leveling effect, and the maximum adjustable margin of each supercharging station will be used first; when the transformer load rate of some supercharging stations is close to the critical value (≥70% or ≤30%), the equipment safety protection weight will be increased, and the adjustment capacity of the station will be reduced by the adjustable coefficient to avoid equipment overload; when a fault of the supercharging pile is detected, the abnormal unit will be automatically removed to ensure that the adjustment command is only issued to the normally operating equipment.

[0121] (2) Edge-end collaborative control device: Real-time acquisition of the working status of supercharging piles, charging power and current, vehicle battery SOC, and real-time load, load rate, energy storage SOC and V2G access status of the distribution transformer, and establish a local two-dimensional adjustable margin assessment model; For charging vehicles with SOC≤80%, the adjustable margin of a single pile is calculated differently according to peak and valley periods, and the total adjustable margin of the cluster is obtained by summing them up; Based on the current load, rated capacity and safe load rate threshold, the expected load rate after adjustment is calculated, and when it exceeds 30%-80%, it is reduced by the adjustable coefficient (0.3-0.9) to obtain the adjustable margin of the transformer; The distribution network load is adjusted upward during the valley period and downward during the peak period. When an instantaneous impact load is detected, the energy storage collaborative smoothing strategy is triggered.

[0122] (3) End-side execution unit: The active control terminal on the pile side dynamically adjusts the charging power for vehicles with SOC≤80% and maintains the current charging state for vehicles with SOC>80%; the energy storage system synchronously charges and discharges to compensate for power mutations in impact scenarios; V2G absorbs surplus power during grid off-peak periods and supplies power in reverse during peak periods; the distribution transformer provides real-time feedback of load rate data, and reduces the adjustable margin of the supercharging station when the expected load rate exceeds the threshold after adjustment; the bottom-level equipment and the middle-level collaborative control device realize real-time data interaction through the IEC 61850 protocol, forming a closed-loop adjustment link of "evaluation-command-execution-feedback".

[0123] (4)Multi - supercharger station collaboration: When the peak - valley difference of the distribution network load ≥ 1000 kW or the adjustable margin of a single station is less than 50 kW, multi - station collaboration is initiated. When the load prediction deviation ≥ 5% or the impact load amplitude ≥ 200 kW, the deviation correction mechanism is triggered; the collaborative control devices of each supercharger station exchange information through the OCPP protocol. When the deviation between the actual load and the target load of the distribution network ≤ 2% and lasts for 10 minutes, or when a certain charging station completes charging / equipment status is abnormal, the current deployment cycle is terminated; the adjustable capacities of each station are aggregated through a distributed communication network, and the global optimal regulation capacity is calculated based on the improved dADMM algorithm.

[0124] S8. A communication control method based on the hybrid communication protocol of IEC 61850 and OCPP.

[0125] At the communication architecture level, the IEC 61850 protocol, as the in - station underlying high - speed communication standard, supports millisecond - level in - station data interaction through the MMS service mechanism, and realizes the linkage protection of in - station equipment through the GOOSE message mechanism, ensuring strong real - time and high reliability of in - station communication. The OCPP protocol focuses on inter - station collaborative communication and upper - layer scheduling interaction, runs on top of the TCP / WebSocket channel, supports the JSON message format and the TLS / SSL encryption transmission mechanism, and realizes the real - time reporting of the adjustable capacity of each supercharger station, the issuance of collaborative adjustment instructions, and the synchronization of operating states.

[0126] At the connection level of the hybrid communication architecture, the IEC 61850 and OCPP protocols are seamlessly connected and data interoperability is achieved through the communication adaptation layer, solving the format difference problem between in - station real - time data and inter - station collaborative data, and completing the semantic correspondence and bidirectional conversion between the IEC 61850 logical node data and the OCPP data model.

[0127] At the operation function level, relying on this communication architecture, this system has three core capabilities: dynamic synchronization of adjustable capacity, real - time feedback of load deviation, and local autonomy during communication interruption. When communication is abnormal, the collaborative control devices of each supercharger station automatically switch to the local autonomous operation mode, operate independently based on local real - time data, and quickly synchronize data through the status synchronization module after communication is restored, and converge back to the globally consistent state again, avoiding load imbalance or equipment overload problems caused by data disconnection.

[0128] S9. A load prediction method based on a supercharger station cluster.

[0129] (1)In the pre - processing link of prediction data, multi - source heterogeneous data of the supercharger station cluster is collected, covering in - station equipment operation data, environmental data, operation data, and distribution network associated data.

[0130] The data collected includes: charging power, current, voltage, SOC, and start / stop status of supercharging / fast charging piles; charging and discharging power and SOC of energy storage systems; and access status and bidirectional power of V2G devices. The sampling period for regular data is 15 minutes, while the sampling period for impact data is increased to 20 ms. Environmental data includes temperature, humidity, and weather type, with a sampling period of 30 minutes. Operational data includes the number of vehicles connected, charging duration, and pile type distribution. Distribution network-related data includes historical load curves for the region, peak and valley time periods, and voltage / frequency fluctuation records. The collected data is standardized by unifying dimensions through linear transformation and using improved wavelet filtering technology to suppress noise in impact data, ensuring noise amplitude ≤3%. Missing data is filled using an LSTM model to ensure data integrity ≥99.5%. A spatiotemporal alignment algorithm is used to correct the time-series deviation of data from each station within the cluster, controlling the deviation within 0.3 seconds. Finally, a standardized prediction dataset is constructed.

[0131] (2) In the load sequence hierarchical decomposition stage, the variational mode decomposition algorithm is used to decompose the load sequence of the supercharging station cluster into three types of components with clear physical meaning, so as to achieve accurate extraction of load characteristics.

[0132] Among them, the trend component reflects the long-term stable change pattern of cluster load, unaffected by short-term fluctuations and instantaneous shocks; the rhythm component corresponds to the multi-peak load characteristics of 7-9 am, 6-8 pm, and midnight, highlighting key parameters such as peak amplitude, duration, and periodic correlation, while adapting to the differences in multi-peak time periods in different regions; the impact component specifically captures the instantaneous fluctuations caused by the start-up of supercharging piles and simultaneous charging of multiple piles, accurately extracting features such as the time of abrupt change, amplitude, and decay rate. Through component separation, the complex load sequence is decomposed into single characteristic components, laying the foundation for subsequent targeted prediction.

[0133] (3) In the multi-model collaborative prediction stage, an appropriate model is selected based on the characteristics of different load components, and a fusion prediction framework of XGBoost, BiLSTM and attention mechanism is constructed.

[0134] For the trend component, the XGBoost model is used, leveraging its strong fitting advantage to stationary time-series data. It inputs multi-source features such as historical load, temperature, and date type, and outputs trend prediction results. For the rhythm component, a BiLSTM model is used to capture periodic correlations, and an attention mechanism is combined to strengthen the weights of features from multiple peak periods, highlighting the load variation patterns during peak periods, and outputting rhythm prediction results. For the impact component, an improved BiLSTM model is used, introducing an impact feature embedding layer. It inputs specific features such as historical impact records, pile type parameters, and equipment start-up and shutdown status, and outputs impact prediction results. A dynamic weight fusion mechanism integrates the prediction results of each component to obtain the final predicted value of the cluster load. The dynamic weights are adaptively adjusted based on the historical prediction errors of each component; the smaller the error, the greater the weight, ensuring that the prediction results more closely match actual load changes.

[0135] (4) In the dynamic correction and iterative optimization process, a closed-loop mechanism of “prediction-feedback-correction” is established.

[0136] The error between the predicted value and the actual load is calculated, and the error distribution characteristics are statistically analyzed using a 24-hour rolling window. When the error exceeds a preset threshold of 5%, model parameter correction is triggered, adjusting the learning rate of XGBoost, the number of hidden layer neurons in BiLSTM, and the attention weight coefficient to optimize model adaptability. For the prediction deviation of impact load, an additional error compensation mechanism is introduced, with the compensation coefficient controlled between 0.3 and 0.7, further improving the accuracy of impact load prediction. Simultaneously, the Guangming large model is used to assist in generating expanded samples to compensate for insufficient samples in rare scenarios such as impact load and extreme weather, significantly improving the model's generalization ability. Prediction results are updated every 15 minutes, providing real-time data support for the coordinated control of supercharging station clusters and the generation of distribution network dispatch instructions.

[0137] Example 2 provides a distributed robust control system for implementing the method described in Example 1, comprising five functional modules that work together: 1. Data Acquisition Unit: Adapted to IEC 61850 and OCPP protocols, it collects multi-source data from charging piles, transformers, and power distribution networks, and supports access for heterogeneous devices from multiple brands.

[0138] 2. Status Analysis Module: Based on the load forecasting model and peak-valley identification algorithm, it outputs the status judgment results of the distribution network and the two-dimensional adjustable margin assessment report within the station.

[0139] 3. Decision-making unit: Runs distributed collaborative algorithm and active load balancing algorithm for distribution lines to generate global optimal regulation strategy and power allocation instructions for each station.

[0140] 4. Communication Unit: Supports adjustable capacity data exchange between stations and transmission of upper-level scheduling instructions, and has dual redundant links to ensure communication continuity.

[0141] 5. Execution Unit: Receives decision instructions and drives the power adjustment of the charging pile cluster, links with the PI controller to achieve dynamic correction, and executes safety protection actions.

[0142] In summary, this invention addresses the combined load characteristics of supercharging stations, including multi-peak rhythms and instantaneous impacts. It achieves proactive load balancing in the distribution network through deep fusion of multi-source data, hierarchical feature extraction, distributed robust optimization, and three-level collaborative scheduling. This includes: collecting and preprocessing multi-source data using a differentiated sampling strategy; decomposing the load into trend, rhythm, and impact components based on VMD and sensing the grid status; performing distributed robust optimization using an improved dADMM algorithm, dynamically adjusting the penalty factor according to load type and incorporating min-max robust control; constructing a three-level scheduling execution architecture at the station level, pile level, and equipment level; and establishing a three-level security protection mechanism with linked PI adaptive dynamic correction. This invention achieves a station bus voltage deviation rate ≤2.5%, instantaneous impact load response time ≤0.5s, and inter-pile power conflict rate ≤0.3% during multi-peak periods, significantly improving the operational stability and equipment coordination efficiency of supercharging stations and demonstrating promising application prospects.

[0143] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A distributed robust control method adapted to the multi-peak and impact load characteristics of supercharging stations, wherein the supercharging station includes supercharging piles, an energy storage system, and V2G equipment, characterized in that, Includes the following steps: S1. Dedicated collection and preprocessing of multi-source heterogeneous data within supercharging stations: For multi-peak load periods and instantaneous impact scenarios, a differentiated sampling strategy is adopted to collect data from core equipment and cross-platforms within the station; the collected data is safely transmitted, standardized, layered filtered and noise suppressed, and abnormal data is removed and missing data is filled through preprocessing and post-verification dual-layer identification; S2. Hierarchical Extraction and Status Awareness of Station Load Characteristics: Based on the variational mode decomposition algorithm, the total load within the station is decomposed into trend components, rhythmic components, and impulse components. The objective function of variational mode decomposition is: in, To decompose the component set and center frequency set Find the minimum value. For time partial derivatives, For the Dirac function, For the first Temporal representation of each decomposed component , The first Each decomposed component and its center frequency The imaginary unit, It is the frequency shift factor; S3. Improved distributed robust control optimization of dADMM: Construct a multi-objective function that balances voltage and / or frequency stability with operating costs, and solve it using an improved distributed alternating direction multiplier method. The multi-objective function is as follows: in, For voltage and frequency weights, Weighting operating costs and photovoltaic power generation revenue, These are the actual voltage and rated voltage of the power grid, respectively. These are the actual frequency and rated frequency of the power grid, respectively. The total operating cost of the supercharging station Photovoltaic grid integration rate; The penalty factor of the algorithm is dynamically adjusted according to the load type and grid status, and min-max robust optimization is incorporated to suppress impulse noise.

2. The method according to claim 1, characterized in that, In S1, the differentiated sampling strategy specifically includes: using a millisecond-level sampling period for supercharging pile impact data and a second-level sampling period for regular operation data; and using sampling periods adapted to the changing characteristics of energy storage systems, V2G equipment, station busbars, and environmental data respectively.

3. The method according to claim 1, characterized in that, In step S1, abnormal data is removed during the preprocessing stage. Outliers are identified, and normalized residuals are calculated during the post-validation phase. Data with excessive residuals was removed, and missing data was filled in based on the LSTM model combined with the operating patterns of similar equipment; among them, This is the raw monitoring data. The mean of the data. This is the threshold for anomaly detection. The standard deviation of the data; To normalize the residuals, For the original residual, It is the square root of the diagonal elements of the covariance matrix.

4. The method according to claim 1, characterized in that, In S2, the load type determination rules are as follows: when the periodic correlation is ≥0.8 and the overlap between the peak period and the historical multi-peak period is ≥90%, it is determined to be a multi-peak rhythm load; when the load change amplitude is ≥200kW, the duration is ≤10s and it matches the start and stop record of the supercharging pile, it is determined to be an instantaneous impact load; when the load fluctuation amplitude is ≤5% and the duration is ≥30s and there are no obvious periodic and change characteristics, it is determined to be a stable load.

5. The method according to claim 1, characterized in that, In S2, the distribution network status sensing includes calculating the voltage deviation rate: Harmonic distortion rate: Three-phase imbalance: in, The first The actual voltage and rated voltage of each monitoring point These are the effective values ​​of the nth harmonic voltage and the fundamental voltage, respectively. The voltage measurements for phases A, B, and C are respectively. This represents the average effective value of the three-phase voltage. The input feature set is based on a random forest and a Transformer model, and the output state result is also based on that.

6. The method according to claim 1, characterized in that, In S3, the dynamic adjustment of the penalty factor includes: adjusting the virtual droop coefficient according to the load type and grid status. in, This is a virtual droop coefficient. This is the rated power of the supercharging station. This represents the maximum allowable frequency deviation of the system. Simultaneously, a min-max objective function is constructed to suppress impact noise and measurement error: in, To monitor the system state Find the infimum. For non-zero disturbances Find the supremum. for System state variables at any given time. for Time-based system control variables This is the weight matrix. This is the noise suppression coefficient. for Constant impact noise and measurement error.

7. The method according to claim 1, characterized in that, It also includes S4 and a three-level scheduling execution system: "site-level-pile-equipment level". The site-level global optimization is executed every 15 minutes, aiming to minimize operating costs, and outputs the maximum available power of the supercharging pile, the energy storage charging and discharging plan, and the photovoltaic consumption ratio. in, For charging loss costs, for Total charging power of the supercharging station at all times. For energy storage dispatch costs, for The charging and discharging power of the energy storage system at all times. For the revenue from photovoltaic power grid integration, for Photovoltaic power output at all times Benefits of reverse power supply for V2G for V2G reverse power supply power at any time; The pile-level real-time control is executed once every 0.1 seconds, generating a single pile power adjustment command based on the impact component and voltage and frequency fluctuations. in, For the first One supercharging station The power adjustment command value at any given time. for The total power adjustment of the station-level global optimization output at any given time. For the first One supercharging station Time priority weight, This is the frequency adjustment coefficient. For power grid frequency deviation, This is the voltage regulation coefficient. This refers to the deviation of the power grid voltage. During equipment operation, the energy storage system synchronously discharges to compensate for voltage drops during peak periods, V2G absorbs surplus power during off-peak periods and supplies power in reverse during peak periods, and the supercharging pile supports 30% to 100% rated power adjustment.

8. The method according to claim 1, characterized in that, It also includes S5, security protection, and dynamic correction, with the security protection employing a three-level response strategy: Warning-level protection is triggered when the transformer load rate is ≥70% or ≤30%, the voltage deviation is ≥3%, and the frequency deviation is ≥±0.015Hz. The adjustable margin allocation ratio is reduced, and the power limit of the overcharging pile is reduced by 10%-20%. Intervention-level protection is triggered when a charging pile fault is detected, the vehicle's SOC is greater than 80%, or the harmonic distortion rate is greater than 5%, to isolate the faulty unit and lock the adjustment authority. Emergency protection is triggered when the transformer load rate is >80% or <30% and a short circuit / ground fault occurs, performing power freeze and GOOSE millisecond-level tripping. The dynamic correction employs PI adaptive control, and the output formula is: in, for The power execution command value at any given time. for The algorithm continuously optimizes the output power value. This is the proportionality coefficient. for Power deviation value at any time The integral coefficient is... This is the integral term of the power deviation, which occurs when the load fluctuates drastically. The system is stable. .

9. The method according to claim 1, characterized in that, It also includes S6 and multi-supercharging station collaborative scheduling: when the peak-valley difference of the distribution network load is ≥1000kW or the adjustable margin of a single station is less than 50kW, multi-station collaboration is initiated, data interaction is carried out through the OCPP protocol, and the globally optimal regulation capacity is calculated based on the improved dADMM algorithm. in, Forecast load for the next period. The average daily load This represents the total adjustable capacity of all stations. This represents the available energy storage capacity.

10. The method according to claim 1, characterized in that, It also includes S7, load prediction based on supercharging station clusters: variational mode decomposition is used to decompose the cluster load into trend, rhythm and impact components, and XGBoost, BiLSTM, attention mechanism and improved BiLSTM model are used for prediction respectively. The prediction results are integrated through dynamic weight fusion mechanism to establish prediction-feedback-correction closed loop mechanism. When the prediction error exceeds 5%, the model parameter correction is triggered.

Images