Split direct current charging multi-split group charging and group control system

Abstract

The invention provides a split type direct current charging multi-split group charging and group control control system, and belongs to the technical field of split type direct current charging piles. The split type direct current charging system comprising a main control cabinet, a power distribution unit, a charging pile terminal and other hardware architectures is constructed; a voltage monitoring unit, a current sensor module, a harmonic filter and other multi-dimensional sensor networks are integrated to acquire power grid state data in real time, and a sliding time window method is used to carry out segmented analysis on the data to extract power grid characteristic parameters; establishing a multi-objective optimization model comprising a power balance equation and a power grid stability equation to calculate an optimal power distribution matrix, adopting a neural network to realize intelligent distribution and dynamic balance adjustment of the power of each charging pile terminal, and starting an emergency power redistribution mechanism to ensure stable operation of the system when the power grid is detected to be abnormal. The technical problem of unbalanced power distribution of the multi-pile charging system caused by power grid voltage fluctuation and harmonic pollution is solved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of split type direct current charging piles, and in particular relates to a split type direct current charging multi-outlet group charging group control system. BACKGROUND

[0002] With the rapid development of electric vehicles and the large-scale construction of charging infrastructure, the split type direct current charging multi-outlet system is widely used in urban charging stations, highway service areas and logistics parks as an intensive charging solution. The system realizes unified scheduling of power resources and effective control of equipment costs by simultaneously supplying power to multiple charging pile terminals through a centralized power distribution unit. Traditional multi-pile charging systems generally use a static power distribution strategy, which presets a fixed distribution ratio according to the rated power of each charging pile. Basic voltage monitoring devices and overcurrent protection mechanisms are configured to maintain basic system operation. This approach can meet basic charging needs under ideal grid conditions. However, in actual operating environments, grid voltage experiences periodic fluctuations and random disturbances, and three-phase voltage imbalance is common. The connection of a large number of nonlinear loads and power electronic devices causes serious harmonic pollution. These factors cause significant differences in input voltage and current characteristics of each charging pile terminal. The traditional fixed distribution strategy cannot be dynamically adjusted according to real-time grid conditions, resulting in insufficient power output for some charging piles when the voltage is low and possible overload operation when the voltage is high. The power distribution between piles is severely unbalanced, the overall system efficiency is reduced, and safety hazards exist. In current multi-pile charging systems, due to the lack of real-time monitoring and compensation mechanisms for grid voltage fluctuations and harmonic pollution, the system cannot maintain the balance of power distribution between piles in complex grid environments. Frequent problems such as large differences in charging power, uneven equipment utilization, and system protection shutdown occur. That is, the existing technology has the technical problem of power distribution imbalance in multi-pile charging systems caused by grid voltage fluctuations and harmonic pollution. SUMMARY

[0003] Therefore, the present application provides a split type direct current charging multi-outlet group charging group control system, which can solve the technical problem of power distribution imbalance in multi-pile charging systems caused by grid voltage fluctuations and harmonic pollution in the prior art.

[0004] The present invention is implemented as follows: the present invention provides a split-type DC charging one-to-many group charging control system, in which a group charging balancing optimization module is provided in the control chip, which is used to collect the three-phase voltage data of the power grid in real time through the voltage monitoring unit, calculate the voltage imbalance and voltage fluctuation amplitude, and collect the real-time current data of each charging pile terminal through the current sensor module, pre-process the collected three-phase voltage data and real-time current data of the power grid, establish a real-time power grid status database, use the sliding time window method to perform segmented analysis on the pre-processed real-time power grid status database, extract the power grid characteristic parameters in each time window, and based on the extracted power grid characteristic parameters, use multiple The target optimization equation group calculates the optimal power allocation matrix of each charging pile terminal to achieve load balancing and minimize grid impact. The calculated optimal power allocation matrix is ​​input into a pre-trained grid stability prediction model to predict grid stability indicators and charging efficiency parameters in the future time period. The filtering parameters of the harmonic filter are adjusted according to the predicted grid stability indicators. The charging power output of each charging pile terminal is optimized by dynamically adjusting the gating weight function. The charging status of each charging pile terminal and changes in grid parameters are monitored in real time. When a grid anomaly or charging pile failure is detected, the emergency power redistribution mechanism is immediately activated to ensure the overall stable operation of the system.

[0005] Among them, the sliding time window method specifically uses a data collection cycle of 30 seconds as the basic unit. Each time window contains grid status data of 900 sampling points, which is used to analyze the short-term change trend of grid parameters.

[0006] Among them, the grid characteristic parameters are specifically quantitative indicators of the grid operation status extracted from the real-time grid status database, including values ​​of four dimensions: voltage effective value, current effective value, power factor and harmonic content.

[0007] Among them, the optimal power allocation matrix is ​​specifically a two-dimensional data structure that describes the power allocation ratio of each charging pile terminal. The number of matrix rows corresponds to the number of charging pile terminals, and the number of columns corresponds to the power allocation time nodes.

[0008] Among them, the grid stability index is a comprehensive quantitative parameter for evaluating the stability of grid operation, including voltage deviation rate, frequency deviation rate, power factor deviation rate and total harmonic distortion rate.

[0009] Among them, the charging efficiency parameters are specifically quantitative indicators that describe the energy conversion efficiency of the charging pile terminal, including charging power utilization rate, energy loss rate and charging time efficiency.

[0010] Among them, the gated weight function is specifically a mathematical function that dynamically adjusts the weight of the neural network based on the grid parameters. The input parameters include the grid voltage fluctuation amplitude, the charging pile load rate and the harmonic content.

[0011] The emergency power redistribution mechanism is specifically a power redistribution strategy started when the system detects an anomaly, which ensures the overall stable operation of the system by reducing the power output of some charging piles.

[0012] The multi-objective optimization equation set includes a power balance equation and a power grid stability equation.

[0013] The power balance equation includes the current load rate of the power grid, charging demand of each pile, voltage fluctuation amplitude of the power grid, and total power capacity of the system.

[0014] The power grid stability prediction model is a time series prediction network based on a Transformer architecture, including an encoder layer, a decoder layer, and an attention mechanism layer.

[0015] The training data set of the power grid stability prediction model includes power grid operation data, charging pile operation data, and environmental parameter data under different working conditions.

[0016] The power grid stability prediction model is trained using the back propagation algorithm and the gradient descent optimizer to update the model parameters.

[0017] The gating weight function is designed as a dynamic adjustment function based on the three key parameters of the power grid voltage fluctuation amplitude, charging pile load rate, and harmonic content.

[0018] When the balance value is in the range of 0.5 to 0.75, a logarithmic function weight adjustment function is used.

[0019] Among them, the main control cabinet is a rectangular box structure with multiple isolation chambers inside. The power distribution unit is set in the first isolation chamber in the main control cabinet, the voltage monitoring unit is set in the second isolation chamber in the main control cabinet, and the control chip is set in the third isolation chamber in the main control cabinet.

[0020] The present invention effectively addresses the limitations of traditional static allocation strategies in complex power grid environments by establishing a real-time power grid status monitoring system and a dynamic power allocation algorithm based on artificial intelligence. The present invention uses a multi-dimensional sensor network to collect key parameters such as the three-phase voltage of the power grid, the current at each charging station, harmonic content, and frequency in real time. The present invention continuously analyzes the power grid status through a sliding time window method, extracting characteristic parameters such as the effective value of voltage, effective value of current, power factor, and harmonic content. A multi-objective optimization model, including a power balance equation and a power grid stability equation, is established. This model can calculate the optimal power allocation matrix for each charging station based on the real-time state of the power grid, achieving dynamic adjustment of load balancing and power allocation under power grid fluctuation conditions. The power grid stability prediction model based on the Transformer architecture introduced in the present invention can learn the temporal characteristics and complex laws of power grid operation, predict changes in power grid status in future time periods, and provide a scientific basis for advance adjustment. The designed four-segment partitioned gating weight function can dynamically select the most suitable control strategy based on the comprehensive evaluation results of the power grid voltage fluctuation amplitude, charging station load rate, and harmonic content, ensuring that each charging station can obtain reasonable power allocation under different power grid conditions, thus achieving balanced and stable overall system operation. In summary, the present invention solves the technical problem mentioned in the background technology that the grid voltage fluctuation and harmonic pollution lead to unbalanced power distribution in a multi-pile charging system. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 The figure is a schematic diagram of the composition of the system involved in the present invention.

[0022] Figure 2 Flowchart of the execution steps of the cluster charge balancing optimization module.

[0023] Figure 3 This is a graph showing the time variation of the grid voltage fluctuation amplitude in Example 2.

[0024] Figure 4 This is the distribution diagram of the dynamic adjustment of the charging pile power allocation coefficient in Example 2. DETAILED DESCRIPTION

[0025] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0026] like Figure 1The application provides a flow chart of a split type direct current charging multi-outlet group charging group control system, and the method comprises the following steps.

[0027] A split type direct current charging multi-outlet group charging group control system comprises a main control cabinet, a power distribution unit, a charging pile terminal, a current sensor module, a voltage monitoring unit, a harmonic filter, a communication interface module, a power metering device, a frequency detection module and a control chip.

[0028] The main control cabinet is in a rectangular box structure, and a plurality of isolated chambers are arranged in the main control cabinet; the power distribution unit is arranged in a first isolated chamber in the main control cabinet, and the power distribution unit comprises a plurality of power output interfaces; each power output interface is connected with a corresponding charging pile terminal through a high-voltage cable; the charging pile terminal is in a columnar structure, and an outer shell is made of aluminum alloy; a direct current output module and a man-machine interface are arranged in the charging pile terminal; the current sensor module is installed at each power output interface of the power distribution unit and is used for monitoring real-time output currents; the voltage monitoring unit is arranged in a second isolated chamber in the main control cabinet, is connected with a power grid incoming line end through a voltage acquisition cable and is used for monitoring voltage fluctuation of the power grid; the harmonic filter is in a cylindrical structure, and a plurality of groups of LC filter circuits and harmonic detection circuits are arranged in the harmonic filter; the harmonic filter is connected with an input end of the power distribution unit through a filter connection line; the communication interface module is arranged on a front panel of the main control cabinet and comprises an Ethernet interface, a serial communication interface and a wireless communication module; the power metering device is installed at the power grid incoming line end in the main control cabinet and is used for measuring power grid input power and a power factor; the frequency detection module is installed beside the voltage monitoring unit, is connected with the power grid incoming line end through a frequency detection cable and is used for monitoring real-time frequency of the power grid.

[0029] The control chip is arranged in a third isolated chamber in the main control cabinet; the control chip is electrically connected with the power distribution unit, the current sensor module, the voltage monitoring unit, the harmonic filter, the communication interface module, the power metering device and the frequency detection module and performs data interaction; a group charging equalization optimization module is arranged in the control chip and is used for realizing dynamic distribution of multi-pile charging power and compensation control of power grid voltage fluctuation, so that overall stability and efficiency of the charging system are finally improved.

[0030] The group charging equalization optimization module is used for executing the following steps:

[0031] S01, real-time three-phase voltage data of the power grid are collected by the voltage monitoring unit, voltage unbalance degree and voltage fluctuation amplitude are calculated, and real-time current data of each charging pile terminal are collected by the current sensor module;

[0032] S02. Preprocessing the collected three-phase voltage data and real-time current data of the power grid, including digital filtering, outlier detection and data standardization, to establish a real-time power grid status database;

[0033] S03, using the sliding time window method to perform segmented analysis on the pre-processed real-time power grid status database, and extracting the power grid characteristic parameters within each time window, including the effective value of voltage, effective value of current, power factor and harmonic content;

[0034] S04. Based on the extracted grid characteristic parameters, a multi-objective optimization equation group is used to calculate the optimal power allocation matrix of each charging pile terminal to achieve load balancing and minimize grid impact;

[0035] S05. Inputting the calculated optimal power allocation matrix into a pre-trained grid stability prediction model to predict grid stability indicators and charging efficiency parameters in a future time period;

[0036] S06. Adjust the filtering parameters of the harmonic filter according to the predicted grid stability index, and optimize the charging power output of each charging pile terminal by dynamically adjusting the gating weight function;

[0037] S07. Monitor the charging status of each charging pile terminal and changes in grid parameters in real time. When a grid anomaly or charging pile failure is detected, immediately activate the emergency power redistribution mechanism to ensure the overall stable operation of the system.

[0038] The sliding time window method specifically refers to a data collection cycle with 30 seconds as the basic unit. Each time window contains grid status data of 900 sampling points, which is used to analyze the short-term change trend of grid parameters.

[0039] The grid characteristic parameters specifically refer to the quantitative indicators of the grid operation status extracted from the real-time grid status database, including the values ​​of four dimensions: voltage RMS, current RMS, power factor and harmonic content.

[0040] The optimal power allocation matrix specifically refers to a two-dimensional data structure that describes the power allocation ratio of each charging pile terminal. The number of matrix rows corresponds to the number of charging pile terminals, and the number of columns corresponds to the power allocation time nodes.

[0041] The grid stability index specifically refers to the comprehensive quantitative parameters for evaluating the stability of grid operation, including voltage deviation rate, frequency deviation rate, power factor deviation rate and total harmonic distortion rate.

[0042] Charging efficiency parameters specifically refer to quantitative indicators that describe the energy conversion efficiency of the charging pile terminal, including charging power utilization rate, energy loss rate and charging time efficiency.

[0043] The gated weight function specifically refers to a mathematical function that dynamically adjusts the weights of the neural network based on grid parameters. The input parameters include the grid voltage fluctuation amplitude, charging pile load rate and harmonic content.

[0044] The emergency power redistribution mechanism specifically refers to the power redistribution strategy that is activated when the system detects an abnormality, which ensures the overall stable operation of the system by reducing the power output of some charging piles.

[0045] The multi-objective optimization equation group includes a power balance equation and a power grid stability equation; the power balance equation is used to calculate the optimal power distribution ratio of each charging pile terminal, and the input includes the current load rate of the power grid, the charging demand of each pile position, the power grid voltage fluctuation amplitude, and the total power capacity of the system, and the output is the power distribution coefficient of each pile position; the power grid stability equation is used to evaluate the impact of the power distribution scheme on the power grid stability, and the input includes the power grid voltage imbalance, the total harmonic distortion rate, the power factor deviation value, and the load change rate, and the output is the power grid stability evaluation index.

[0046] The current load rate of the power grid is calculated by the ratio of the power grid input power measured by the power metering device to the rated power of the power distribution unit. The charging demand of each pile position is obtained by inputting the human-computer interaction interface of the charging pile terminal. The power grid voltage fluctuation amplitude is calculated by analyzing the three-phase voltage data of the power grid collected by the voltage monitoring unit. The total power capacity of the system is the rated power output capacity of the power distribution unit. The power grid voltage imbalance is calculated by the three-phase voltage data of the power grid collected by the voltage monitoring unit. The total harmonic distortion rate is obtained by the harmonic detection circuit built into the harmonic filter. The power factor deviation value is calculated by comparing the power factor measured by the power metering device with the standard power factor. The load change rate is calculated by comparing the load data of two consecutive time windows. The frequency deviation rate is calculated by the difference between the power grid frequency measured by the frequency detection module and the standard frequency of 50Hz.

[0047] The power allocation coefficient of each charging pile position is used to control the power ratio allocated by the power distribution unit to each charging pile terminal, and the power grid stability evaluation index is used to evaluate whether the current power allocation scheme meets the requirements for stable operation of the power grid.

[0048] The specific structure of the power grid stability prediction model is a time series prediction network based on a Transformer architecture, including an encoder layer, a decoder layer, and an attention mechanism layer, wherein the hierarchical fusion weight of the attention mechanism layer is dynamically adjusted according to three parameters of the power grid voltage fluctuation amplitude, the charging pile load rate, and the harmonic content, and when the power grid state is stable, the hierarchical fusion weight is biased towards historical data features, and when the power grid state fluctuates, the hierarchical fusion weight is biased towards real-time data features; the steps of establishing the training data set of the power grid stability prediction model specifically include collecting power grid operation data, charging pile operation data, and environmental parameter data under different working conditions, establishing a training sample set containing multi-dimensional features such as power grid voltage fluctuation, load change, harmonic pollution, and power factor change, expanding the sample number to 1 million groups through data enhancement technology, and labeling according to time sequence characteristics to form a supervised learning data set corresponding to input and output; the steps of training the power grid stability prediction model specifically include updating model parameters using a back propagation algorithm and a gradient descent optimizer, setting the learning rate to 0.001, the batch size to 64, and the training rounds to 1000, using an early stopping mechanism to prevent overfitting during training, evaluating model performance using a validation set, stopping training when the validation set loss function does not decrease for 10 consecutive rounds, and finally obtaining a neural network model that can accurately predict power grid stability.

[0049] The power grid voltage fluctuation amplitude is calculated from three-phase voltage data collected by a voltage monitoring unit, the charging pile load rate is calculated from the ratio of real-time current data collected by a current sensor module to the rated current of the charging pile terminal, and the harmonic content is obtained by analyzing power grid voltage and current waveforms using a harmonic detection circuit built into a harmonic filter.

[0050] The gating weight function is designed as a dynamic adjustment function based on three key parameters of the power grid voltage fluctuation amplitude, the charging pile load rate, and the harmonic content, and a balance value is obtained by calculating the weighted average of the three parameters, when the balance value is in the range of 0 to 0.25, a linearly increasing weight adjustment function is used, when the balance value is in the range of 0.25 to 0.5, a quadratic function weight adjustment function is used, when the balance value is in the range of 0.5 to 0.75, a logarithmic function weight adjustment function is used, and when the balance value is in the range of 0.75 to 1.0, an exponential decay weight adjustment function is used, and the switching of different weight adjustment functions realizes accurate control of the hierarchical fusion weight in the power grid stability prediction model.

[0051] It should be noted that the gating weight function adopts a four-section partition adjustment strategy, and the corresponding weight adjustment function is selected according to the different ranges of the balance value, which fully considers the complexity of the power grid operating state and the response characteristics of the charging system.

[0052] When the balance value is between 0 and 0.25, the system uses a linearly increasing weight adjustment function. At this stage, the grid is relatively stable, with minimal voltage fluctuations, low charging pile load, and normal harmonic content. The linearly increasing function is simple and intuitive, with smooth and predictable weight changes, which promotes stable system operation. During this stage, the neural network's hierarchical fusion weights are gradually increased, allowing the system to gradually adapt to load changes and preventing sudden weight jumps from impacting the charging process.

[0053] When the balance value is between 0.25 and 0.5, the system uses a quadratic weight adjustment function. At this point, the grid begins to fluctuate slightly, and the charging pile load factor increases, requiring a more refined control strategy. The quadratic function has nonlinear characteristics, enabling a more sensitive response under moderate load conditions. The curvature of the quadratic function causes the weight adjustment to change relatively slowly initially, but accelerates as the peak value is approached. This characteristic perfectly matches the transition from stable to fluctuating grid loads.

[0054] When the balance value is between 0.5 and 0.75, the system uses a logarithmic weight adjustment function. At this point, the grid enters a medium-to-high load state, with increased voltage fluctuations and rising harmonic content, requiring a more cautious control strategy. The logarithmic function exhibits rapid initial changes followed by a gradual flattening. This characteristic helps the system respond quickly to rapid load changes while maintaining relatively stable control under high load conditions. The logarithmic function's convergence properties ensure that weight adjustments are not overly aggressive, preventing system oscillations.

[0055] When the balance value is between 0.75 and 1.0, the system adopts an exponential decay weight adjustment function. At this point, the grid's operating conditions deteriorate significantly, with severe voltage fluctuations, charging pile loads approaching full capacity, and severe harmonic pollution. The exponential decay function's rapid decline can quickly reduce the weight of unstable factors in the neural network, prioritizing safe system operation. This sharp decline allows the system to quickly enter protection mode, maintaining basic charging functionality by reducing the weight of uncertain factors.

[0056] The core principle of this four-segment design lies in matching the nonlinear characteristics of power grid operation. The power grid itself is a complex nonlinear system, exhibiting distinct stability and operating characteristics under varying load conditions. By using different mathematical functions to describe the characteristics of each operating range, it can better adapt to the dynamic changes of the power grid.

[0057] Furthermore, this partitioning strategy considers the safety requirements of the charging system. It optimizes efficiency at low loads, balances efficiency and stability at medium loads, and prioritizes safety at high loads. Four different function forms provide the system with a rich selection of control strategies, enabling the cluster charging balancing optimization module to find the most appropriate control parameters under various operating conditions, thereby achieving the optimal balance between grid stability and charging efficiency.

[0058] The specific implementation of the above steps is described in detail below. Step S01 is implemented by using a voltage monitoring unit (VMU) to collect instantaneous values ​​of the three-phase grid voltage in real time using a high-precision voltage sensor array. The sampling frequency is set to 30kHz to ensure that even small fluctuations in the grid voltage can be captured. The VMU's built-in digital signal processor performs fast Fourier transform analysis on the collected voltage data, calculating the effective value, phase difference, and spectral characteristics of each phase voltage. Voltage imbalance is calculated using the negative-sequence component method, which decomposes the three-phase voltage into positive-sequence, negative-sequence, and zero-sequence components through symmetrical component transformation. The ratio of the negative-sequence component to the positive-sequence component is used as the voltage imbalance indicator. When this ratio exceeds 2%, a voltage imbalance warning is triggered. Voltage fluctuation amplitude is calculated using a sliding average algorithm, calculating the standard deviation of the effective voltage value within a 1-second time window. Significant voltage fluctuation is considered present when the standard deviation exceeds 1% of the rated voltage. The current sensor module uses a high-precision Hall-effect current sensor with a measurement accuracy of 0.2% and a response time of less than 1 millisecond, enabling real-time capture of current trends at each charging station terminal. Each current data is digitized by a 16-bit analog-to-digital converter, and the sampling frequency is synchronized with the voltage sampling to ensure the time consistency of the voltage and current data.

[0059] The specific implementation of step S02 involves performing multi-stage preprocessing on the collected raw data to improve data quality and analysis accuracy. Digital filtering utilizes a Butterworth low-pass filter to remove high-frequency noise interference. The cutoff frequency is set to 5kHz, and the filter order is 8th. This effectively suppresses high-frequency harmonics generated by switching power supplies and power electronics. Outlier detection utilizes the statistically based 3σ criterion to calculate the mean and standard deviation of the data sequence. Data points that deviate from the mean by more than three standard deviations are marked as outliers. The outlier detection window is set to 5 seconds, and a data quality alert is triggered when the proportion of outliers exceeds 10%. Data normalization utilizes a maximum-minimum normalization method to map data of different physical quantities to a uniform range of 0 to 1, facilitating subsequent mathematical operations and pattern recognition. The real-time power grid status database utilizes a ring buffer structure to store preprocessed data. The buffer capacity is set to store the most recent hour of historical data. The data storage format utilizes compression encoding technology to reduce storage space while supporting rapid data retrieval and analysis.

[0060] The specific implementation of step S03 involves segmented feature extraction from grid status data using a sliding time window method. The time window length is set to 30 seconds, with a window sliding step of 5 seconds, ensuring a 25-second overlap between adjacent time windows. This overlapping design improves the continuity and stability of feature extraction. Each time window contains 900 sampling points of grid status data. The frequency domain characteristics of the voltage and current signals are analyzed using the Fast Fourier Transform algorithm to extract the amplitude and phase information of the fundamental component and each harmonic component. The RMS voltage value is calculated using the root mean square (RMS) algorithm, squaring, averaging, and square rooting the instantaneous voltage value within the time window. The RMS current value is calculated using the same RMS algorithm for each charging pile terminal. The power factor is calculated based on the phase difference between voltage and current. The zero-point detection method is used to measure the time difference between the zero crossings of the voltage and current waveforms, and the power factor value is calculated from this phase difference. The harmonic content analysis uses a harmonic analysis algorithm to calculate the total harmonic distortion (THD) of the 2nd to 50th harmonic components. Harmonic mitigation measures are initiated when the THD exceeds 5%.

[0061] The specific implementation of step S04 involves establishing a mathematical model for power distribution based on multi-objective optimization theory and employing a genetic algorithm to solve the optimal power allocation matrix. The power balance equation aims to minimize the uneven distribution of power across charging pile terminals. Constraints include the sum of the power allocation coefficients at each charging pile being equal to 1, the maximum power per pile not exceeding 95% of the rated power, and the total system power not exceeding the rated capacity of the power distribution unit. The grid stability equation aims to minimize grid shocks. Constraints include voltage imbalance less than 2%, total harmonic distortion less than 5%, and a power factor above 0.9. The genetic algorithm uses a population size of 100 individuals, each representing a power allocation solution. The algorithm iteratively optimizes through selection, crossover, and mutation operations. The maximum number of iterations is set to 500, and the convergence criterion is that the optimal solution remains unchanged by less than 0.1% over 50 consecutive generations. The fitness function comprehensively considers power balance and grid stability indicators, taking the form of a weighted sum. The weight coefficients are dynamically adjusted based on the current grid operating status, with a higher weight given to power balance when the grid is stable and a higher weight given to stability when the grid is experiencing disturbances.

[0062] The specific implementation of step S05 involves feeding the optimal power allocation matrix as input data into a grid stability prediction model to predict stability for future time periods. The prediction model, based on a long-short-term memory neural network architecture, is capable of capturing the temporal variations and long-term dependencies of grid status. The model inputs include the current power allocation matrix, historical grid status parameters, and environmental factor data. The output is a 15-minute forecast of grid stability indicators and charging efficiency parameters. Grid stability indicators include voltage deviation rate, frequency deviation rate, power factor deviation rate, and total harmonic distortion rate, with a prediction accuracy requirement exceeding 90%. Charging efficiency parameters include charging power utilization rate, energy loss rate, and charging time efficiency; these parameters reflect the operational effectiveness and economic efficiency of the charging system. Model inference utilizes batch processing, processing 16 time steps of data at a time. The inference time is kept within 100 milliseconds, meeting the latency requirements of real-time control.

[0063] The specific implementation of step S06 involves dynamically adjusting the harmonic filter parameters and charging power output based on the grid stability indicator output by the prediction model. Harmonic filter parameter adjustment utilizes an adaptive filtering algorithm, adjusting the filter's center frequency and quality factor based on the predicted harmonic content trend. The filter utilizes a multi-cascade structure, including 2nd, 3rd, 5th, and 7th harmonic filter branches. The inductor and capacitor parameters of each branch can be adjusted online via electronic switches. The gated weight function calculates the grid voltage fluctuation amplitude, charging pile load factor, and harmonic content as input parameters, and outputs the weight adjustment coefficients for each layer of the neural network. The gated weight function utilizes a piecewise function design, employing different mathematical functions for weight calculation when the weighted average of the input parameters falls within different intervals, ensuring optimal control under all operating conditions. Charging power output regulation is achieved through pulse width modulation technology, with an adjustment accuracy of 1% of the rated power and a response time of less than 50 milliseconds, enabling rapid tracking of changes in power allocation instructions.

[0064] The specific implementation of step S07 involves establishing a multi-level condition monitoring and fault diagnosis system to enable rapid detection and emergency response to system anomalies. Condition monitoring utilizes multi-sensor fusion technology, integrating the changing trends of multiple physical quantities such as voltage, current, temperature, and vibration to assess system health. Grid anomaly detection is based on statistical process control theory, establishing control charts for grid parameters. When monitored parameters exceed control limits, an anomaly alarm is triggered. The control limits are set to the mean plus or minus three standard deviations. Charging pile fault diagnosis utilizes a model-based fault detection approach, identifying the fault type and severity by comparing the deviation between actual operating parameters and theoretical model outputs. The emergency power redistribution mechanism employs a hierarchical response strategy, with three levels of response depending on the severity of the anomaly: Level 1 is a minor adjustment of the power output of a single charging pile; Level 2 is the suspension of some charging piles; and Level 3 is an emergency system shutdown for protection. The power redistribution algorithm, based on the principle of maximizing remaining capacity, prioritizes the charging needs of critical loads while ensuring stable grid operation. The redistribution calculation time is kept within one second, meeting the real-time requirements of emergency response.

[0065] The power grid stability prediction model uses a deep neural network structure based on the Transformer architecture, which has powerful sequence modeling capabilities and parallel computing advantages. The model mainly consists of three core components: the encoder layer, the decoder layer, and the multi-head attention mechanism layer. The encoder layer is responsible for extracting feature representations of the input data and contains six identical encoder blocks. Each encoder block consists of a multi-head self-attention sublayer and a feedforward neural network sublayer. Residual connections and layer normalization techniques are used to improve training stability. The decoder layer is responsible for generating the predicted output and also contains six decoder blocks. Each decoder block has an encoder-decoder attention sublayer added to it to associate input features with output predictions. The multi-head attention mechanism uses eight attention heads, each with a dimension of 64, which can capture feature relationships in different subspaces in parallel. The hierarchical fusion weights of the attention mechanism layer are dynamically adjusted according to the operating status of the power grid. When the grid voltage fluctuation amplitude is less than 1%, the charging pile load rate is less than 50%, and the harmonic content is less than 3%, the hierarchical fusion weights are biased towards historical data characteristics, and the weight distribution ratio is 70% for historical data and 30% for real-time data; when the power grid status fluctuates greatly, the weight distribution is adjusted to 30% for historical data and 70% for real-time data, ensuring that the model can quickly respond to changes in the power grid status.

[0066] The training dataset construction process involves multiple steps, including data collection, feature engineering, data augmentation, and annotation. The data collection phase involves collecting operational data from power grid systems across different regions, seasons, and load levels, covering a wide range of operating conditions, including normal operation, minor disturbances, and severe faults. The total raw data volume reaches 10 TB. Grid operational data includes basic electrical parameters such as three-phase voltage, three-phase current, active power, reactive power, and frequency. Charging pile operational data includes information such as charging current, voltage, power, and charging status at each charging station. Environmental parameters include external factors affecting grid operation, such as temperature, humidity, and wind speed. The feature engineering phase preprocesses and extracts features from the raw data, calculating derived features such as voltage imbalance, total harmonic distortion, and power factor to generate a feature vector containing 156 feature dimensions. Data augmentation techniques employ methods such as noise injection, time warping, and amplitude scaling to expand the training sample size, increasing the number of original samples from 100,000 to 1 million, enhancing the model's generalization and robustness. During the data labeling process, power system experts label the stability level of each sample based on grid stability theory. The labeled levels are divided into five levels: excellent, good, general, poor, and dangerous, forming the input-output correspondence required for supervised learning.

[0067] The model training adopts a phased training strategy. First, a pre-training phase is carried out, using large-scale unlabeled data for self-supervised learning to learn the inherent laws and representation methods of power grid data; then a fine-tuning phase is carried out, using labeled data for supervised learning to optimize the prediction accuracy of the model. The training process uses the Adam optimizer, the initial value of the learning rate is set to 0.001, and the cosine annealing strategy is used to dynamically adjust the learning rate, with a minimum learning rate of 10 -6 . The batch size is set to 64, the number of training rounds is 1000 rounds, and each round contains 1562 batches. To prevent overfitting, an early stopping mechanism is used to monitor the validation set loss function, and training is automatically stopped when the validation set loss does not decrease for 10 consecutive rounds. The loss function adopts a weighted combination of mean square error and cross entropy loss with a weight ratio of 7:3, taking into account both prediction accuracy and classification accuracy. Gradient clipping technology is used during training to prevent gradient explosion, and the clipping threshold is set to 1.0. The final trained model has a prediction accuracy of 92.3% on the test set, which meets the accuracy requirements of practical applications.

[0068] It should be noted that the first technical idea of ​​the present invention is a multidimensional feature extraction method based on a sliding time window. This method realizes the continuity analysis and feature extraction of power grid state parameters through the design of a 30-second time window and a 5-second sliding step. Compared with the traditional fixed time point sampling method, the sliding window method can capture the dynamic change trend of power grid parameters and avoid the interference of instantaneous outliers on system judgment. The window overlapping design ensures the continuity of feature extraction and eliminates the problem of information loss caused by time segmentation. Multidimensional feature extraction covers key power grid parameters such as voltage, current, power factor, harmonic content, etc., forming a complete power grid status portrait, which provides a rich information basis for subsequent optimization decisions.

[0069] The second technical approach is a power allocation strategy that combines multi-objective optimization with genetic algorithms. This strategy takes power balance and grid stability as dual optimization objectives, and finds the optimal power allocation solution through the global optimization capability of the genetic algorithm. Compared with traditional single-objective optimization methods, multi-objective optimization can ensure charging efficiency while taking into account grid stability, avoiding the problem of local optimal solutions. The random search characteristics of the genetic algorithm enable the system to escape the local optimum and find the global optimal solution, which is particularly suitable for dealing with complex optimization problems under multiple constraints. The dynamic weight adjustment mechanism adjusts the importance of the optimization objectives in real time according to the operating status of the grid, realizing adaptive optimization control under different working conditions.

[0070] The third technical approach is a Transformer-based grid stability prediction model. This model utilizes an advanced attention mechanism to capture the complex relationships between grid parameters and is capable of handling long-sequence data and multivariate time series prediction. Compared to traditional time series prediction methods, the Transformer architecture boasts stronger feature learning capabilities and improved generalization performance. The multi-head attention mechanism simultaneously focuses on feature relationships across different subspaces, enhancing the model's expressive power and prediction accuracy. The dynamic adjustment of hierarchical fusion weights enables the model to adaptively balance the importance of historical and real-time data based on grid status, improving both prediction accuracy and real-time performance.

[0071] The fourth technical approach is the design of a piecewise gating weight function. This function divides the control strategy into four intervals based on the grid's operating state, with each interval using a different mathematical function for weight adjustment. Compared to traditional linear control methods, piecewise function design can better match the nonlinear characteristics of the grid system and provide a more refined control strategy. The four function forms of linear, quadratic, logarithmic, and exponential cover the entire process of the grid from stability to deterioration, ensuring optimal control effects under various operating conditions. This design fully considers the complexity of the grid's operating state and the response characteristics of the charging system, achieving an organic unity of precise control and system safety.

[0072] The synergy of these four technical approaches forms a complete intelligent group charging control system. Sliding window feature extraction provides the system with accurate state perception capabilities, multi-objective optimization ensures scientific and comprehensive decision-making, the Transformer prediction model provides a forward-looking control basis, and the piecewise gating function enables refined execution control. The organic combination of these four technologies realizes closed-loop control from state perception, decision optimization, predictive analysis, to precise execution, with higher control accuracy, stronger adaptability, and better stability than existing technologies. This synergy not only improves the operating efficiency of the charging system, but also enhances the stability and reliability of the power grid, providing advanced technical support for the construction and operation of large-scale electric vehicle charging infrastructure.

[0073] Specifically, the principle of the present application is that the present application can solve the problem of uneven power distribution caused by power grid voltage fluctuation and harmonic pollution, and the core is to build a complete power grid state perception and intelligent power regulation system. The root cause of the uneven power distribution caused by power grid voltage fluctuation and harmonic pollution is that the traditional system lacks accurate perception of the real-time state of the power grid and dynamic response capability. The present application realizes real-time collection of key parameters such as three-phase voltage of the power grid, current of each stake, harmonic content and frequency through a multi-dimensional sensor network including a voltage monitoring unit, a current sensor module, a harmonic filter and a frequency detection module, etc., providing comprehensive and accurate power grid state information for the system. The collected data is analyzed by using the sliding time window method, and each 30-second time window contains 900 sampling points, which can effectively capture the short-term change trend and dynamic characteristics of the power grid parameters, avoiding the influence of randomness of instantaneous sampling, and providing a reliable data basis for subsequent intelligent decision-making. The multi-objective optimization model established includes a power balance equation and a power grid stability equation. The power balance equation calculates the optimal power distribution coefficient according to the power grid load rate, the demand of each stake, the voltage fluctuation amplitude and the system capacity, and the power grid stability equation evaluates the impact of the distribution scheme on the operation of the power grid. The two equations are mutually constrained and balanced to ensure that the power distribution meets the demand of each stake and does not impact the power grid. The power grid stability prediction model based on the Transformer architecture introduced can learn the complex rules and time sequence characteristics of the power grid operation through the encoder-decoder structure and the attention mechanism, and predict the changes of the power grid state in the future period, providing a scientific basis for the system to adjust in advance. The gating weight function adopts a four-section partition strategy, which dynamically selects different weight adjustment functions such as linear increase, quadratic function, logarithmic function or exponential decay according to the comprehensive balance value of the power grid voltage fluctuation amplitude, the load rate of the charging pile and the harmonic content. This design fully considers the nonlinear characteristics of the power grid system, so that the system can find the most suitable control parameters in different operating states, realize the dynamic balance of power distribution of each charging pile, and effectively solve the problem of uneven power distribution caused by power grid fluctuation and harmonic pollution.

[0074] A specific embodiment 1 of the present application is provided below, and the specific implementation of each step in the embodiment 1 is described in detail as follows.

[0075] The specific implementation of step S01 is to collect three-phase voltage data of the power grid in real time through a voltage monitoring unit. The voltage unbalance degree U unb The calculation formula is:

[0076]

[0077] In the formula, U2 is the negative sequence voltage component, and the unit is V; U1 is the positive sequence voltage component, and the unit is V. The negative sequence component U2 is obtained by symmetric component transformation, and the specific calculation is as follows:

[0078]

[0079] Where U A 、U B 、U C They are the instantaneous values ​​of phase A, phase B, and phase C voltages, respectively, acquired in real time by voltage sensors, in units of V; a = e j2π / 3 is a unit complex operator, where j is an imaginary unit. Voltage fluctuation amplitude ΔU fluc The calculation uses the sliding standard deviation method:

[0080]

[0081] Where U rms，i is the effective value of the voltage at the i-th sampling point, in V; is the mean value of the effective voltage in the time window, calculated as The unit is V; n is the number of sampling points, which is set to 30; i is the sampling point number, ranging from 1 to n. The current data I collected by the current sensor module for each charging pile terminal k Obtained by a Hall effect sensor, where k represents the kth charging pile terminal, k=1, 2, ..., m, and m is the total number of charging pile terminals.

[0082] The specific implementation of step S02 is to pre-process the collected data, and the outlier detection adopts the 3σ criterion, and the judgment condition is:

[0083] |x data,i -μ data |>3σ data ;

[0084] Where x data，i is the value of the i-th data point; μ data is the mean of the data series, calculated as σ data is the standard deviation of the data series, calculated as N is the total length of the data sequence. Data normalization uses maximum and minimum value normalization:

[0085]

[0086] Where x norm is the normalized value, ranging from 0 to 1; x raw is the original input value; x min and x max are the minimum and maximum values ​​of the data series, respectively.

[0087] The specific implementation of step S03 is to use the sliding time window method to extract the grid characteristic parameters, voltage RMS

[0088] U rms The calculation formula is:

[0089]

[0090] Where u(t) is the instantaneous voltage function, in V; T is the integration time, set to 30s; t is the time variable, in s. rms Using the same calculation method:

[0091]

[0092] Where i(t) is the instantaneous value of the current in A. The power factor cosφ is calculated from the phase difference:

[0093] cosφ=cos(φ u -φ i );

[0094] Where, φ u is the voltage phase, obtained by the zero-point detection method, in rad; φ i is the current phase, obtained by the zero-point detection method, in rad; φ is the phase difference between voltage and current, in rad. The total harmonic distortion THD calculation formula is:

[0095]

[0096] Where U h is the effective value of the h-th harmonic voltage, obtained by fast Fourier transform, in V; u1 is the effective value of the fundamental voltage, in V; H is the highest harmonic order, set to 50; h is the harmonic order, ranging from 2 to H.

[0097] The specific implementation of step S04 is to establish a multi-objective optimization equation group to solve the optimal power allocation matrix. The power balance equation is:

[0098]

[0099] Where, P k is the allocated power of the kth charging pile, in kW; To distribute power evenly, we can calculate The unit is kW; m is the number of charging piles; f1 is the power balance objective function. The grid stability equation is:

[0100]

[0101] In the formula, w1, w2, w3 are weight coefficients, respectively taking 0.4, 0.3, 0.3, and all being dimensionless; cosφ ref is a reference power factor, set to 0.95, and dimensionless; f2 is a power grid stability target function. The optimal power distribution matrix P is expressed as:

[0102]

[0103] In the formula, P ij represents the power distribution value of the i th charging pile at the j th time node, with the unit of kW; n is the number of time nodes; i is the charging pile serial number, ranging from 1 to m; j is the time node serial number, ranging from 1 to n.

[0104] The specific implementation of step S05 is the same as the foregoing, and will not be described in detail here.

[0105] The specific implementation of step S06 is to dynamically adjust the neural network weight by using a gating weight function, and the balance value β calculation formula is:

[0106] β = α1ΔU fluc + α2ρ load + α3THD;

[0107] In the formula, α1, α2, α3 are weight coefficients, respectively taking 0.4, 0.3, 0.3, and all being dimensionless; ρ load is a charging pile load rate, calculated as ρ load = I actual / I rated , wherein I actual is an actual operating current of the charging pile, with the unit of A, and I rated is a rated current of the charging pile, with the unit of A. The gating weight function g(β) adopts a piecewise function form:

[0108] When 0≤β<0.25, g(β)=k1β+b1;

[0109] When 0.25≤β<0.5, g(β)=k2β 2 +b2;

[0110] When 0.5≤β<0.75, g(β)=k3ln(β+1)+b3;

[0111] When 0.75≤β≤1.0, g(β)=k4e -λβ +b4;

[0112] Where k1, k2, k3, and k4 are slope coefficients, which are 2.0, 1.5, 0.8, and 0.6, respectively, and are all dimensionless; b1, b2, b3, and b4 are intercept parameters, which are 0.1, 0.15, 0.2, and 0.25, respectively, and are all dimensionless; λ is the exponential decay coefficient, which is 3.0 and is dimensionless.

[0113] The specific implementation of step S07 is to establish an emergency power redistribution mechanism. The power redistribution algorithm is based on the principle of maximum remaining capacity. The power after redistribution is The calculation formula is:

[0114]

[0115] Where, is the original distributed power, in kW; η k is the adjustment coefficient of the kth charging pile, dimensionless, and is calculated as:

[0116]

[0117] Where, is the remaining capacity of the kth charging pile, which is measured in real time by the capacity detection module and is expressed in kWh. is the total capacity of the kth charging pile, in kWh; γ is the safety margin coefficient, which is 0.85 and dimensionless.

[0118] The voltage unbalance calculation formula is based on the symmetrical component transformation theory, where the symmetrical component transformation The degree of voltage imbalance is quantified by decomposing the three-phase asymmetric voltage into positive sequence, negative sequence, and zero sequence components. This formula takes into account the combined effects of the three-phase voltage amplitude and phase. The ratio of the negative sequence component to the positive sequence component can accurately reflect the imbalance state of the three-phase voltage of the power grid. Compared with the traditional simple amplitude comparison method, it has higher accuracy and a stronger theoretical basis, providing a reliable quantitative indicator for power grid stability assessment. Voltage fluctuation amplitude calculation formula The sliding standard deviation method can reflect the degree of voltage fluctuation in real time and quantify the discrete degree of voltage change into specific values ​​through statistical principles. Compared with the peak detection method, it can better suppress the influence of transient interference and improve the stability and accuracy of voltage fluctuation detection.

[0119] Power balance equation Based on the principle of least squares, the load balancing is realized by minimizing the sum of squares of the deviation of each charging pile power allocation from the average value. The equation considers the overall power allocation fairness of the system, avoiding the situation of single pile overload or light load. Compared with the traditional average allocation method, it can dynamically adjust according to the actual demand, improving the overall efficiency and stability of the charging system. Power grid stability equation The three key stability indicators of voltage unbalance, harmonic distortion rate and power factor deviation are comprehensively considered. Multiple physical quantities are unified into a single optimization target through weighted summation. The setting of weight coefficient is based on the stability theory of power system. Compared with the single index optimization method, it can comprehensively evaluate the operation state of power grid, ensuring that the power allocation scheme meets the load demand and ensures the stable operation of power grid.

[0120] The gating weight function g(β) adopts a piecewise function design, which adjusts the weight according to the different operation states of the power grid. This design is based on the nonlinear characteristics of the power grid system and the piecewise control principle in control theory. The linear function g(β) = k1β + b1 is suitable for gradual adjustment in stable state, the quadratic function g(β) = k2β 2 +b2 provides smooth transition under medium load, the logarithmic function g(β) = k3ln(β + 1) + b3 provides convergence control at high load, and the exponential decay function g(β) = k4e -λβ +b4 provides fast response in extreme state. Compared with the traditional linear control method, it can more accurately match the dynamic characteristics of the power grid, improving the adaptability and response speed of the control system. Emergency power redistribution formula Based on the capacity constraint optimization theory, the adjustment range is determined by the product of the residual capacity ratio and the safety margin coefficient. This formula not only considers the actual carrying capacity of the equipment but also reserves a safety margin. Compared with the fixed proportion adjustment method, it can be differentiated according to the actual situation, and maximize the maintenance of charging service capacity under the premise of ensuring system safety.

[0121] It should be noted that, in this embodiment, the following hardware equipment can be used: the main control cabinet uses the Chint Electric XL-21 series low-voltage power distribution cabinet, the power distribution unit uses the Huichuan Technology MD310T4B variable frequency speed regulator as the power control core, the charging pile terminal uses the Teladian TCDZ-DCO.7 series DC charging pile, the current sensor module uses the Haiguang Electronics HIT-200 Hall current sensor, the voltage monitoring unit uses the Beijing Sifang Relay Protection CSC-326 voltage monitoring device, the harmonic filter uses the Beijing Haidexin HDX-APF300 active power filter, the communication interface module uses the Advantech ADAM-3600 industrial communication gateway, the power metering device uses the Wasion Group DTSD341-MC3 three-phase smart electricity meter, the frequency detection module uses the Qingdao Qingzhi QS5100 frequency measurement module, and the control chip core uses the STMicroelectronics STM32H757AI dual-core microcontroller, which is based on ARM The Cortex-M7 and Cortex-M4 architectures, with main frequencies of 480MHz and 240MHz, support the STM32Cube.AI neural network deployment framework, fully meeting the real-time inference computing needs of the Transformer architecture neural network model in the cluster charging balance optimization module.

[0122] To better understand and implement the present invention, Example 2 of a specific application scenario is provided below: A technical team developed a split-type DC charging system with a single-to-multiple charging cluster to address the practical challenges of large grid load fluctuations, concentrated charging demand, and high power quality requirements in a region. The system includes a main control cabinet, a power distribution unit, and eight charging pile terminals, with a total installed capacity of 480kW.

[0123] After the system was installed, the technical team first conducted detailed testing of the basic grid parameters. The three-phase grid voltage was rated at 380V. Continuous monitoring over 24 hours using the voltage monitoring unit revealed that the grid voltage fluctuated between 370V and 390V, with voltage imbalance varying between 0.8% and 2.3%. The frequency detection module measured a stable grid frequency between 49.8Hz and 50.2Hz. The power metering device indicated that the power factor at the access point fluctuated between 0.85 and 0.92, and the total harmonic distortion rate detected by the harmonic filter ranged from 2.1% to 4.8%.

[0124] Before the system was officially put into operation, the technical team configured the parameters of the group charging balance optimization module. The sliding time window was set to 30 seconds, each window contained 900 sampling points, and the sampling frequency was 30kHz. The digital signal processor processing capacity of the voltage monitoring unit was 2.7×10 7The current sensor module has a measurement accuracy of 0.2% and a response time of 0.8ms, ensuring accurate capture of current changes at each charging station.

[0125] On the first day of system operation, eight charging pile terminals were connected simultaneously, with charging demands of 45kW, 38kW, 52kW, 41kW, 48kW, 35kW, 44kW, and 39kW, respectively. The cluster charging balance optimization module calculated the power allocation coefficient for each charging pile based on parameters such as the current grid load of 73%, a grid voltage fluctuation of 1.8%, and a total system power capacity of 480kW, as shown in Table 1.

[0126] Table 1 Power distribution parameters of each charging pile

[0127] Charging pile number Charging power requirement / kW Power allocation coefficient Actual distributed power / kW Charging station 1 45 0.128 46.2 Charging station 2 38 0.115 41.5 Charging station 3 52 0.135 48.6 Charging station 4 41 0.120 43.2 Charging station 5 48 0.131 47.3 Charging station 6 35 0.109 39.3 Charging station 7 44 0.126 45.4 Charging station 8 39 0.118 42.6

[0128] As can be seen from Table 1, the system optimizes power allocation according to the grid status and charging demand. The total allocated power is 354.1kW and the system power utilization rate reaches 73.8%.

[0129] During the operation of the system, the technical team focused on monitoring the performance of the power grid stability prediction model. The model is based on the Transformer architecture and consists of an encoder layer, a decoder layer, and 8 attention heads, each with a dimension of 64. The model input has 156 feature dimensions, including key parameters such as the grid voltage fluctuation amplitude, charging pile load rate, and harmonic content. In actual operation, when the grid voltage fluctuation amplitude is 1.8%, the average charging pile load rate is 73.8%, and the harmonic content is 3.2%, the weighted average calculation result of the gated weight function is 0.32, which is within the range of 0.25 to 0.5. The system automatically adopts the quadratic function weight adjustment function.

[0130] like Figure 3 As shown in the figure, the change trend of the grid voltage fluctuation amplitude during the operation of the system shows obvious periodic characteristics. The peak load period is from 9:00 to 11:00 in the morning and from 15:00 to 17:00 in the afternoon, and the voltage fluctuation amplitude is relatively large. Figure 4 As shown in the figure, the power distribution coefficient of each charging pile is dynamically adjusted over time. When the power grid is stable, the power distribution of each pile is relatively even. When the power grid fluctuates, the system gives priority to ensuring the stable power supply of important loads.

[0131] On the third day of the system's operation, the technical team encountered a typical emergency power redistribution scenario. At 2:30 PM that day, the grid voltage suddenly dropped to 365V, the voltage imbalance rose to 3.1%, and the total harmonic distortion rate reached 6.2%, triggering the system's emergency response mechanism. Upon detecting the anomaly, the cluster charging balance optimization module immediately activated secondary response mode, suspending operation of three charging stations and redistributing power to the remaining five, ensuring stable grid operation. The entire emergency response process took 0.8 seconds, with the power redistribution calculation time being 0.6 seconds, meeting real-time control requirements.

[0132] After a week of system operation, the technical team conducted a statistical analysis of the grid stability prediction model's accuracy. The model achieved 91.7% accuracy in predicting grid stability indicators for the next 15 minutes, including 93.2% for voltage deviation rate, 94.1% for frequency deviation rate, 89.8% for power factor deviation rate, and 90.4% for total harmonic distortion rate. The prediction of charging efficiency parameters also performed well, with accuracy of 92.1% for charging power utilization rate, 88.9% for energy loss rate, and 91.3% for charging time efficiency.

[0133] The system demonstrates excellent dynamic response in adaptive harmonic filter adjustment. When detecting an increase in the second harmonic content, the filter automatically adjusts the inductance of the second harmonic filter branch from 15mH to 18mH and the capacitance from 220μF to 180μF, effectively suppressing the propagation of the second harmonic. The fifth harmonic filter branch has an adjustable parameter range of 8mH to 12mH inductance and 150μF to 200μF capacitance, meeting the harmonic suppression requirements under different operating conditions.

[0134] The system also demonstrated excellent performance in handling large load surges. When eight charging stations were simultaneously activated, the instantaneous power surge reached 420kW. The cluster charging balancing optimization module implemented a time-sharing startup strategy, staggering the startup times of the eight charging stations, with each charging station starting up every 200ms, effectively reducing the impact on the power grid. During the startup process, the maximum grid voltage drop was kept within 2.8%, well below the safety limit of 5%.

[0135] The technical team also tested the system's temperature adaptability. At an ambient temperature of 35°C, the internal temperature of the main control cabinet remained stable at 42°C, and the operating temperature of each power module remained below 55°C. When the ambient temperature dropped to -10°C, the system initiated a preheating process to ensure that all electronic components operated within the optimal operating temperature range.

[0136] After one month of continuous operation, the system's reliability indicators met design requirements. The system achieved an average trouble-free operation time of 720 hours, an average single-fault repair time of 1.2 hours, and a system availability rate of 99.8%. The average charging efficiency of the charging pile terminals was 94.2%, and the power conversion loss rate was kept below 5.8%.

[0137] This invention represents a significant technological advancement compared to traditional charging systems. Traditional charging systems typically utilize an independent control mode, lacking coordination between charging stations, which can easily cause grid shocks and uneven power distribution. This invention achieves coordinated control of multiple charging stations through a group charging balance optimization module, dynamically adjusting the power distribution of each station based on grid conditions, effectively reducing grid shocks. Traditional systems have limited harmonic suppression capabilities and often employ passive filters with fixed parameters, which are unable to adapt to dynamic changes in harmonic components. This invention utilizes an adaptive harmonic filter that can adjust filter parameters in real time based on harmonic detection results, significantly improving harmonic suppression. In terms of predictive control, traditional systems rely primarily on historical experience and simple feedback control, lacking the ability to predict future grid conditions. This invention utilizes a grid stability prediction model that can predict grid state changes in advance, enabling proactive control and improving system stability and reliability. In terms of fault handling, traditional systems have a slow response speed, often requiring manual intervention to complete power redistribution. The emergency power redistribution mechanism of this invention can complete fault detection and power redistribution in seconds, significantly improving the system's emergency response capabilities.

[0138] It should be noted that the present invention also addresses the following three key technical issues: The first is the passive response delay caused by the lack of grid state prediction capabilities in multi-pile charging systems. Traditional multi-pile charging systems typically adopt a passive response mode based on real-time monitoring and immediate adjustment. When the grid state changes, the system requires a certain amount of time to detect, process, and respond. This delay can lead to short-term power distribution imbalances and system shocks. The present invention introduces a grid stability prediction model based on the Transformer architecture. This model uses an encoder-decoder structure and an attention mechanism to learn the temporal characteristics and complex laws of grid operation. By analyzing historical data and real-time status, it predicts grid stability indicators and charging efficiency parameters for future time periods. This enables the system to identify grid state trends in advance and pre-adjust power allocation strategies, achieving a transition from passive response to active prediction, effectively reducing system response delays and power allocation fluctuations. The second technical issue is the impact of harmonic pollution and power factor deterioration on the operating efficiency of charging systems. Harmonic pollution in the grid can cause charging equipment to have a lower power factor and increase energy loss. Traditional systems typically use fixed-parameter harmonic filters for passive filtering, which cannot dynamically adjust filtering parameters based on real-time harmonic content, resulting in poor filtering effectiveness. The present invention integrates a harmonic detection circuit in the harmonic filter, which can monitor the harmonic content in the grid voltage and current waveforms in real time. The group charging balance optimization module dynamically adjusts the filtering parameters of the harmonic filter according to the predicted grid stability index. Through the four-segment partition adjustment strategy of the gated weight function, when the harmonic content is at different levels, the system automatically selects the corresponding filter parameter combination, which realizes the active suppression of harmonic pollution and dynamic optimization of the power factor, and significantly improves the operating efficiency and power quality of the charging system. The third technical problem is the lack of effective fault isolation and power redistribution mechanism for the multi-pile charging system under abnormal grid conditions. When a single or multiple charging piles fail or the grid fluctuates severely, the traditional system often adopts an overall shutdown protection strategy, which affects the continued operation of the normal charging piles and reduces the system availability and service continuity. The present invention establishes an emergency power redistribution mechanism. By real-time monitoring of the charging status of each charging pile terminal and changes in grid parameters, when a grid anomaly or charging pile failure is detected, the group charging balance optimization module can quickly identify the fault source and isolate it from the power allocation matrix, while recalculating the optimal power allocation plan for the remaining normal charging piles. By reducing the power output of some charging piles, the overall stable operation of the system is guaranteed, ensuring that the system can still maintain basic charging service functions under abnormal circumstances.

[0139] It should be noted that the variables involved in the present invention are explained in detail as shown in Table 2 below.

[0140] Table 2 Variable explanation table

[0141]

[0142] The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed by the present invention, which should be covered by the scope of protection of the present invention.

Claims

1. A split-type DC charging one-to-many group charging control system, including a main control cabinet, a power distribution unit, a charging pile terminal, a current sensor module, a voltage monitoring unit, a harmonic filter, a communication interface module, a power metering device, a frequency detection module and a control chip, characterized in that: The control chip includes a group charging balancing optimization module, which uses a voltage monitoring unit to collect real-time three-phase grid voltage data, calculate voltage imbalance and voltage fluctuation amplitude, and simultaneously collects real-time current data from each charging pile terminal through a current sensor module. The collected three-phase grid voltage and real-time current data are preprocessed to establish a real-time grid status database. The preprocessed real-time grid status database is segmented and analyzed using a sliding time window method to extract grid characteristic parameters within each time window. Based on the extracted grid characteristic parameters, a multi-objective optimization equation system is used to calculate the optimal power allocation matrix for each charging pile terminal to achieve load balancing and minimize grid impact. The calculated optimal power allocation matrix is ​​input into a pre-trained grid stability prediction model to predict grid stability indicators and charging efficiency parameters for future time periods. The filtering parameters of the harmonic filter are adjusted based on the predicted grid stability indicators. The charging power output of each charging pile terminal is optimized by dynamically adjusting the gating weight function. The charging status of each charging pile terminal and changes in grid parameters are monitored in real time. When a grid anomaly or charging pile failure is detected, the emergency power redistribution mechanism is immediately activated to ensure the overall stable operation of the system.

2. The split-type DC charging one-to-many group charging control system according to claim 1 is characterized in that: The sliding time window method uses a data collection cycle of 30 seconds as the basic unit. Each time window contains grid status data of 900 sampling points, which is used to analyze the short-term change trend of grid parameters.

3. The split-type DC charging one-to-many group charging control system according to claim 2 is characterized in that: The grid characteristic parameters are specifically quantitative indicators of the grid operation status extracted from the real-time grid status database, including the values ​​of four dimensions: voltage RMS, current RMS, power factor and harmonic content.

4. The split-type DC charging one-to-many group charging control system according to claim 3 is characterized in that: The optimal power allocation matrix is ​​a two-dimensional data structure that describes the power allocation ratio of each charging pile terminal. The number of matrix rows corresponds to the number of charging pile terminals, and the number of columns corresponds to the power allocation time nodes.

5. The split-type DC charging one-to-many group charging control system according to claim 4 is characterized in that: The grid stability index is a comprehensive quantitative parameter for evaluating the stability of grid operation, including voltage deviation rate, frequency deviation rate, power factor deviation rate and total harmonic distortion rate.

6. The split-type DC charging one-to-many group charging control system according to claim 5 is characterized in that: Charging efficiency parameters are quantitative indicators that describe the energy conversion efficiency of the charging pile terminal, including charging power utilization rate, energy loss rate and charging time efficiency.

7. The split-type DC charging one-to-many group charging control system according to claim 6 is characterized in that: The gated weight function is a mathematical function that dynamically adjusts the weights of the neural network based on grid parameters. The input parameters include the grid voltage fluctuation amplitude, charging pile load rate and harmonic content.

8. The split-type DC charging one-to-many group charging control system according to claim 7 is characterized in that: The emergency power redistribution mechanism is a power redistribution strategy that is activated when the system detects an abnormality. It ensures the overall stable operation of the system by reducing the power output of some charging piles.

9. The split-type DC charging one-to-many group charging control system according to claim 8 is characterized in that: The multi-objective optimization equation group includes the power balance equation and the grid stability equation. The power balance equation is used to calculate the optimal power allocation ratio of each charging pile terminal, and the grid stability equation is used to evaluate the impact of the power allocation plan on the grid stability.

10. The split-type DC charging one-to-many group charging control system according to claim 9 is characterized in that: The inputs to the power balance equation include the current load rate of the power grid, the charging demand of each charging station, the grid voltage fluctuation amplitude, and the total power capacity of the system. The output is the power distribution coefficient of each charging station. The inputs to the power grid stability equation include the grid voltage imbalance, total harmonic distortion rate, power factor deviation value, and load change rate. The output is the power grid stability evaluation index.

Images