Power distribution management system of intelligent charging pile

Abstract

The invention discloses a power distribution management system of an intelligent charging pile, and belongs to the technical field of electric vehicle charging facilities and intelligent power grids. The load prediction module carries out space-time alignment fusion on historical data and real-time monitoring data to generate a power distribution demand prediction value, and the edge calculation controller executes a model prediction control algorithm based on multi-source data to generate a real-time control strategy containing relay time sequence parameters and a capacitance compensation scheme. And the dynamic power distribution adjustment module executes strategy parameters through the solid-state relay array and the parallel compensation capacitor bank. Self-adaptive adjustment of the power distribution network is achieved through a real-time monitoring-prediction-optimization closed-loop control mechanism, the harmonic content of the power grid is effectively reduced, the energy distribution efficiency of the charging pile group is improved, and the method is suitable for intelligent electric energy management of public charging stations and other scenes.

Description

Technical Field

[0001] The present invention belongs to the technical field of electric vehicle charging facilities and smart grids, and particularly relates to a power distribution management system for smart charging piles. Background Art

[0002] In the power distribution system of electric vehicle charging facilities, the random start and stop and power fluctuations of multiple charging piles can easily cause three-phase imbalance problems in the power supply network. Traditional power distribution systems usually adopt a three-phase load balancing strategy with a fixed threshold, which cannot adapt to the dynamically changing working conditions of the charging pile group. When the charging load suddenly changes in a short period of time, the phase current difference may exceed the rated capacity of the transformer, resulting in the risk of neutral line overload. The existing method alleviates this problem by regularly rotating the phase sequence distribution of the charging piles, but it is difficult to complete dynamic adjustment within a time scale of seconds due to communication delays and control cycles. In particular, when multiple charging piles switch to high-power mode at the same time, the rate of change of the phase current difference can reach more than 10% per minute, while the mechanical action time of traditional relays is usually more than 100ms, resulting in the superposition effect of regulation lag and transient inrush current. In addition, the fixed threshold control strategy does not take into account the time drift of the impedance parameters of the power supply node, further reducing the robustness of the dynamic balance. Therefore, the present invention was produced. Summary of the Invention

[0003] An object of the present invention is to solve at least the above problems and / or disadvantages and to provide at least the advantages which will be described hereinafter.

[0004] Another object of the present invention is to provide a power distribution management system for an intelligent charging pile.

[0005] Solve the problems of three-phase imbalance, harmonic distortion and delayed response of the power distribution system caused by the dynamic load of the charging pile group; Solve the problem of insufficient prediction accuracy caused by the spatiotemporal mismatch between historical data and real-time monitoring data; Solve the problems of inrush current superposition and uneven thermal stress during relay switching; Solve the contradiction between delay uncertainty and algorithm complexity in centralized control architecture; Solve the forecast fusion problem where nonlinear characteristics and periodic fluctuations are difficult to model together; Solve the problem of constructing input vectors caused by inconsistent timestamps and discrete spatial distribution of multi-source data; Solve the problems of dynamic constraint coupling and real-time solution efficiency in multi-objective optimization; Solve the mismatch problem between dynamic changes in harmonic spectrum and fixed compensation scheme; Solve the control jitter problem caused by insufficient synchronization accuracy of cross-node monitoring data; Resolve the risk of control command loss caused by insufficient communication link reliability.

[0006] To this end, the technical solution provided by the present invention is: A power distribution management system for a smart charging pile, comprising: The current and voltage monitoring module is configured at each charging pile power supply node to collect three-phase current and voltage waveform data in real time to obtain real-time current and voltage data; The load prediction module is connected to the current and voltage monitoring module to align and integrate the historical usage data of charging piles with the real-time current and voltage data in time and space to generate a distribution demand forecast value; a dynamic power distribution adjustment module, which is in communication with the load prediction module and receives the power distribution demand prediction value output by the load prediction module; An edge computing controller is communicatively connected to the current and voltage monitoring module, the load prediction module, and the dynamic power distribution adjustment module, wherein the edge computing controller receives real-time current and voltage data from the current and voltage monitoring module, receives power distribution demand forecast values ​​from the load prediction module, and executes a model predictive control algorithm based on multi-source data to generate power distribution real-time control strategy parameters including relay timing parameters and capacitor compensation schemes, and sends the control strategy parameters to the dynamic power distribution adjustment module. The dynamic power distribution adjustment module receives the power distribution real-time control strategy parameters from the edge computing controller and executes the power distribution real-time control strategy parameters through the solid-state relay array and the parallel compensation capacitor group.

[0007] Preferably, in the power distribution management system of the smart charging pile, the load prediction module includes: The data fusion unit is used to align and fuse the discrete historical usage data of the charging pile with the real-time current and voltage data in time and space to construct a multi-dimensional time series input vector; A hybrid prediction model, which is composed of an LSTM neural network and an ARIMA algorithm in parallel through a weighted fusion unit, receives the multidimensional time series input vector, wherein the LSTM network processes nonlinear charging behavior characteristics and the ARIMA algorithm captures periodic load fluctuation patterns. The result output of the hybrid prediction model is synchronized with the policy update cycle of the edge computing controller; The prediction window dynamic adjustment unit dynamically selects a 15-30 minute prediction window based on real-time load fluctuation data. When the fluctuation rate exceeds the three-phase imbalance change rate threshold Δ>3% / second, it sends a mode switching signal to the edge computing controller and automatically switches to the 15-minute high-frequency prediction mode.

[0008] Preferably, in the power distribution management system of the smart charging pile, the dynamic power distribution adjustment module includes: The dynamic balancing control unit uses an intelligent timing optimization strategy based on a genetic algorithm to generate a Pareto-optimal on-off timing combination based on the control strategy parameters issued by the edge computing controller, ensuring that the load differences between phases dynamically meet the rated value interval constraints. The optimized timing parameters are then fed back to the hierarchical prediction model architecture of the edge computing controller. The harmonic compensation decision unit integrates an FFT harmonic analysis core and a fuzzy logic controller. When the 5%-10% harmonic distortion threshold is detected, it generates a hierarchical switching instruction linked to the capacitor compensation scheme and transmits the compensation effect data back to the data fusion unit of the load prediction module. A phase-to-phase coupling suppression circuit that injects reverse offset current during the switching process of the solid-state relay array and inputs inrush current suppression data into the optimization objective function of the edge computing controller; The thermal stress balancing module monitors the junction temperature distribution of solid-state relays in real time and dynamically adjusts the on-off timing duty cycle to control the temperature rise difference of each relay in the array within the range of ±15°C. It also feeds back the thermal status data to the edge computing controller as a real-time constraint condition for the model predictive control algorithm.

[0009] Preferably, in the power distribution management system of the smart charging pile, the edge computing controller includes: A hierarchical prediction model architecture decomposes the model predictive control algorithm into a second-level (5-30 seconds) rolling optimization layer and a minute-level (1-15 minutes) strategic planning layer. The time window of the rolling optimization layer dynamically matches the prediction output period of the load prediction module and receives the thermal stress data output by the thermal stress equalization module in real time as a constraint condition. The strategic planning layer integrates the historical calibration data of the confidence feedback mechanism to generate long-term optimization constraints, whose maximum time window does not exceed the 30-minute upper limit set by the prediction window dynamic adjustment unit; A dynamic event triggering mechanism unit constructs an adaptive threshold function based on real-time load fluctuation data. When the three-phase imbalance change rate Δ>3% / second or the harmonic distortion gradient ΔTHD>1.5% / second, it performs an emergency update of the control strategy parameters and synchronously adjusts the LSTM-ARIMA fusion weight distribution ratio of the hybrid prediction model and the operating mode of the prediction window dynamic adjustment unit.

[0010] Preferably, in the power distribution management system of the smart charging pile, the hybrid prediction model includes: A dynamic weight allocation unit, whose input end is in communication with the dynamic event trigger mechanism unit of the edge computing controller, receives trigger signals of the three-phase imbalance change rate Δ and the harmonic distortion gradient ΔTHD in real time, and whose output end is connected to the prediction result fusion interface of the LSTM neural network and the ARIMA algorithm; Two-channel feature extractor, including: The first feature channel embeds a bidirectional LSTM neural network with a temporal attention mechanism to extract temporal features of non-steady-state current spikes and sudden changes in user charging behavior in the charging pile historical data; The second characteristic channel is configured with an ARIMA algorithm module with adaptive difference order adjustment to track the periodic fluctuation of the power grid load through a sliding time window; The weighted fusion unit includes: The weight calculation subunit dynamically adjusts the fusion weight ratio of the LSTM neural network and the ARIMA algorithm based on the real-time values ​​of the three-phase imbalance change rate Δ and the harmonic distortion gradient ΔTHD output by the dynamic event trigger mechanism; The tensor fusion subunit superimposes the nonlinear feature vector output by LSTM and the periodic component generated by ARIMA according to real-time weights to generate a fused prediction value; The clock synchronization interface precisely aligns the time resolution of the fusion prediction value output by the weighted fusion unit to the policy update period of the edge computing controller.

[0011] Preferably, in the power distribution management system of the smart charging pile, the data fusion unit includes: Spatiotemporal alignment engine, including: The multi-source timestamp calibration module dynamically matches the discrete timestamps of the charging pile's historical usage data with the continuous sampling timestamps of the real-time current and voltage data, and uses a sliding window interpolation algorithm to fill the time gaps in the historical data. The node topology mapping module maps the non-uniformly distributed discrete historical data to the spatial coordinate system of the real-time monitoring node according to the physical location relationship of the charging pile power supply nodes; Data integrity enhancers, including: The missing data compensation subunit performs Gaussian process regression based on the correlation coefficient of the current of adjacent nodes when it detects that the historical data is missing for more than 10% of the sampling period; The abnormal data cleaning subunit removes outlier data points exceeding ±20% of the rated current through a dynamic threshold comparison method; The multidimensional vector generator constructs input vectors from the time-space aligned data according to the following dimensions: the first dimension is the amplitude ratio of the fundamental component to the harmonic components of the real-time three-phase current; the second dimension is the second-order difference characteristics of the historical load curve; the third dimension is the time-varying parameters of the power supply node impedance matrix; and the fourth dimension is the relay junction temperature gradient fed back by the thermal stress balancing module. The timing constraint verification interface uses a hardware timer to check the clock synchronization deviation of each dimension of the input vector to ensure that the closed-loop control timing jitter is less than 5ms.

[0012] Preferably, in the power distribution management system of the smart charging pile, the execution of the model predictive control algorithm includes the following steps: a) Multi-objective optimization function construction: The first optimization goal is to minimize the transient inrush current peak value and junction temperature gradient variance of the solid-state relay array; The second optimization goal is to maximize the tracking accuracy of the power distribution demand forecast value output by the load forecast module; The third optimization goal is to suppress the 5%-10% harmonic distortion rate detected by the harmonic compensation decision unit; b) Layered scrolling optimization mechanism: Second-level optimization layer: Using a 5-30 second window, it integrates multi-dimensional time series input vectors and thermal stress data to solve the Pareto frontier solution set of the solid-state relay on-off timing in real time; Minute-level planning layer: Using a dynamically adjusted 15-30 minute forecast window as the constraint boundary, the robustness of the capacitor compensation scheme is optimized and anti-disturbance threshold parameters are generated; c) Dynamic constraint coupling module: The inrush current suppression data of the phase-to-phase coupling suppression circuit and the temperature rise difference data of the thermal stress equalization module are used as real-time inequality constraints for the second-level optimization layer. Dynamically map the fusion weight ratio of the hybrid prediction model to the objective function attenuation factor of the minute-level planning layer; d) Self-healing transmission of policy parameters: When the activation of the dynamic event trigger mechanism is detected, the iterative process of the minute-level planning layer is skipped, and the rolling optimization results of the previous cycle are directly superimposed with the emergency compensation amount, and redundant instructions are issued through the dual channels of the multimodal communication unit.

[0013] Preferably, in the power distribution management system of the smart charging pile, the harmonic compensation decision unit further includes: The adaptive harmonic spectrum analysis module is configured to: use a sliding window fast Fourier transform (FFT) to decompose the current waveform in real time and generate an amplitude distribution spectrum of harmonic components from 0 to 50; Construct a set of harmonic weight factors, where the weight value of each harmonic is inversely proportional to the corresponding harmonic impedance value in the power supply node impedance matrix. When the distortion rate of a specific harmonic order exceeds a threshold, a capacitor compensation capacity value adaptively matching the grid impedance is generated based on the frequency characteristics of the harmonic order and the line impedance parameters. Dynamic capacitor matching array, including: programmable switching capacitor group clusters, each group of capacitance value is configured in a geometric step-by-step manner, and the compensation accuracy control required by the anti-disturbance threshold parameter is achieved through a binary combination strategy; The capacitor bank switching logic controller selects the optimal capacitor combination scheme according to the compensation capacity value output by the adaptive harmonic spectrum analysis module.

[0014] Preferably, in the power distribution management system of the smart charging pile, the current voltage monitoring module includes: A multimodal sensing unit, which is configured at each charging pile power supply node, includes: a current transformer array, a broadband voltage sensor group, and a temperature-current coupling compensator; An adaptive synchronous acquisition engine configured to: achieve clock synchronization across power supply nodes; dynamically adjust the sampling frequency based on the edge computing controller's policy update cycle; The online harmonic extraction unit is embedded with a sliding window fast Fourier transform (FFT) core to separate the fundamental wave and the 2nd to 50th harmonic components in the current and voltage waveforms in real time, and generate a harmonic spectrum matrix with a time stamp.

[0015] Preferably, in the power distribution management system of the smart charging pile, the edge computing controller further includes: The multimodal communication unit integrates the enhanced HPLC protocol for power line carrier communication and the LoRa wireless mesh network dual channel. It automatically switches the wireless link when the carrier communication signal-to-noise ratio is less than 15dB, ensuring data synchronization latency less than 100ms and compatibility with the requirements of the bidirectional communication link between modules. The bidirectional communication link includes: a command channel for issuing control strategy parameters to the dynamic power distribution adjustment module; and a monitoring channel for receiving the current and voltage data stream collected in real time by the current and voltage monitoring module. The redundant communication guarantee module is configured to activate the Bluetooth Mesh emergency communication link when it is detected that the dual-channel communication quality of the enhanced HPLC protocol and the LoRa wireless network decreases at the same time. The emergency link adopts a time-sliced ​​retransmission mechanism.

[0016] The present invention has at least the following beneficial effects: The present invention improves the response speed of the power distribution system to dynamic loads by constructing a real-time monitoring-prediction-adjustment closed-loop control architecture, while reducing the impact of three-phase imbalance and harmonic distortion on the power grid.

[0017] The present invention enhances the spatiotemporal correlation of load prediction through the synergistic effect of spatiotemporal alignment fusion and hybrid prediction model, and effectively copes with the complex working conditions of charging pile groups with nonlinear and periodic superposition.

[0018] The present invention reduces transient inrush current and equipment loss during relay switching and improves the reliability and life of solid-state relay arrays through the combined control of genetic algorithm optimization and thermal stress balancing.

[0019] The present invention uses a hierarchical prediction architecture and a dynamic event triggering mechanism to reduce computational complexity while ensuring optimization accuracy, thereby achieving real-time distributed control of large-scale charging pile groups.

[0020] The present invention adaptively matches the nonlinear and periodic characteristics of load changes through dynamic weight allocation and dual-channel feature extraction, thereby improving the generalization ability of the prediction model under sudden change conditions.

[0021] The present invention solves the problem of integrating multi-source heterogeneous data through the cooperation of the spatiotemporal alignment engine and the multidimensional vector generator, and provides input data with high integrity and consistency for the control algorithm.

[0022] The present invention balances the real-time and robustness of multi-objective optimization through a hierarchical optimization and dynamic constraint coupling mechanism, ensuring the synergy of control strategies at different time scales.

[0023] The present invention realizes real-time adaptation of harmonic compensation capacity and grid impedance through a combination strategy of adaptive harmonic spectrum analysis and dynamic capacitor matching, thus avoiding the resonance risk of traditional fixed compensation schemes.

[0024] The present invention improves the synchronization accuracy of cross-node monitoring data through multi-modal sensing and adaptive synchronous acquisition technology, providing high-reliability basic data guarantee for closed-loop control.

[0025] The present invention enhances the reliability of control instruction transmission through multimodal communication and redundancy protection mechanism, ensuring the continuous and stable operation of the system under extreme working conditions.

[0026] Other advantages, objectives and features of the present invention will be reflected in part from the following description and will be understood by those skilled in the art through study and practice of the present invention. DETAILED DESCRIPTION

[0027] The present invention is further described in detail below with reference to the embodiments so that those skilled in the art can implement the invention with reference to the description.

[0028] It should be understood that terms such as “having”, “including” and “comprising” used herein do not preclude the existence or addition of one or more other elements or combinations thereof.

[0029] It should be noted that the experimental methods described in the following embodiments are conventional methods unless otherwise specified, and the reagents and materials can be obtained from commercial channels unless otherwise specified.

[0030] The present invention provides a power distribution management system for an intelligent charging pile, comprising the following steps: The current and voltage monitoring module is configured at each charging pile power supply node to collect three-phase current and voltage waveform data in real time to obtain real-time current and voltage data; The load prediction module is connected to the current and voltage monitoring module to align and integrate the historical usage data of charging piles with the real-time current and voltage data in time and space to generate a distribution demand forecast value; a dynamic power distribution adjustment module, which is in communication with the load prediction module and receives the power distribution demand prediction value output by the load prediction module; An edge computing controller is communicatively connected to the current and voltage monitoring module, the load prediction module, and the dynamic power distribution adjustment module, wherein the edge computing controller receives real-time current and voltage data from the current and voltage monitoring module, receives power distribution demand forecast values ​​from the load prediction module, and executes a model predictive control algorithm based on multi-source data to generate power distribution real-time control strategy parameters including relay timing parameters and capacitor compensation schemes, and sends the control strategy parameters to the dynamic power distribution adjustment module. The dynamic power distribution adjustment module receives the power distribution real-time control strategy parameters from the edge computing controller and executes the power distribution real-time control strategy parameters through the solid-state relay array and the parallel compensation capacitor group.

[0031] An open-type current transformer with a range of 0-1000A and a wide-band voltage sensor can be selected and installed on the incoming terminal block of the charging pile power supply node. The sampling frequency is set to 4kHz, and cross-node synchronous acquisition is achieved through a hardware timer, with the time deviation controlled within ±50μs. The core material of the current transformer can be a laminated silicon steel sheet structure, and the shell is made of flame-retardant polycarbonate material. The voltage divider resistor of the voltage sensor uses a metal film resistor with a temperature coefficient of ±50ppm / ℃ and is installed on an insulating bracket on the inner wall of the distribution cabinet. During operation, the three-phase current signal is converted into a voltage signal by the Hall element, and is input into a high-precision analog-to-digital conversion chip together with the voltage sampling value to generate a waveform data packet with a timestamp. The data fusion unit can be configured with a microcontroller and run a sliding window interpolation algorithm to align the 15-minute sampling points of historical data with the 1-second sampling points of real-time data, with a maximum tolerance of ±2 seconds for filling the time gap. The hybrid prediction model's LSTM neural network can use a two-layer structure with 128 hidden units, and the ARIMA algorithm's differencing order can be dynamically adjusted from 1st to 3rd order. The prediction window dynamic adjustment unit sets the three-phase imbalance change rate threshold to Δ>3% / second. When the fluctuation rate exceeds the threshold, a mode switching signal is sent to the edge computing controller via a general-purpose input / output pin. When constructing the input vector, the power supply node impedance parameters are measured using an online frequency sweep method with a measurement frequency range of 50 Hz to 2.5 kHz. The impedance matrix update period is set to 10 minutes. In the multi-objective optimization function of the model predictive control algorithm, the weight coefficient for the variance of the solid-state relay junction temperature gradient is set to 0.6, and the weight coefficient for harmonic distortion suppression is set to 0.3. A solid-state relay module with a rated current of 200A can be selected and installed on the forced-air cooling baseplate of the power distribution cabinet. The on-off timing is controlled by the driver chip. The parallel compensation capacitor bank uses metallized polypropylene film capacitors (with a withstand voltage of 450V) and implements staged switching via power semiconductor modules. The switching command transmission delay is ≤5ms. The thermal stress equalization module collects junction temperature data using a temperature sensor mounted on the relay heat sink. The temperature rise differential control algorithm uses PID regulation with a proportional coefficient set to 0.8. During operation, if the harmonic distortion rate exceeds 7%, the capacitor bank switches to a matching compensation unit within 20ms, and the relay array simultaneously switches the load phase according to the optimized timing. This implementation utilizes high-precision sensing and real-time optimized control to enhance the distribution system's adaptability to dynamic load changes and mitigate the impact of transient events on the grid. A time-space aligned data fusion mechanism ensures the temporal and spatial consistency of the predictive model's input data, while a layered optimization architecture balances control accuracy and response speed. A modular hardware design improves system maintainability, while thermal stress balancing extends the life of key components.

[0032] In one of the solutions of the present invention, preferably, the load prediction module includes: The data fusion unit is used to align and fuse the discrete historical usage data of the charging pile with the real-time current and voltage data in time and space to construct a multi-dimensional time series input vector; A hybrid prediction model, which is composed of an LSTM neural network and an ARIMA algorithm in parallel through a weighted fusion unit, receives the multidimensional time series input vector, wherein the LSTM network processes nonlinear charging behavior characteristics and the ARIMA algorithm captures periodic load fluctuation patterns. The result output of the hybrid prediction model is synchronized with the policy update cycle of the edge computing controller; The prediction window dynamic adjustment unit dynamically selects a 15-30 minute prediction window based on real-time load fluctuation data. When the fluctuation rate exceeds the three-phase imbalance change rate threshold Δ>3% / second, it sends a mode switching signal to the edge computing controller and automatically switches to the 15-minute high-frequency prediction mode.

[0033] A microcontroller can be used to run a sliding window interpolation algorithm to align the 15-minute sampling points of the charging pile's historical data with the 1-second sampling points of the real-time monitoring module. The timestamp calibration module's maximum interpolation tolerance is set to ±2 seconds, and the node topology mapping uses the Euclidean distance algorithm with a spatial positioning accuracy of ±0.5 meters. The missing data compensation subunit can perform Gaussian process regression based on the correlation coefficient of adjacent node currents (threshold set to ≥0.85). The dynamic threshold for abnormal data cleaning is set to ±20% of the rated current. The multidimensional vector generator is installed in the embedded industrial computer in the power distribution cabinet. The real-time three-phase current harmonic components are collected using a high-precision analog-to-digital conversion chip. The fundamental wave and harmonic amplitude ratio are calculated at a frequency of 50Hz±1Hz. The bidirectional LSTM neural network can be configured with two hidden layers, each with 128 units. The temporal attention mechanism has a time window of 60 minutes. The ARIMA algorithm's differencing order is adaptively adjusted from 1 to 3, and the sliding time window is set to 24 hours. The dynamic weight allocation unit adjusts the LSTM and ARIMA weight ratios in real time by comparing the three-phase imbalance change rate Δ (threshold 3% / second) with the harmonic distortion gradient ΔTHD (threshold 1.5% / second). The weight calculation subunit uses a sigmoid function mapping, outputting an LSTM weight coefficient of 0.7-0.8 when Δ>3% / second. The tensor fusion subunit performs matrix superposition of the LSTM output feature vector and the ARIMA periodic component within the programmable logic chip. The timing deviation of the clock synchronization interface is controlled within ±5ms. The prediction window switching threshold is set at a three-phase imbalance change rate Δ>3% / second. The mode switching signal is transmitted to the edge computing controller via an optocoupler isolation circuit. The 15-minute window in high-frequency prediction mode uses a data update mechanism with a sliding step of 1 minute, and the historical data buffer depth is set to 48 hours. A high-precision clock module can be used to ensure time base synchronization and is installed on the back of the load prediction module circuit board. During operation, if real-time monitoring indicates that Δ exceeds the threshold five times in a row, the window switching is triggered via a general-purpose input / output pin, and the initial differencing order of the ARIMA algorithm is reset to 1. The second dimension of the input vector, the second-order difference characteristics of the historical load, are calculated using the central difference method, with a differencing interval set to 10 minutes. This implementation improves the load forecasting model's adaptability to sudden operating conditions by combining dynamic weight allocation with multi-dimensional feature fusion. A spatiotemporal alignment mechanism effectively eliminates temporal deviations in multi-source data, while a hybrid forecasting architecture accounts for both nonlinear characteristics and periodicity. A dynamic switching strategy for forecasting windows enhances the system's ability to track load fluctuations, while optocoupler isolation ensures interference resistance in mode switching signals.

[0034] In one of the solutions of the present invention, preferably, the dynamic power distribution adjustment module includes: The dynamic balancing control unit uses an intelligent timing optimization strategy based on a genetic algorithm to generate a Pareto-optimal on-off timing combination based on the control strategy parameters issued by the edge computing controller, ensuring that the load differences between phases dynamically meet the rated value interval constraints. The optimized timing parameters are then fed back to the hierarchical prediction model architecture of the edge computing controller. The harmonic compensation decision unit integrates an FFT harmonic analysis core and a fuzzy logic controller. When the 5%-10% harmonic distortion threshold is detected, it generates a hierarchical switching instruction linked to the capacitor compensation scheme and transmits the compensation effect data back to the data fusion unit of the load prediction module. A phase-to-phase coupling suppression circuit that injects reverse offset current during the switching process of the solid-state relay array and inputs inrush current suppression data into the optimization objective function of the edge computing controller; The thermal stress balancing module monitors the junction temperature distribution of solid-state relays in real time and dynamically adjusts the on-off timing duty cycle to control the temperature rise difference of each relay in the array within the range of ±15°C. The thermal state data is fed back to the edge computing controller as a real-time constraint condition of the model predictive control algorithm. A temperature sensor is used to monitor the relay junction temperature, and the PID parameters are set to Kp=0.8 / Ki=0.05 / Kd=0.1. The temperature rise difference is controlled within ±15℃. A microcontroller can be used to run the genetic algorithm, with a population size of 80 and 100 iterations. The Pareto frontier solution selection criteria include interphase current difference (threshold ±5%), switching frequency (≤120 times per hour), and energy consumption coefficient (≤0.3). The optimization results are transmitted via a communication interface to the solid-state relay driver board, which is installed in the control compartment on the left side of the power distribution cabinet. During operation, the three-phase current RMS values ​​are collected every 5 seconds. When the current difference between any two phases exceeds 8%, the genetic algorithm is triggered to recalculate the on-off timing combination. The drive signal for the relay array is transmitted via an optocoupler isolation circuit, with the rise time controlled to within 1μs. A digital signal processor can be used for FFT harmonic analysis, with an analysis window length of 10 power frequency cycles (200ms) and a frequency resolution of 0.5Hz. The fuzzy logic controller's input variables include the THD value (domain 5%-10%) and its rate of change (gradient threshold 1.5% / second), and the output variable is the capacitor compensation capacity (step size ±25kVar). Compensation instructions are transmitted to the driver module via an optical fiber interface, installed in the right unit of the capacitor compensation cabinet. When the 13th harmonic component is detected to exceed 5% of the fundamental, the filter branch with a series reactance of 7% is prioritized. The capacitor bank uses metallized polypropylene film capacitors with a withstand voltage rating of 450V and a capacitance tolerance of ±3%. The reverse compensation current generator can utilize an IGBT module with a drive voltage of 15V ±5%. The reverse current amplitude is dynamically adjusted based on the inrush current peak (adjustment steps of 10A). A smoothing reactor (2mH inductance) is connected in series with the compensation current injection circuit and installed at the busbar connection of the relay array. The temperature sensor of the thermal stress equalization module is mounted on the bottom of the relay ceramic substrate, with a temperature sampling interval of 1 second. The output signal of the PID regulator controls the relay duty cycle through PWM modulation. When the temperature difference between adjacent relays exceeds 10°C, the switching frequency of the high-temperature relay is automatically reduced. The heat dissipation substrate is made of aluminum alloy with a 0.5mm thick oxide insulation layer sprayed on the surface. This implementation improves the dynamic balancing accuracy of three-phase loads through the synergistic effect of genetic algorithm optimization and fuzzy decision-making. A reverse current injection mechanism effectively reduces the impact of transient processes on the power grid, and a PID temperature control strategy extends relay life. A modular harmonic compensation design enhances adaptability to different distortion modes, and a multi-parameter constraint mechanism ensures a safe margin for the optimization process.

[0035] In one of the solutions of the present invention, preferably, the edge computing controller includes: A hierarchical prediction model architecture decomposes the model predictive control algorithm into a second-level (5-30 seconds) rolling optimization layer and a minute-level (1-15 minutes) strategic planning layer. The time window of the rolling optimization layer dynamically matches the forecast output period of the load forecast module and receives the thermal stress data output by the thermal stress equalization module in real time as a constraint. The strategic planning layer integrates historical calibration data of the confidence feedback mechanism to generate long-term optimization constraints, with a maximum time window not exceeding the 30-minute upper limit set by the forecast window dynamic adjustment unit. A dynamic event triggering mechanism unit constructs an adaptive threshold function based on real-time load fluctuation data. When the three-phase imbalance change rate Δ>3% / second or the harmonic distortion gradient ΔTHD>1.5% / second, it performs an emergency update of the control strategy parameters and synchronously adjusts the LSTM-ARIMA fusion weight distribution ratio of the hybrid prediction model and the operating mode of the prediction window dynamic adjustment unit.

[0036] A microcontroller can be used to implement a second-level rolling optimization layer. The default time window is set to 15 seconds, which is automatically shortened to 5 seconds when the thermal stress data change rate exceeds 2°C / second. The strategic planning layer runs on an embedded industrial computer with a maximum time window limit of 30 minutes. Historical calibration data is stored in a flash memory module installed on the upper shelf of the control cabinet. The objective function weight coefficient of the rolling optimization layer is dynamically adjusted based on the real-time load rate. When the load rate exceeds 80%, the junction temperature gradient variance weight is increased to 0.7. During operation, thermal stress data is received via the bus every 5 seconds. If the constraints are exceeded for three consecutive cycles, the strategic planning layer is triggered to recalculate. The confidence threshold is set at 85%, and historical data is stored using a sliding window mechanism with a 30-day window length and a one-hour data update interval. Calibration data, including load factor distribution, relay switching statistics, and capacitor compensation success rate for the past 30 days, is stored in a memory chip. The constraint generation algorithm for the strategic planning layer uses Monte Carlo simulation with 1000 iterations, and is installed in the expansion slot of the industrial computer. When the confidence level of historical data falls below 75%, it automatically switches to the rolling optimization layer's standalone operating mode and sends an alarm signal through the interface. The threshold for the three-phase imbalance change rate Δ is set to 3% / second, and the threshold for the harmonic distortion gradient ΔTHD is set to 1.5% / second. Emergency updates are activated after the trigger condition persists for 3 seconds. A differential transceiver can be used to transmit the mode switching signal, installed on the communication daughterboard of the edge computing controller. When adjusting the weight distribution ratio, the weight coefficient of the LSTM neural network is increased from 0.5 to 0.7, and the order of the ARIMA algorithm is reset to 2nd order. After the event is triggered, the prediction window dynamic adjustment unit immediately switches to a 15-minute high-frequency mode, while the capacitor compensation decision unit enters a fast-response state, reducing the switching command delay to less than 10ms. During operation, when the communication channel bit error rate exceeds 1E-6, the data verification and retransmission mechanism is automatically activated. This implementation balances the requirements of real-time control and long-term planning through the collaborative optimization of hierarchical time windows. A confidence feedback mechanism improves the effective utilization of historical data, and a dynamic event triggering strategy enhances the system's response to sudden disturbances. A threshold-driven mode switching mechanism ensures the adaptability of the control strategy, and a communication interface design improves data transmission reliability.

[0037] In one of the solutions of the present invention, preferably, the hybrid prediction model includes: A dynamic weight allocation unit, whose input end is in communication with the dynamic event trigger mechanism unit of the edge computing controller, receives trigger signals of the three-phase imbalance change rate Δ and the harmonic distortion gradient ΔTHD in real time, and whose output end is connected to the prediction result fusion interface of the LSTM neural network and the ARIMA algorithm; Two-channel feature extractor, including: The first feature channel embeds a bidirectional LSTM neural network with a temporal attention mechanism to extract temporal features of non-steady-state current spikes and sudden changes in user charging behavior in the charging pile historical data; The second characteristic channel is configured with an ARIMA algorithm module with adaptive difference order adjustment to track the periodic fluctuation of the power grid load through a sliding time window; The weighted fusion unit includes: The weight calculation subunit dynamically adjusts the fusion weight ratio of the LSTM neural network and the ARIMA algorithm according to the real-time values ​​of the three-phase imbalance change rate Δ and the harmonic distortion gradient ΔTHD output by the dynamic event trigger mechanism, where: When the three-phase imbalance change rate Δ>3% / second or the harmonic distortion gradient ΔTHD>1.5% / second is detected, the weight ratio of the LSTM neural network is increased to above 70%; under steady-state conditions, the weight ratio of the ARIMA algorithm is maintained in a periodic fluctuation range of 50%-70%; The tensor fusion subunit superimposes the nonlinear feature vector output by LSTM and the periodic component generated by ARIMA according to real-time weights to generate a fused prediction value; The clock synchronization interface precisely aligns the time resolution of the fusion prediction value output by the weighted fusion unit to the policy update period of the edge computing controller.

[0038] A microcontroller can be used to implement the weight allocation logic. When the three-phase imbalance change rate Δ exceeds 3% / second, a trigger signal is sent via a general-purpose input / output pin. The weight calculation subunit uses a sigmoid function mapping, with an input Δ value range of 0-5% / second corresponding to an output weight coefficient of 0.5-0.8. The threshold for the harmonic distortion gradient ΔTHD is set at 1.5% / second. When both Δ>3% and ΔTHD>1.5% are met, the LSTM weight is locked at 0.75. The trigger signal is transmitted via an optocoupler isolation circuit to a programmable logic chip installed in the slot of the load prediction module. During operation, Δ and ΔTHD data are collected once per second. If the threshold is exceeded three times consecutively, a forced weight adjustment is initiated. The bidirectional LSTM neural network can be configured with two hidden layers, each with 128 units. The temporal attention mechanism has a time window of 60 minutes, and the activation function uses Reluctant Unit (ReLU). The differencing order of the ARIMA algorithm is automatically selected through a verification method, the sliding window length is set to 24 hours, and the residual standard deviation threshold is set to 0.15. The LSTM input data preprocessing unit can use a high-precision analog-to-digital converter chip, installed in the analog signal interface area of ​​the control board. The ARIMA module runs in the processor core and obtains historical load data through a direct memory access channel. During operation, the detection threshold for non-steady-state current spikes is set to 150% of the rated current. When a sudden event is detected, the sampling frequency of the LSTM feature channel is temporarily increased to 2kHz. The tensor fusion subunit utilizes a programmable logic chip, with the matrix superposition operation clock frequency set to 100 MHz and the data bus width configured to 64 bits. The clock synchronization interface utilizes a high-precision clock module, with signal error controlled to within ±1 μs. It is installed in a designated location on the circuit board. Timestamp alignment of the fused prediction values ​​is achieved using a hardware timer. When a timing deviation exceeding 5 ms is detected, an empty data packet is automatically inserted to maintain queue continuity. The output interface of the weighted fusion unit is transmitted to the edge computing controller via differential signaling, with a signal transmission delay compensation value set to 2 ms. This implementation enhances the predictive model's adaptability to sudden operating conditions through the synergy of dynamic weight adjustment and dual-channel feature extraction. A threshold trigger mechanism ensures optimal feature weighting under critical operating conditions, while fusion computing improves real-time data processing. High-precision clock synchronization eliminates timing mismatches in multi-source data, and a modular hardware layout enhances system maintainability.

[0039] In one solution of the present invention, preferably, the data fusion unit includes: Spatiotemporal alignment engine, including: The multi-source timestamp calibration module dynamically matches the discrete timestamps of the charging pile's historical usage data with the continuous sampling timestamps of the real-time current and voltage data, and uses a sliding window interpolation algorithm to fill the time gaps in the historical data. The node topology mapping module maps the non-uniformly distributed discrete historical data to the spatial coordinate system of the real-time monitoring node according to the physical location relationship of the charging pile power supply nodes; Data integrity enhancers, including: The missing data compensation subunit performs Gaussian process regression based on the correlation coefficient of the current of adjacent nodes when it detects that the historical data is missing for more than 10% of the sampling period; The abnormal data cleaning subunit removes outlier data points exceeding ±20% of the rated current through a dynamic threshold comparison method; The multidimensional vector generator constructs input vectors from the time-space aligned data according to the following dimensions: the first dimension is the amplitude ratio of the fundamental component to the harmonic components of the real-time three-phase current; the second dimension is the second-order difference characteristics of the historical load curve; the third dimension is the time-varying parameters of the power supply node impedance matrix; and the fourth dimension is the relay junction temperature gradient fed back by the thermal stress balancing module. The timing constraint verification interface uses a hardware timer to check the clock synchronization deviation of each dimension of the input vector to ensure that the closed-loop control timing jitter is less than 5ms.

[0040] A microcontroller can be used to run a sliding window interpolation algorithm. The timestamp calibration module has a maximum tolerance of ±2 seconds and an interpolation step size of 30 seconds. Node topology mapping uses a Euclidean distance algorithm, with a spatial coordinate system resolution of 0.1 meters. The system is installed in a slot on an embedded industrial computer. When the discrete timestamps of historical data align with the 1-second sampling interval of real-time data, missing data points are filled using cubic spline interpolation. During operation, if a node position offset exceeding 1 meter is detected, topology recalculation is automatically triggered, and mapping parameters are updated via the communication interface. The missing data compensation subunit performs Gaussian process regression based on the correlation coefficient of adjacent node currents (threshold ≥ 0.85), using a radial basis function (length scale 0.5) as the kernel function. The dynamic threshold for abnormal data cleaning is set at ±20% of the rated current, and outliers exceeding the threshold are replaced by median filtering. The data completion algorithm runs on the processor core with a 5-second calculation cycle and is installed on the circuit board of the data fusion unit. During operation, when historical data is missing for more than 15%, a cross-node data collaborative compensation mechanism is activated, and the compensation results are transmitted to the multidimensional vector generator via an interface. The first dimension of the input vector, the fundamental-to-harmonic ratio, is acquired via an analog-to-digital conversion chip, with the harmonic analysis frequency range set to 50 Hz-2.5 kHz. The second dimension, the second-order differential features, are calculated using the central difference method, with a 10-minute differential interval and three decimal places of accuracy. The timing constraint verification interface can utilize a high-precision clock module, with a hardware timer check period set to 1 second. A data buffer reset is triggered when a clock deviation exceeding 5 ms is detected. The output interface of the multidimensional vector generator is connected to the load prediction module via differential signals, with the signal transmission delay compensation value set to 2 ms. This implementation improves the spatiotemporal consistency of multi-source data through the coordinated processing of spatiotemporal calibration and data compensation. A dynamic threshold cleaning mechanism effectively filters out abnormal interference data, and Gaussian regression completion enhances the integrity of input vectors. The structured design of multidimensional vectors accommodates the dynamic characteristics of power grid parameters, and hardware-level timing verification ensures the synchronization accuracy of the control system.

[0041] In one embodiment of the present invention, preferably, the execution of the model predictive control algorithm includes the following steps: a) Multi-objective optimization function construction: The first optimization goal is to minimize the transient inrush current peak value and junction temperature gradient variance of the solid-state relay array; The second optimization goal is to maximize the tracking accuracy of the power distribution demand forecast value output by the load forecast module; The third optimization goal is to suppress the 5%-10% harmonic distortion rate detected by the harmonic compensation decision unit; b) Layered scrolling optimization mechanism: Second-level optimization layer: Using a 5-30 second window, it integrates multi-dimensional time series input vectors and thermal stress data to solve the Pareto frontier solution set of the solid-state relay on-off timing in real time; Minute-level planning layer: Using a dynamically adjusted 15-30 minute forecast window as the constraint boundary, the robustness of the capacitor compensation scheme is optimized and anti-disturbance threshold parameters are generated; c) Dynamic constraint coupling module: The inrush current suppression data of the phase-to-phase coupling suppression circuit and the temperature rise difference data of the thermal stress equalization module are used as real-time inequality constraints for the second-level optimization layer. Dynamically map the fusion weight ratio of the hybrid prediction model to the objective function attenuation factor of the minute-level planning layer; d) Self-healing transmission of policy parameters: When the dynamic event trigger mechanism is detected to be activated, the iterative process of the minute-level planning layer is skipped, and the rolling optimization result of the previous cycle is directly used to superimpose the emergency compensation amount, and redundant instructions are issued through the dual channels of the multimodal communication unit. ### Specific implementation of claim 7 A microcontroller can be used to construct a multi-objective function, with the junction temperature gradient variance weighting coefficient set to 0.6 and the harmonic distortion suppression weighting to 0.3. The solid-state relay's actuation time threshold is set to 10ms, with the driver chip controlling the conduction timing. The capacitor compensation capacity step accuracy is set to ±5kVar, using metallized polypropylene film capacitors (with a withstand voltage of 450V) installed in designated locations within the capacitor compensation cabinet. During operation, thermal stress data is collected every 5 seconds. If the junction temperature gradient variance exceeds 0.4 three times consecutively, its weight is automatically increased to 0.7. The default time window for the second-level optimization layer is 15 seconds, shortened to 5 seconds when the load factor exceeds 80%. The anti-disturbance threshold parameters for the minute-level planning layer are generated through Monte Carlo simulations with 200 iterations, running on the processor core of an embedded industrial computer. The rolling optimization results are transmitted to the dynamic power distribution adjustment module via a bus, and a verification algorithm is used to verify the data packets. During operation, when the prediction window switches to the 15-minute high-frequency mode, the bounds of the robustness optimization are tightened to ±3%. Inrush current suppression data is sampled at a 10kHz frequency and collected by a high-precision analog-to-digital conversion chip. This data is then used as an inequality constraint for second-level optimization (with a threshold set at 120% of the rated current). Temperature rise differential data (with a threshold of ±15°C) from the thermal stress equalization module is collected by a temperature sensor mounted on the surface of the relay heat sink. The weight ratios in the hybrid prediction model are converted to attenuation factors for the minute-level objective function using a linear mapping relationship, with the mapping coefficient set to 0.8. During operation, when the LSTM weight exceeds 0.7, the attenuation factor is automatically reduced to 0.6 to enhance robustness. Emergency mode is triggered when the three-phase imbalance change rate exceeds 3% / second for 5 seconds. At this point, the minute-level planning iterations are skipped and the optimization results from the previous cycle are directly applied with a 10% compensation. Redundant command channels can utilize both bus and fiber interfaces and are installed on the control cabinet's communication backplane. When a communication bit error rate exceeding 1E-6 is detected, forward error correction is activated, with a maximum retransmission limit of three. During self-healing transmission, capacitor compensation commands are prioritized, reducing response latency to less than 15ms. This implementation balances conflicting optimization objectives through dynamic configuration of multi-objective weights. A layered time window design balances real-time response and long-term stability, while a constraint coupling mechanism enhances the relevance of system parameters. Redundant communication and self-healing strategies enhance system reliability under extreme operating conditions, while a modular optimization algorithm reduces computing resource usage.

[0042] In one of the solutions of the present invention, preferably, the harmonic compensation decision unit further includes: The adaptive harmonic spectrum analysis module is configured to: use a sliding window fast Fourier transform (FFT) to decompose the current waveform in real time and generate an amplitude distribution spectrum of harmonic components from 0 to 50; Construct a set of harmonic weight factors, where the weight value of each harmonic is inversely proportional to the corresponding harmonic impedance value in the power supply node impedance matrix. When the distortion rate of a specific harmonic order exceeds a threshold, a capacitor compensation capacity value adaptively matching the grid impedance is generated based on the frequency characteristics of the harmonic order and the line impedance parameters. Dynamic capacitor matching array, including: programmable switching capacitor group clusters, each group of capacitance value is configured in a geometric step-by-step manner, and the compensation accuracy control required by the anti-disturbance threshold parameter is achieved through a binary combination strategy; The capacitor bank switching logic controller selects the optimal capacitor combination scheme according to the compensation capacity value output by the adaptive harmonic spectrum analysis module.

[0043] A digital signal processor can be used to perform a sliding window FFT, with the analysis window length set to 10 power frequency cycles (200ms) and a frequency resolution of 0.5Hz. The harmonic distortion rate trigger threshold is set to 5%, and a compensation command is generated when the 13th harmonic component exceeds 5% of the fundamental. The input signal to the FFT core is acquired by an analog-to-digital converter chip installed on the front-end signal board of the harmonic compensation cabinet. During operation, the harmonic spectrum matrix is ​​updated every 100ms. When the total harmonic distortion rate exceeds 7%, the sampling frequency is automatically increased to 2kHz. Harmonic impedance measurement uses a swept frequency method with a frequency range of 50Hz-2.5kHz and an impedance measurement accuracy of ±5%. A weighting factor calculation unit runs within the processor core. When a harmonic impedance is detected below 10Ω, its weighting factor is increased to 0.8. The compensation capacity calculation step size is set to ±25kVar. Capacitor combinations are matched using a lookup table. The table data is stored in a memory chip installed in a designated location on the control board. During operation, if the line impedance temperature drift exceeds ±10%, an impedance re-measurement is automatically triggered and the weighting factor set is updated. The capacitor bank can use metallized polypropylene film capacitors (withstand voltage 450V), with the capacity of each bank configured in a 50kVar ratio, and the capacitance deviation controlled within ±3%. A microcontroller can be used as the switching logic controller, with a binary combination strategy search depth set to 4 levels. It is installed in the control unit on the right side of the capacitor compensation cabinet. The driver module has a response time of ≤20ms and a drive voltage of 15V±5%. Switching commands are received via a fiber optic interface. During operation, if the capacitor bank temperature exceeds 85°C, it automatically switches to the backup capacitor branch and triggers the cooling fan to accelerate. This implementation improves the compatibility of the compensation scheme with grid parameters through real-time harmonic spectrum analysis and impedance matching calculations. A dynamic capacitor combination strategy enables refined control of compensation capacity, avoiding the overcompensation risk associated with traditional fixed compensation. A binary search algorithm reduces the computational complexity of the switching logic, and a temperature monitoring mechanism enhances the operational reliability of the capacitor bank.

[0044] In one of the solutions of the present invention, preferably, the flow voltage monitoring module includes: A multimodal sensing unit, which is configured at each charging pile power supply node, includes: a current transformer array, a broadband voltage sensor group, and a temperature-current coupling compensator; An adaptive synchronous acquisition engine configured to: achieve clock synchronization across power supply nodes; dynamically adjust the sampling frequency based on the edge computing controller's policy update cycle; The online harmonic extraction unit is embedded with a sliding window fast Fourier transform (FFT) core to separate the fundamental wave and the 2nd to 50th harmonic components in the current and voltage waveforms in real time, and generate a harmonic spectrum matrix with a time stamp.

[0045] An open-type current transformer (range 0-1000A, accuracy level 0.5) can be selected. The core material uses a laminated silicon steel structure, and the casing is flame-retardant polycarbonate. It is installed on the incoming terminal block of the charging pile power supply node. The wide-band voltage sensor can be configured with a voltage divider resistor network (temperature coefficient ±50ppm / °C), with a bandwidth covering 0-5kHz, and is installed on the insulating bracket on the inner wall of the distribution cabinet. The temperature-current coupling compensator uses a temperature sensor (temperature measurement range -40°C to 150°C) mounted on the surface of the current transformer casing, and a three-wire wiring system is used to eliminate the influence of lead resistance. During operation, the current signal is converted into a voltage signal by the Hall element and input into a high-precision analog-to-digital conversion chip together with the voltage sampling value to generate waveform data with temperature compensation. A clock module (with a timing accuracy of ±50μs) can be used as the master clock source, and synchronization signals are distributed to each monitoring node via a bus. The sampling frequency is dynamically adjusted based on the edge computing controller's policy cycle: when the policy update cycle is 5 seconds, the sampling frequency is set to 4kHz; when the cycle is extended to 30 seconds, the frequency is reduced to 1kHz to reduce power consumption. The clock synchronization engine runs on a microcontroller installed on the monitoring module's main control board. During operation, the inter-node clock deviation is checked every 5 minutes. If the deviation exceeds ±100μs, a recalibration of the synchronization signal is triggered, and the calibration data is stored in the memory chip. Sliding window FFT processing can be performed using a digital signal processor, with an analysis window length set to 10 power frequency cycles and a frequency resolution of 0.5 Hz. Harmonic separation covers the 2nd to 50th harmonic order, with fundamental frequency tracking accuracy of ±0.1 Hz. The generated harmonic spectrum matrix is ​​timestamped with 1 ms accuracy and stored in a buffer installed in a slot on the signal processing board. During operation, if a harmonic amplitude is detected exceeding 8% of the fundamental, the resolution of the harmonic analysis is automatically increased to 0.1 Hz, and the abnormal spectrum data is uploaded via the interface. This implementation combines multimodal sensing with high-precision synchronous data acquisition to enhance the spatial and temporal consistency of grid parameter monitoring. The wideband measurement design meets the requirements of harmonic analysis, while the temperature compensation mechanism reduces the impact of environmental factors on sampling accuracy. The real-time processing capabilities of online harmonic extraction provide a reliable data foundation for dynamic compensation, while the modular hardware layout facilitates on-site maintenance and upgrades.

[0046] In one of the solutions of the present invention, preferably, the edge computing controller further includes: The multimodal communication unit integrates the enhanced HPLC protocol for power line carrier communication and the LoRa wireless mesh network dual channel. It automatically switches the wireless link when the carrier communication signal-to-noise ratio is less than 15dB, ensuring data synchronization latency less than 100ms and compatibility with the requirements of the bidirectional communication link between modules. The bidirectional communication link includes: a command channel for issuing control strategy parameters to the dynamic power distribution adjustment module; and a monitoring channel for receiving the current and voltage data stream collected in real time by the current and voltage monitoring module. The redundant communication guarantee module is configured to activate the Bluetooth Mesh emergency communication link when it is detected that the dual-channel communication quality of the enhanced HPLC protocol and the LoRa wireless network decreases at the same time. The emergency link adopts a time-sliced ​​retransmission mechanism.

[0047] A power line carrier communication module (operating in the 2-30MHz frequency band) can be used. The signal-to-noise ratio detection threshold is set to 15dB. If the signal-to-noise ratio falls below 12dB for three seconds, the system automatically switches to the wireless communication module. The wireless communication module can be configured for a specific frequency band (transmit power 20dBm), with a receiving sensitivity sufficient for long-distance transmission. It is installed on the communication sub-board in the control cabinet. The carrier coupling circuit utilizes a high-permeability magnetic ring and high-temperature-resistant insulation material and is installed in an isolation slot on the communication backplane. During operation, the carrier signal is coupled to the power line via a current transformer. When the channel bit error rate exceeds the set threshold, the forward error correction mechanism is activated. The low-power emergency communication link uses a module that supports multi-node protocols (maximum 32 nodes), with a time-sliced ​​retransmission period set to 50ms. It is installed in a backup communication slot. When the bit error rate of the primary communication channel simultaneously exceeds 1E-6, emergency network initialization is triggered via a general-purpose input / output pin. The emergency link utilizes an encrypted transmission protocol, with a packet size limit of 64 bytes and a routing table update interval of 10 seconds. During operation, inter-node communication quality is dynamically assessed based on signal strength, and relay paths are automatically switched when continuous packet loss occurs. The dual-channel quality monitoring unit can be configured with a wireless receiver chip and a carrier demodulation chip, with a signal-to-noise ratio sampling period of 1 second. The channel switching decision algorithm runs on a microcontroller installed on the communication module's main control board. When the main channel latency exceeds 150ms and the wireless signal strength falls below a threshold, an isolation circuit activates the emergency communication network. The emergency link retransmission limit is set to 5 times, and the data confirmation timeout is set to 200ms. A dual-color LED is installed on the front of the control panel to indicate the communication status. This implementation enhances the communication reliability of the power distribution management system in complex electromagnetic environments through a redundant design of multiple communication media. A dynamic channel switching mechanism ensures the continuous transmission of critical control commands, and a time-sliced ​​retransmission strategy optimizes the transmission efficiency of emergency links. Encrypted transmission and quality monitoring enhance data security, and a modular communication architecture facilitates adaptive configuration in different scenarios.

[0048] Example An intelligent charging pile power distribution management system includes a current and voltage monitoring module configured at the incoming terminal block of the charging pile power supply node, utilizing an open-type current transformer and a broadband voltage sensor. The current transformer's core utilizes a laminated silicon steel structure, while its housing is constructed of flame-retardant polycarbonate. The voltage sensor's voltage divider resistors utilize metal film resistors with a temperature coefficient of ±50ppm / °C, mounted on insulating brackets on the inner wall of the distribution cabinet. The module synchronously acquires three-phase current and voltage waveforms at a 4kHz sampling frequency, generating time-stamped data packets using an AD7606 analog-to-digital conversion chip. Cross-node clock synchronization deviation is controlled within ±50μs. The load forecasting module's embedded industrial computer runs a sliding window interpolation algorithm, aligning the 15-minute sampling intervals of historical data with the 1-second intervals of real-time data, filling the time gaps within a maximum tolerance of ±2 seconds. The hybrid forecasting model utilizes a 128-unit bidirectional LSTM neural network in parallel with a 1st- to 3rd-order adaptive ARIMA algorithm. The LSTM weight coefficient is dynamically increased to above 0.7 based on the three-phase imbalance change rate (Δ) exceeding 3% / second. If Δ exceeds the limit for five consecutive times, the forecast window switches from 30 minutes to a high-frequency 15-minute mode. The switching signal is transmitted via an optocoupler isolation circuit. The edge computing controller uses an STM32H7 microcontroller to implement a multi-objective optimization function. The junction temperature gradient variance weight is set to 0.6, and the harmonic distortion suppression weight is set to 0.3. Thermal stress data is received via the CAN bus every 5 seconds. If the junction temperature gradient variance exceeds 0.4 three times consecutively, the weight is automatically increased to 0.7. The generated relay on-off timing (operation time ≤ 10ms) is controlled by the IR2110 driver chip. The CPSS-200A solid-state relay module is mounted on a forced air cooling baseplate with an on-resistance of ≤ 0.5mΩ. The dynamic power distribution adjustment module's parallel compensation capacitor bank has a total capacity of 500 kVar and utilizes metallized polypropylene film capacitors. IGBT modules enable graded switching with ±5 kVar step accuracy. A PT100 temperature sensor monitors relay junction temperature in real time, with PID parameters set to Kp=0.8 / Ki=0.05 / Kd=0.1. When the temperature difference between adjacent relays exceeds 10°C, the duty cycle is automatically adjusted to maintain a temperature rise difference of ≤15°C. During execution, if harmonic distortion exceeds 7%, matching capacitor units are activated within 20 ms, and load phases are switched according to Pareto optimal timing. The system achieves dynamic balance of power distribution in charging pile groups through high-precision sensing and hierarchical optimization control, reduces the impact of transient processes on power grid equipment, and improves power quality and equipment operation reliability.

[0049] Although the embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and exemplary embodiments. They can be applied to a variety of fields suitable for the embodiments of the present invention. Those skilled in the art will readily realize further modifications. Therefore, without departing from the general concept defined by the claims and their equivalents, the embodiments of the present invention are not limited to the specific details and embodiments shown and described herein.

Claims

1. A power distribution management system for an intelligent charging pile, characterized in that: include: The current and voltage monitoring module is configured at each charging pile power supply node to collect three-phase current and voltage waveform data in real time to obtain real-time current and voltage data; The load prediction module is connected to the current and voltage monitoring module to align and integrate the historical usage data of charging piles with the real-time current and voltage data in time and space to generate a distribution demand forecast value; a dynamic power distribution adjustment module, which is in communication with the load prediction module and receives the power distribution demand prediction value output by the load prediction module; An edge computing controller is communicatively connected to the current and voltage monitoring module, the load prediction module, and the dynamic power distribution adjustment module, wherein the edge computing controller receives real-time current and voltage data from the current and voltage monitoring module, receives power distribution demand forecast values ​​from the load prediction module, and executes a model predictive control algorithm based on multi-source data to generate power distribution real-time control strategy parameters including relay timing parameters and capacitor compensation schemes, and sends the control strategy parameters to the dynamic power distribution adjustment module. The dynamic power distribution adjustment module receives the power distribution real-time control strategy parameters from the edge computing controller and executes the power distribution real-time control strategy parameters through the solid-state relay array and the parallel compensation capacitor group.

2. The power distribution management system for the intelligent charging pile according to claim 1, characterized in that: The load prediction module includes: The data fusion unit is used to align and fuse the discrete historical usage data of the charging pile with the real-time current and voltage data in time and space to construct a multi-dimensional time series input vector; A hybrid prediction model, which is composed of an LSTM neural network and an ARIMA algorithm in parallel through a weighted fusion unit, receives the multidimensional time series input vector, wherein the LSTM network processes nonlinear charging behavior characteristics and the ARIMA algorithm captures periodic load fluctuation patterns. The result output of the hybrid prediction model is synchronized with the policy update cycle of the edge computing controller; The prediction window dynamic adjustment unit dynamically selects a 15-30 minute prediction window based on real-time load fluctuation data. When the fluctuation rate exceeds the three-phase imbalance change rate threshold Δ>3% / second, it sends a mode switching signal to the edge computing controller and automatically switches to the 15-minute high-frequency prediction mode.

3. The power distribution management system for the smart charging pile according to claim 2, characterized in that: The dynamic power distribution adjustment module includes: The dynamic balancing control unit uses an intelligent timing optimization strategy based on a genetic algorithm to generate a Pareto-optimal on-off timing combination based on the control strategy parameters issued by the edge computing controller, ensuring that the load differences between phases dynamically meet the rated value interval constraints. The optimized timing parameters are then fed back to the hierarchical prediction model architecture of the edge computing controller. The harmonic compensation decision unit integrates an FFT harmonic analysis core and a fuzzy logic controller. When the 5%-10% harmonic distortion threshold is detected, it generates a hierarchical switching instruction linked to the capacitor compensation scheme and transmits the compensation effect data back to the data fusion unit of the load prediction module. A phase-to-phase coupling suppression circuit that injects reverse offset current during the switching process of the solid-state relay array and inputs inrush current suppression data into the optimization objective function of the edge computing controller; The thermal stress balancing module monitors the junction temperature distribution of solid-state relays in real time and dynamically adjusts the on-off timing duty cycle to control the temperature rise difference of each relay in the array within the range of ±15°C. It also feeds back the thermal status data to the edge computing controller as a real-time constraint condition for the model predictive control algorithm.

4. The power distribution management system for the smart charging pile according to claim 3, characterized in that: The edge computing controller includes: A hierarchical prediction model architecture decomposes the model predictive control algorithm into a second-level rolling optimization layer and a minute-level strategic planning layer. The time window of the rolling optimization layer dynamically matches the forecast output period of the load forecast module and receives the thermal stress data output by the thermal stress equalization module in real time as a constraint. The strategic planning layer integrates historical calibration data of the confidence feedback mechanism to generate long-term optimization constraints, with a maximum time window not exceeding the 30-minute upper limit set by the forecast window dynamic adjustment unit. A dynamic event triggering mechanism unit constructs an adaptive threshold function based on real-time load fluctuation data. When the three-phase imbalance change rate Δ>3% / second or the harmonic distortion gradient ΔTHD>1.5% / second, it performs an emergency update of the control strategy parameters and synchronously adjusts the LSTM-ARIMA fusion weight distribution ratio of the hybrid prediction model and the operating mode of the prediction window dynamic adjustment unit.

5. The power distribution management system for the smart charging pile according to claim 2, characterized in that: The hybrid prediction model includes: A dynamic weight allocation unit, whose input end is in communication with the dynamic event trigger mechanism unit of the edge computing controller, receives trigger signals of the three-phase imbalance change rate Δ and the harmonic distortion gradient ΔTHD in real time, and whose output end is connected to the prediction result fusion interface of the LSTM neural network and the ARIMA algorithm; Two-channel feature extractor, including: The first feature channel embeds a bidirectional LSTM neural network with a temporal attention mechanism to extract temporal features of non-steady-state current spikes and sudden changes in user charging behavior in the charging pile historical data; The second characteristic channel is configured with an ARIMA algorithm module with adaptive difference order adjustment to track the periodic fluctuation of the power grid load through a sliding time window; The weighted fusion unit includes: The weight calculation subunit dynamically adjusts the fusion weight ratio of the LSTM neural network and the ARIMA algorithm according to the real-time values ​​of the three-phase imbalance change rate Δ and the harmonic distortion gradient ΔTHD output by the dynamic event trigger mechanism, where: When the three-phase imbalance change rate Δ>3% / second or the harmonic distortion gradient ΔTHD>1.5% / second is detected, the weight ratio of the LSTM neural network is increased to above 70%; under steady-state conditions, the weight ratio of the ARIMA algorithm is maintained in a periodic fluctuation range of 50%-70%; The tensor fusion subunit superimposes the nonlinear feature vector output by LSTM and the periodic component generated by ARIMA according to real-time weights to generate a fused prediction value; The clock synchronization interface precisely aligns the time resolution of the fusion prediction value output by the weighted fusion unit to the policy update period of the edge computing controller.

6. The power distribution management system for the intelligent charging pile according to claim 2, characterized in that: The data fusion unit includes: Spatiotemporal alignment engine, including: The multi-source timestamp calibration module dynamically matches the discrete timestamps of the charging pile's historical usage data with the continuous sampling timestamps of the real-time current and voltage data, and uses a sliding window interpolation algorithm to fill the time gaps in the historical data. The node topology mapping module maps the non-uniformly distributed discrete historical data to the spatial coordinate system of the real-time monitoring node according to the physical location relationship of the charging pile power supply nodes; Data integrity enhancers, including: The missing data compensation subunit performs Gaussian process regression based on the correlation coefficient of the current of adjacent nodes when it detects that the historical data is missing for more than 10% of the sampling period; The abnormal data cleaning sub-unit removes outlier data points exceeding ±20% of the rated current through a dynamic threshold comparison method; The multidimensional vector generator constructs input vectors from the time-space aligned data according to the following dimensions: the first dimension is the amplitude ratio of the fundamental component to the harmonic components of the real-time three-phase current; the second dimension is the second-order difference characteristics of the historical load curve; the third dimension is the time-varying parameters of the power supply node impedance matrix; and the fourth dimension is the relay junction temperature gradient fed back by the thermal stress balancing module. The timing constraint verification interface uses a hardware timer to check the clock synchronization deviation of each dimension of the input vector to ensure that the closed-loop control timing jitter is less than 5ms.

7. The power distribution management system for the smart charging pile according to claim 4, characterized in that: The execution of the model predictive control algorithm includes the following steps: a) Multi-objective optimization function construction: The first optimization goal is to minimize the transient inrush current peak value and junction temperature gradient variance of the solid-state relay array; The second optimization goal is to maximize the tracking accuracy of the power distribution demand forecast value output by the load forecast module; The third optimization goal is to suppress the 5%-10% harmonic distortion rate detected by the harmonic compensation decision unit; b) Layered scrolling optimization mechanism: Second-level optimization layer: Using a 5-30 second window, it integrates multi-dimensional time series input vectors and thermal stress data to solve the Pareto frontier solution set of the solid-state relay on-off timing in real time; Minute-level planning layer: Using a dynamically adjusted 15-30 minute forecast window as the constraint boundary, the robustness of the capacitor compensation scheme is optimized and anti-disturbance threshold parameters are generated; c) Dynamic constraint coupling module: The inrush current suppression data of the phase-to-phase coupling suppression circuit and the temperature rise difference data of the thermal stress equalization module are used as real-time inequality constraints for the second-level optimization layer. Dynamically map the fusion weight ratio of the hybrid prediction model to the objective function attenuation factor of the minute-level planning layer; d) Self-healing transmission of policy parameters: When the activation of the dynamic event trigger mechanism is detected, the iterative process of the minute-level planning layer is skipped, and the rolling optimization results of the previous cycle are directly superimposed with the emergency compensation amount, and redundant instructions are issued through the dual channels of the multimodal communication unit.

8. The power distribution management system for the intelligent charging pile according to claim 3, characterized in that: The harmonic compensation decision unit also includes: The adaptive harmonic spectrum analysis module is configured to: use a sliding window fast Fourier transform (FFT) to decompose the current waveform in real time and generate an amplitude distribution spectrum of harmonic components from 0 to 50; Construct a set of harmonic weight factors, where the weight value of each harmonic is inversely proportional to the corresponding harmonic impedance value in the power supply node impedance matrix. When the distortion rate of a specific harmonic order exceeds a threshold, a capacitor compensation capacity value adaptively matching the grid impedance is generated based on the frequency characteristics of the harmonic order and the line impedance parameters. Dynamic capacitor matching array, including: programmable switching capacitor group clusters, each group of capacitance value is configured in a geometric step-by-step manner, and the compensation accuracy control required by the anti-disturbance threshold parameter is achieved through a binary combination strategy; The capacitor bank switching logic controller selects the optimal capacitor combination scheme according to the compensation capacity value output by the adaptive harmonic spectrum analysis module.

9. The power distribution management system for the smart charging pile according to claim 4, characterized in that: The current voltage monitoring module includes: A multimodal sensing unit, which is configured at each charging pile power supply node, includes: a current transformer array, a broadband voltage sensor group, and a temperature-current coupling compensator; An adaptive synchronous acquisition engine configured to: achieve clock synchronization across power supply nodes; dynamically adjust the sampling frequency based on the edge computing controller's policy update cycle; The online harmonic extraction unit is embedded with a sliding window fast Fourier transform (FFT) core to separate the fundamental wave and the 2nd to 50th harmonic components in the current and voltage waveforms in real time, and generate a harmonic spectrum matrix with a time stamp.

10. The power distribution management system for the intelligent charging pile according to claim 2, characterized in that: The edge computing controller further includes: The multimodal communication unit integrates the enhanced HPLC protocol for power line carrier communication and the LoRa wireless mesh network dual channel. It automatically switches the wireless link when the carrier communication signal-to-noise ratio is less than 15dB, ensuring data synchronization latency less than 100ms and compatibility with the requirements of the bidirectional communication link between modules. The bidirectional communication link includes: a command channel for issuing control strategy parameters to the dynamic power distribution adjustment module; and a monitoring channel for receiving the current and voltage data stream collected in real time by the current and voltage monitoring module. The redundant communication guarantee module is configured to activate the Bluetooth Mesh emergency communication link when it is detected that the dual-channel communication quality of the enhanced HPLC protocol and the LoRa wireless network decreases at the same time. The emergency link adopts a time-sliced ​​retransmission mechanism.