Photovoltaic high penetration rate power distribution network multi-point voltage comprehensive control system and method

Abstract

The application discloses a photovoltaic high-permeability power distribution network multi-point voltage comprehensive control system and method, which comprises a photovoltaic output and load data acquisition module, a space-time coupling voltage sensitivity dynamic graph convolution calculation module, a source-load collaborative probability prediction module, a digital twin mapping module, a distributed feeder automation 5G collaborative control module and a voltage control execution module. The method collects real-time data, calculates a voltage sensitivity matrix at multiple times through dynamic graph convolution, combines source-load probability prediction results, simulates voltage distribution in multiple scenes in a digital twin body, and generates optimized adjustment instructions through 5G collaborative control to control equipment to realize multi-point voltage adjustment. The application greatly improves the sensing accuracy of photovoltaic and load changes, voltage simulation accuracy and control response efficiency, comprehensively improves the voltage regulation stability and reliability of the power distribution network, and guarantees the safe operation of the power distribution network under high photovoltaic permeability.

Description

Technical Field

[0001] This invention relates to the field of multi-point voltage control technology in power distribution networks, and particularly to a comprehensive multi-point voltage control system and method for power distribution networks with high photovoltaic penetration. Background Technology

[0002] As the penetration rate of photovoltaic (PV) energy in distribution networks continues to increase, the operating characteristics of these networks are undergoing significant changes. PV power output is intermittent and fluctuating, and coupled with real-time changes in load power, this leads to frequent voltage fluctuations at various nodes in the distribution network, even exceeding limits, thus affecting the safe and stable operation of the network. Currently, distribution network voltage control increasingly relies on technologies such as data acquisition, model calculation, and collaborative control. It requires combining spatiotemporal voltage sensitivity analysis, source-load power prediction, digital twin simulation, and 5G communication collaboration to achieve precise voltage regulation at multiple points. This is necessary to adapt to the complex operating scenarios of distribution networks with high PV penetration and meet the requirements for real-time, accuracy, and collaborative voltage control.

[0003] Existing technologies for multi-point voltage control in high-penetration photovoltaic distribution networks suffer from two significant drawbacks. Firstly, current voltage control methods largely rely on voltage data from single moments or localized areas for analysis and decision-making, failing to adequately consider the dynamic changes in the distribution network topology and the spatiotemporal coupling of voltages between different nodes. This results in incomplete voltage sensitivity calculations, making it difficult to accurately reflect the combined impact of photovoltaic output and load power changes on the voltage of each node, thus affecting the accuracy of voltage regulation commands. Secondly, existing control methods lack effective deep integration of source-load collaborative prediction and digital twin mapping. They cannot accurately simulate voltage distribution under different photovoltaic output and load power scenarios using digital twins. Furthermore, during control command transmission and execution, the low-latency characteristics of 5G communication are not fully utilized to achieve efficient coordination among feeder terminal units, easily leading to control command response delays or execution conflicts, making it difficult to achieve rapid and coordinated multi-point voltage regulation in the distribution network. Summary of the Invention

[0004] In order to overcome the shortcomings and deficiencies of the existing technology, the present invention provides a multi-point voltage integrated control system and method for photovoltaic high penetration distribution networks.

[0005] The technical solution adopted in this invention is a multi-point voltage integrated control system for a high-penetration photovoltaic distribution network, comprising: a photovoltaic output and load data acquisition module, which acquires real-time data on the output power of each photovoltaic inverter, the load power consumption of each node, and the bus voltage within the distribution network, and transmits the acquired data to a data processing module; a spatiotemporally coupled voltage sensitivity dynamic graph convolution calculation module, which receives standardized data output from the data processing module, constructs a distribution network topology graph structure, calculates the sensitivity matrix of each node voltage to photovoltaic output and load power at different times based on graph convolution operators, and transmits the sensitivity matrix to a source-load collaborative probability prediction module; and a source-load collaborative probability prediction module, which receives the sensitivity matrix, combines it with historical photovoltaic output and load data, establishes a probability prediction model, and outputs the photovoltaic output and load power of each node within a preset future time period. The probability distribution results are mapped to the digital twin mapping module. This module receives the probability distribution results, constructs a digital twin of the distribution network, maps the results to the digital twin model, simulates the voltage distribution of the distribution network under different photovoltaic output and load power scenarios, and outputs the voltage simulation results to the distributed feeder automation 5G collaborative control module. The distributed feeder automation 5G collaborative control module receives the voltage simulation results, obtains real-time status information of each feeder terminal unit through the 5G communication link, generates voltage regulation commands based on the voltage simulation results and terminal status information, and transmits them to the voltage control execution module. The voltage control execution module receives the voltage regulation commands and controls the reactive power output of the photovoltaic inverter, the tap changer switching of the on-load tap changer, and the switching of the static var compensator, performing multi-point voltage regulation of the distribution network.

[0006] Furthermore, the model expression used in the spatiotemporal coupled voltage sensitivity dynamic graph convolution calculation module is as follows: ,in, Let be the voltage sensitivity coefficient of node i to node j at time t. It is the Sigmoid activation function. Let i be the set of adjacent nodes of node i. These are the adjacency matrix elements of the distribution network topology graph. The graph convolution weight matrix, This provides the combined photovoltaic output and load power data for node k at time t. The bias vector is used; the probability distribution model adopted by the source-load collaborative probability prediction module is: ,in, for Photovoltaic power output at all times With load power The joint probability density, Let covariance matrix be the variance matrix. It is the mean vector. It is a random vector.

[0007] Furthermore, the expression for the distribution network voltage simulation model in the digital twin mapping module is as follows: ,in, for The analog voltage at time node n, Let be the reference voltage at node n. This refers to the number of photovoltaic access nodes. for The deviation of photovoltaic output from the reference value at time node i. For the number of load nodes, for The deviation of the load power at time node j from the reference value. The nodes at time t are respectively The voltage sensitivity coefficient for node n; the instruction generation model for the distributed feeder automation 5G collaborative control module is as follows: ,in, This represents the voltage adjustment command for node n at time t. For adjustment coefficients, The voltage reference value for node n. For the number of historical moments, for The analog voltage at time node n, Let be the simulated voltage at node n at time t.

[0008] Furthermore, the dynamic topology update model of the spatiotemporal coupled voltage sensitivity dynamic graph convolution calculation module is as follows: ,in, Let be the adjacency matrix elements of nodes i and k at time t. For topological inertia coefficient, The nodes at time t are respectively The combined data of photovoltaic output and load power The maximum value of the fused data difference between nodes i and k. As an indicator variable, it takes the value 1 when nodes i and k are directly connected, and 0 otherwise; the error correction model for the source-load collaborative probability prediction module is as follows: ,in, for Predict the mean vector at any given time. For correction factor, for The vector of actual photovoltaic output versus load power at any given time. for Predict the photovoltaic output and load power vector in real time. for Predict the mean vector at any given time.

[0009] Furthermore, the scenario iteration model for the digital twin mapping module is as follows: ,in, for Real-time digital twin simulation results The simulation results are at time t. The iteration step size, The voltage data of the distribution network actually observed at time t. The simulation result at time t is given by the input vector. The gradient; the communication delay compensation model for the distributed feeder automation 5G collaborative control module is: ,in, This is the voltage regulation command after compensation. Let be the 5G communication delay of node n at time t. As a reference delay, This is the original voltage adjustment command for node n at time t.

[0010] Furthermore, the multi-scale feature fusion model of the spatiotemporal coupled voltage sensitivity dynamic graph convolution calculation module is as follows: ,in, Let i be the multi-scale fusion feature of node i at time t. For scale quantity, Let be the weight at the s-th scale. For graph convolution operations at scale s, ... Here is the distribution network topology at time t; the edge probability calculation model of the source-load coordination probability prediction module is: ,in, for The edge probability density of photovoltaic power output at any given moment Let be the joint probability density.

[0011] Furthermore, the source-load coordinated probability prediction module includes: a data preprocessing unit, which receives the sensitivity matrix output by the spatiotemporally coupled voltage sensitivity dynamic graph convolution calculation module, identifies outliers in the sensitivity matrix, fills in missing data using interpolation methods, and aligns the processed sensitivity matrix with historical photovoltaic power output and load data in time to form a dataset with a unified time dimension; a feature extraction unit, which performs feature engineering on the dataset, extracts the intraday variation features of photovoltaic power output, the seasonal fluctuation features of load power, and the spatiotemporal correlation features of the sensitivity matrix, uses principal component analysis to reduce the dimensionality of the extracted features, retains the main feature information, and generates feature vectors; a probability model construction unit, which constructs a Gaussian mixture probability model based on the feature vectors, determines the number of model components and initial parameters, and uses the expectation-maximization algorithm to iteratively optimize the model parameters so that the similarity between the probability distribution output by the model and the historical data distribution meets a preset threshold; and a prediction result output unit, which inputs the time features of a future preset period into the optimized probability model, calculates the probability distribution parameters of photovoltaic power output and load power at each time, including mean, variance, and quantiles, and encapsulates the probability distribution results into a standard data format for transmission to the digital twin mapping module.

[0012] Furthermore, the digital twin mapping module includes: a twin construction unit, which acquires the geographical wiring diagram, equipment parameters, and topology data of the distribution network, builds a digital twin framework consistent with the actual distribution network in a virtual environment, maps the physical parameters of photovoltaic inverters, transformers, and feeder equipment to the twin model, and establishes a one-to-one correspondence between the equipment model and the actual equipment; and a data mapping unit, which receives the probability distribution results output by the source-load coordination probability prediction module, maps the photovoltaic output and load power data in the probability distribution results to the corresponding equipment models and node positions in the digital twin according to the time series and spatial node coordinates, and performs data mapping to the twin. Spatial matching; Scene simulation unit, based on the mapped data, calls the circuit simulation engine in the digital twin to simulate the voltage calculation process of each node in the distribution network under different combinations of photovoltaic output and load power, considering the impact of line impedance and equipment loss on voltage, and outputs the time-domain waveform data of node voltage under each scenario; Result analysis unit, performs feature extraction on the voltage waveform data output by the scenario simulation, calculates the maximum value, minimum value and fluctuation range of the voltage of each node, compares it with the allowable range of distribution network voltage, identifies voltage over-limit nodes and over-limit periods, and integrates the voltage simulation results with the over-limit information and transmits them to the distributed feeder automation 5G collaborative control module.

[0013] Furthermore, the distributed feeder automation 5G collaborative control module includes: a status acquisition unit, which establishes a communication connection with each feeder terminal unit in the distribution network through a 5G communication module, acquires switch status, current, voltage, and equipment operating status data uploaded by the terminal units according to a preset sampling frequency, verifies the acquired data, removes invalid data, and generates a real-time status dataset; and an instruction generation unit, which receives the voltage simulation results and over-limit information output by the digital twin mapping module, combines them with the real-time status dataset, uses a fuzzy control algorithm to calculate the voltage regulation amount of each over-limit node, and determines the reactive power regulation amplitude of the photovoltaic inverter and the transformer based on the regulation amount. The system switches the connection gear and the reactive power compensation device switching capacity to generate an initial voltage regulation command. The command optimization unit performs a feasibility analysis on the initial voltage regulation command, considering the equipment adjustment range, adjustment speed, and synergistic effects between different devices. It uses a particle swarm optimization algorithm to optimize the regulation command, avoiding frequent equipment actions and adjustment conflicts, and generates the optimal voltage regulation command. The command issuing unit distributes the optimal voltage regulation command to the corresponding feeder terminal unit according to the equipment address code, issues the command in real time through the 5G communication link, records the command issuance time and content, and receives command execution feedback information from the terminal unit, forming a command execution closed loop.

[0014] A multi-point voltage integrated control system for photovoltaic high-penetration distribution networks. This system is implemented through a multi-point voltage integrated control method for photovoltaic high-penetration distribution networks, including:

[0015] Step 1: The photovoltaic output and load data acquisition module collects real-time data on the output power of each photovoltaic inverter, the load power consumption of each node, and the bus voltage within the distribution network. The collected data is then transmitted to the data processing module for standardization, resulting in standardized data. Step 2: The spatiotemporal coupled voltage sensitivity dynamic graph convolution calculation module receives the standardized data, constructs the distribution network topology, and calculates the sensitivity matrix of each node's voltage to photovoltaic output and load power at different times based on graph convolution operators. This sensitivity matrix is ​​then transmitted to the source-load collaborative probability prediction module. Step 3: The source-load collaborative probability prediction module receives the sensitivity matrix, combines it with historical photovoltaic output and load data to establish a probability prediction model, and outputs the probability distribution results of photovoltaic output and load power at each node within a preset future time period to the digital twin mapping module. Step 4: ... The digital twin mapping module receives the probability distribution results, constructs a digital twin of the distribution network, maps the probability distribution results to the digital twin model, simulates the voltage distribution of the distribution network under different scenarios, and outputs the voltage simulation results to the distributed feeder automation 5G collaborative control module; Step 5: The distributed feeder automation 5G collaborative control module receives the voltage simulation results, obtains the real-time status information of each feeder terminal unit through the 5G communication link, generates and optimizes the voltage regulation command based on the voltage simulation results and terminal status information, and transmits it to the voltage control execution module; Step 6: The voltage control execution module receives the voltage regulation command, controls the reactive power output of the photovoltaic inverter, the tap changer switching of the on-load tap changer, and the switching of the static var compensator, and simultaneously collects the voltage data of each node after regulation and feeds it back to the data processing module to form a control closed loop.

[0016] Beneficial Effects: This invention proposes a multi-point voltage integrated control system and method for photovoltaic high-penetration distribution networks. It acquires real-time data across all dimensions through a photovoltaic output and load data acquisition module, and constructs a dynamic distribution network topology using a spatiotemporally coupled voltage sensitivity dynamic graph convolution calculation module. This accurately calculates multi-time, multi-node voltage sensitivity matrices, fully considering the spatiotemporal coupling relationship between topology dynamic changes and node voltages. This solves the problem of incomplete sensitivity calculations in existing technologies, significantly improving the perception accuracy of the impact of photovoltaic output and load power changes, thereby enhancing the accuracy of voltage regulation command generation. Simultaneously, the source-load collaborative probability prediction module and the digital twin mapping module deeply integrate... This system integrates and outputs a high-precision source-load power probability distribution, simulating voltage distribution across multiple scenarios in a digital twin. Combined with a distributed feeder automation 5G collaborative control module, it leverages the low latency of 5G to achieve efficient collaboration among feeder terminal units. This addresses the shortcomings of existing technologies in source-load prediction and digital twin integration, as well as poor collaboration, significantly improving the accuracy of voltage simulation and the response speed of control commands, while avoiding execution conflicts. Ultimately, the voltage control execution module precisely regulates equipment, comprehensively enhancing the stability and reliability of multi-point voltage regulation in high-penetration photovoltaic distribution networks, ensuring the safe operation of the distribution network, and fully meeting the high requirements of voltage control for real-time performance, accuracy, and collaboration. Attached Figure Description

[0017] Figure 1 This is a diagram showing the system module composition of the present invention;

[0018] Figure 2 This is a flowchart of the method steps of the present invention. Detailed Implementation

[0019] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0020] like Figure 1 As shown, the multi-point voltage integrated control system for a photovoltaic high-penetration distribution network includes:

[0021] The photovoltaic output and load data acquisition module collects real-time data on the output power of each photovoltaic inverter, the power consumption of each node load, and the bus voltage within the distribution network, and transmits the collected data to the data processing module.

[0022] Specifically, the photovoltaic output and load data acquisition module is the foundation of the entire control system's data input. It is responsible for acquiring key real-time data related to voltage regulation within the distribution network, providing accurate and comprehensive data support for subsequent calculations and decisions by various modules. The completeness and timeliness of its data acquisition directly affect the accuracy of subsequent voltage sensitivity calculations, source-load predictions, and control command generation. The module needs to cover all photovoltaic access nodes and load nodes within the distribution network. The data acquisition targets include the output power of each photovoltaic inverter, the load power consumption of each node, and the bus voltage. The acquired data must accurately reflect the operating status of the distribution network, providing raw data for subsequent steps such as constructing the distribution network topology and calculating the voltage sensitivity matrix. This is a prerequisite for achieving precise multi-point voltage control in a high-penetration photovoltaic distribution network. In the specific implementation of this module, a data acquisition terminal with high-frequency sampling capabilities is used, with a sampling frequency set to acquire data every 200 milliseconds to ensure the capture of rapid fluctuations in photovoltaic output and load power. The data acquisition range for photovoltaic inverter output power covers 0 to 1000 kW to adapt to the output of photovoltaic power plants of different capacities; the data acquisition range for load power consumption is 0 to 800 kW, including power variations of different types of loads such as residential and commercial loads; the data acquisition range for bus voltage is 0.9 to 1.1 times the rated voltage, i.e., 0.342 to 0.418 kV for a 0.38 kV distribution network and 9 to 11 kV for a 10 kV distribution network. The acquisition terminal is connected to the photovoltaic inverter, load monitoring device, and bus voltage sensor via an RS485 communication interface. After data preprocessing (removing obvious outliers), the acquired data is transmitted to the data processing module via Ethernet at a transmission rate of no less than 100 Mbps to ensure that the data is transmitted from the acquisition terminal to the data processing module within 1 second, guaranteeing data timeliness.

[0023] The spatiotemporal coupled voltage sensitivity dynamic graph convolution calculation module receives standardized data output from the data processing module, constructs a distribution network topology graph structure, calculates the sensitivity matrix of each node voltage to photovoltaic output and load power at different times based on the graph convolution operator, and transmits the sensitivity matrix to the source-load collaborative probability prediction module.

[0024] Specifically, the spatiotemporal coupled voltage sensitivity dynamic graph convolution calculation module is the core component for achieving accurate voltage analysis. It receives standardized data from the data processing module, constructs a distribution network topology, and uses a dynamic graph convolution algorithm to calculate the sensitivity matrix of each node's voltage to photovoltaic output and load power at different times. The calculation results reflect the degree of impact of photovoltaic and load changes on voltage under different spatiotemporal conditions, providing key sensitivity parameters for the source-load coordination probability prediction module. It serves as a crucial bridge connecting data acquisition and source-load prediction. The module must fully consider the dynamic changes in the distribution network topology (such as topology adjustments caused by line switching and equipment failures) and the spatiotemporal coupling characteristics of voltage (the mutual influence of voltage between different nodes at different times) to ensure that the sensitivity matrix accurately reflects the dynamic characteristics of the distribution network operation, providing a scientific basis for subsequent voltage regulation strategy formulation. In practice, the module first receives standardized data from the data processing module (including normalized photovoltaic output, load power, and bus voltage data), and constructs an initial topology based on the distribution network geographical wiring diagram and equipment parameters. The topology includes at least 50 nodes, including photovoltaic access nodes, load nodes, and interconnection nodes. When using the dynamic graph convolution algorithm, a time window length of 10 minutes is set, meaning the topology graph and sensitivity matrix are updated every 10 minutes to adapt to dynamic changes in the distribution network topology. The convolution kernel size is set to 3×3 to ensure sufficient capture of voltage correlation characteristics between adjacent nodes. During the calculation, the impact of photovoltaic output and load power changes of its neighboring nodes (no more than 8) on the node's voltage is analyzed, with matrix operations used to obtain the voltage sensitivity coefficient of each node at each time point. The calculation accuracy of the sensitivity coefficient is controlled within ±0.001 kV / kW. After the calculation is completed, the sensitivity matrix is ​​transmitted to the source-load cooperative probability prediction module in the form of a two-dimensional array. The data transmission volume per batch does not exceed 100KB to ensure efficient data transmission.

[0025] The source-load collaborative probability prediction module receives the sensitivity matrix, combines it with historical photovoltaic power output and load data, establishes a probability prediction model, and outputs the probability distribution results of photovoltaic power output and load power of each node within a preset future time period to the digital twin mapping module.

[0026] Specifically, the source-load collaborative probability prediction module is responsible for predicting future trends in the source-load changes of the distribution network. It receives the sensitivity matrix output from the spatiotemporally coupled voltage sensitivity dynamic graph convolution calculation module, and constructs a probability prediction model by combining historical photovoltaic (PV) output and load data. It outputs the probability distribution results of PV output and load power at each node within a preset future time period, providing predictive data support for the digital twin mapping module. Its prediction accuracy directly affects the accuracy of the digital twin simulation and the rationality of subsequent control commands. The module needs to achieve collaborative analysis of source-load data and the sensitivity matrix, fully utilizing the voltage-source-load correlation characteristics reflected by the sensitivity matrix to improve the adaptability of the prediction model to the operating patterns of the distribution network. This avoids prediction biases caused by traditional prediction models relying solely on historical data, providing a guarantee for early warning and adjustment strategies for voltage fluctuations in the distribution network under high PV penetration. During implementation, the module first receives the sensitivity matrix and simultaneously retrieves historical PV output and load data from the past three months (data sampling interval consistent with the acquisition module, 200 milliseconds). The historical data is time-series aligned to ensure consistency with the time dimension of the sensitivity matrix. When constructing the probabilistic prediction model, the future prediction period is set to 1 hour, and the prediction time interval is 5 minutes. This means the model outputs the probability distribution of source load power for each node every 5 minutes within the next hour. The model training sample size is no less than 10,000 sets, including different weather conditions such as sunny, cloudy, and rainy days, as well as different load characteristics such as weekdays and holidays. By analyzing the sensitivity of each node's voltage to changes in source load in the sensitivity matrix, differentiated weights are assigned to the source load prediction of different nodes (the weight coefficient for sensitive nodes is set to 1.2, and the weight coefficient for non-sensitive nodes is set to 0.8), improving the prediction accuracy of the model for sensitive nodes. After prediction, the mean, variance, and 95% confidence interval of the source load power for each node are output as probability distribution parameters. These parameters are transmitted to the digital twin mapping module in tabular form, with data transmission latency controlled within 2 seconds to ensure that the prediction results can support the digital twin simulation in a timely manner.

[0027] The digital twin mapping module receives probability distribution results, constructs a digital twin of the distribution network, maps the probability distribution results to the digital twin model, simulates the voltage distribution of the distribution network under different photovoltaic output and load power scenarios, and outputs the voltage simulation results to the distributed feeder automation 5G collaborative control module.

[0028] Specifically, the digital twin mapping module is crucial for achieving linkage between virtual simulation and actual operation of the distribution network. It receives the probability distribution results output by the source-load collaborative probability prediction module, constructs a digital twin highly consistent with the actual distribution network, and maps the predicted source-load probability distribution results into the digital twin model. This simulates the distribution network voltage distribution under different source-load scenarios. The simulation results provide an intuitive and accurate voltage status reference for the distributed feeder automation 5G collaborative control module, serving as the core link between source-load prediction and control execution. The module must ensure that the topology and equipment parameters of the digital twin completely match the actual distribution network, while possessing high-precision voltage simulation capabilities to realistically reproduce the impact of different source-load changes on voltage. This provides a virtual experimental environment for control strategy verification and optimization, avoiding operational risks to the distribution network caused by actual trial and error. In practical implementation, the module first constructs a digital twin based on the actual geographical wiring diagram of the distribution network and equipment parameters (including line resistance, reactance, transformer ratio and capacity, photovoltaic inverter capacity and regulation range, etc.). The parameter error of the equipment model in the twin is controlled within ±2%, and the topology matches the actual distribution network 100%. After receiving the source-load probability distribution results, the predicted source-load power values ​​(including the mean and extreme values ​​within the 95% confidence interval) of each node are mapped to the corresponding node and equipment models in the digital twin according to the time series (every 5 minutes). The time delay of the mapping process does not exceed 1 second. When starting the voltage simulation calculation, the Newton-Raphson power flow calculation method is adopted, the iteration accuracy is set to 10^-6, and the number of iterations does not exceed 10 to ensure the accuracy of the simulation results. The simulation scope covers all nodes of the distribution network (no less than 50), and outputs the voltage value and voltage fluctuation range of each node at each time point. After the simulation is completed, the voltage simulation results of each node (including voltage values, whether the limit is exceeded, etc.) are compiled into a data report and transmitted to the distributed feeder automation 5G collaborative control module. The data volume of each batch of reports does not exceed 50KB to ensure data transmission efficiency.

[0029] The distributed feeder automation 5G collaborative control module receives voltage simulation results, obtains real-time status information of each feeder terminal unit through the 5G communication link, generates voltage regulation commands based on voltage simulation results and terminal status information, and transmits them to the voltage control execution module.

[0030] Specifically, the distributed feeder automation 5G collaborative control module is the core of voltage regulation command generation and collaborative execution. It receives voltage simulation results from the digital twin mapping module, acquires real-time status information of each feeder terminal unit via the 5G communication link, and generates voltage regulation commands by combining the voltage simulation results with the terminal status. The rationality and timeliness of the commands directly determine the effectiveness of voltage control, making it a crucial link between digital twin simulation and actual voltage regulation. The module must fully utilize the low latency and high bandwidth characteristics of 5G communication to achieve real-time monitoring and command issuance to each feeder terminal unit. It must also possess command optimization capabilities to avoid conflicts when different terminal units execute commands, ensuring the coordination and efficiency of multi-point voltage regulation in the distribution network and adapting to the regulation needs of rapid voltage fluctuations in the distribution network under high photovoltaic penetration. During implementation, the module establishes communication connections with each feeder terminal unit (no fewer than 30) within the distribution network via a 5G base station. The downlink rate of 5G communication is no less than 100Mbps, the uplink rate is no less than 50Mbps, and the communication latency is controlled within 20 milliseconds to ensure real-time transmission of status information and rapid command issuance. First, the voltage simulation results output by the digital twin mapping module are received to identify voltage over-limit nodes (the over-limit standard is 0.95 to 1.05 times the rated voltage; exceeding this range is considered an over-limit) and the degree of over-limit. Simultaneously, real-time status information of each feeder terminal unit is collected via the 5G link, including switch status (open / closed), equipment operating temperature (range 0 to 80℃), and current adjustment quantities (such as inverter reactive power output value and transformer tap position). Based on the voltage over-limit situation and terminal status, a fuzzy control algorithm is used to generate preliminary adjustment commands. The commands include the terminal unit number to be adjusted, the adjustment type (reactive power adjustment, tap switching, reactive power compensation device switching), and the adjustment quantity. Then, a conflict detection algorithm (detecting whether different commands conflict with equipment actions) optimizes the preliminary commands, with an optimization time not exceeding 500 milliseconds. Finally, the optimal adjustment command is generated and sent to the corresponding feeder terminal unit via the 5G link, ensuring a 100% command delivery success rate.

[0031] The voltage control execution module receives voltage regulation commands and controls the reactive power output of the photovoltaic inverter, the tap switching of the on-load tap changer, and the switching of the static var compensator to perform multi-point voltage regulation of the distribution network.

[0032] Specifically, the voltage control execution module is the final execution link of voltage regulation commands. It receives voltage regulation commands output from the distributed feeder automation 5G collaborative control module and achieves precise voltage regulation at multiple points in the distribution network by controlling the operating status of equipment such as photovoltaic inverters, on-load tap-changing transformers, and static var compensators. The accuracy and response speed of its execution directly determine the effectiveness of the voltage control target and are a key component of the entire closed-loop control system. The module needs to establish reliable control connections with various regulation devices to ensure accurate reception and execution of regulation commands. It also needs to have status feedback capabilities, feeding back the execution results and regulated voltage data to the data processing module to form a control closed loop. This ensures that the distribution network voltage remains continuously stable within the allowable range, adapting to the frequent voltage fluctuations in distribution networks under high photovoltaic penetration. In specific implementation, the module establishes control connections with photovoltaic inverters (no fewer than 20 units), on-load tap-changing transformers (no fewer than 5 units), and static var compensators (no fewer than 3 units) via industrial Ethernet, with a communication rate of no less than 100Mbps and a control command execution delay of no more than 1 second. Upon receiving the adjustment command, the module executes corresponding control actions according to the command type: For photovoltaic inverters, it controls their reactive power output, with an adjustment range of 0 to 50% of their rated capacity (e.g., 0 to 500 kvar for a 1000 kW inverter), and an adjustment accuracy of ±10 kvar; for on-load tap changers, it controls their tap switching, with no fewer than 17 tap positions, each tap adjusting by 0.5% of the rated voltage, and a switching time not exceeding 5 seconds; for static var compensators (SVCs), it controls their switching capacity, with a switching step size of 50 kvar and a switching response time not exceeding 200 milliseconds. After the adjustment action is completed, the module collects the adjusted voltage data of each node through voltage sensors (the collection frequency is consistent with the data acquisition module, at 200 milliseconds / time), organizes the equipment execution status (e.g., whether the adjustment is completed, whether the adjustment amount meets the standard) and voltage data, and feeds it back to the data processing module. The amount of feedback data per batch does not exceed 30KB to ensure the real-time performance and integrity of the control closed loop.

[0033] Preferably, the model expression used in the spatiotemporal coupled voltage sensitivity dynamic graph convolution calculation module is: ,in, Let be the voltage sensitivity coefficient of node i to node j at time t. It is the Sigmoid activation function. Let i be the set of adjacent nodes of node i. These are the adjacency matrix elements of the distribution network topology graph. The graph convolution weight matrix, This provides the combined photovoltaic output and load power data for node k at time t. The bias vector is used; the probability distribution model adopted by the source-load collaborative probability prediction module is: ,in, for Photovoltaic power output at all times With load power The joint probability density, Let covariance matrix be the variance matrix. It is the mean vector. It is a random vector.

[0034] Specifically, the spatiotemporal coupled voltage sensitivity dynamic graph convolution calculation module and the source-load collaborative probability prediction module improve the module's calculation accuracy and prediction reliability by constructing specific model expressions. The spatiotemporal coupled voltage sensitivity dynamic graph convolution calculation module employs a model that introduces an activation function to process the calculation results of adjacent node data and weight matrix. The elements of the adjacent matrix accurately reflect the topological connection relationship of the distribution network. The weight matrix is ​​determined through training with a large amount of historical data, and the bias vector is used to correct calculation deviations. This ensures that the calculation accuracy of the sensitivity coefficient of each node voltage to photovoltaic output and load power at different times is controlled within ±0.001 kV / kW, providing accurate sensitivity parameters for subsequent predictions. The probability distribution model of the source-load collaborative probability prediction module reflects the correlation between photovoltaic output and load power through the covariance matrix, and the mean vector reflects the average level of both. This model is built based on historical data from the past 3 months with a sampling interval of 200 milliseconds and no less than 10,000 training samples. It can output the joint probability density of source-load power of each node every 5 minutes in the next hour, and the transmission delay of the prediction results is controlled within 2 seconds. This provides high-quality prediction data for the digital twin mapping module, ensures the accuracy of voltage simulation, and effectively overcomes the problems of one-sided sensitivity calculation and disconnect between prediction and control in existing technologies.

[0035] Preferably, the expression for the distribution network voltage simulation model in the digital twin mapping module is: ,in, for The analog voltage at time node n, Let be the reference voltage at node n. This refers to the number of photovoltaic access nodes. for The deviation of photovoltaic output from the reference value at time node i. For the number of load nodes, for The deviation of the load power at time node j from the reference value. The nodes at time t are respectively The voltage sensitivity coefficient for node n; the instruction generation model for the distributed feeder automation 5G collaborative control module is as follows: ,in, This represents the voltage adjustment command for node n at time t. For adjustment coefficients, The voltage reference value for node n. For the number of historical moments, for The analog voltage at time node n, Let be the simulated voltage at node n at time t.

[0036] Specifically, the digital twin mapping module and the distributed feeder automation 5G collaborative control module improve the system's simulation capabilities and control efficiency by constructing voltage simulation models and command generation models. The voltage simulation model of the digital twin mapping module is based on the node reference voltage, combined with sensitivity coefficients and source-load power deviation to calculate the simulated voltage. The reference voltage is determined according to the rated voltage of the distribution network (e.g., 0.38 kV for a 0.38 kV distribution network and 10 kV for a 10 kV distribution network). The source-load power deviation is the difference between the predicted value and the reference value. This model uses the Newton-Raphson power flow calculation method, with an iteration accuracy of 10^-6, no more than 10 iterations, and a simulation range covering all nodes (no fewer than 50). It outputs the simulated voltage of each node at each time point, with the simulation result error controlled within ±2%. The instruction generation model of the 5G collaborative control module for distributed feeder automation introduces an adjustment coefficient to balance the impact of real-time simulated voltage and historical simulated voltage on the instructions. The adjustment coefficient is determined through on-site debugging (α is usually 0.6, β is 0.4), and the number of historical time points is set to 10. Based on the terminal status information (including switch status, equipment temperature, etc.) obtained by 5G communication, the model generates adjustment instructions by combining the voltage simulation results. The instruction generation time does not exceed 500 milliseconds, and the transmission delay is controlled within 20 milliseconds through the 5G link, ensuring timely and accurate execution of instructions and solving the problems of insufficient simulation and delayed control response in existing technologies.

[0037] Preferably, the dynamic topology update model of the spatiotemporal coupled voltage sensitivity dynamic graph convolution calculation module is as follows: ,in, Let be the adjacency matrix elements of nodes i and k at time t. For topological inertia coefficient, The nodes at time t are respectively The combined data of photovoltaic output and load power The maximum value of the fused data difference between nodes i and k. As an indicator variable, it takes the value 1 when nodes i and k are directly connected, and 0 otherwise; the error correction model for the source-load collaborative probability prediction module is as follows: ,in, for Predict the mean vector at any given time. For correction factor, for The vector of actual photovoltaic output versus load power at any given time. for Predict the photovoltaic output and load power vector in real time. for Predict the mean vector at any given time.

[0038] Specifically, the topology update of the spatiotemporally coupled voltage sensitivity dynamic graph convolution calculation module and the error correction of the source-load cooperative probability prediction module improve the system's adaptability and prediction accuracy by constructing a dynamic topology update model and an error correction model. The dynamic topology update model of the spatiotemporal coupled voltage sensitivity dynamic graph convolution calculation module introduces a topology inertia coefficient to retain historical topology information. The inertia coefficient is set to 0.7. At the same time, it combines the source-load fusion data difference of nodes and indicator variables to adjust the adjacency matrix elements. The source-load fusion data difference is the absolute value of the data of two nodes. The maximum value is determined by statistical historical data. The indicator variable is set according to the node connection relationship (1 for direct connection, 0 otherwise). This model updates the topology graph and adjacency matrix every 10 minutes to ensure that the topology structure and the actual distribution network match 100% at all times, improving the adaptability of sensitivity calculation. The error correction model of the source-load cooperative probability prediction module uses a correction coefficient to adjust the prediction mean vector. The correction coefficient is set to 0.1. The difference between the actual value and the predicted value is used to correct the subsequent prediction mean. This model updates the prediction mean vector every 5 minutes to keep the deviation between the prediction mean and the actual value within ±5%, which greatly improves the prediction accuracy and avoids voltage regulation deviation caused by dynamic changes in topology and prediction errors. It overcomes the shortcomings of poor topology adaptability and low prediction accuracy in existing technologies.

[0039] Preferably, the scenario iteration model for the digital twin mapping module is as follows: ,in, for Real-time digital twin simulation results The simulation results are at time t. The iteration step size, The voltage data of the distribution network actually observed at time t. The simulation result at time t is given by the input vector. The gradient; the communication delay compensation model for the distributed feeder automation 5G collaborative control module is: ,in, This is the voltage regulation command after compensation. Let be the 5G communication delay of node n at time t. As a reference delay, This is the original voltage adjustment command for node n at time t.

[0040] Specifically, the scenario iteration of the digital twin mapping module and the communication delay compensation of the distributed feeder automation 5G collaborative control module improve the simulation accuracy and control reliability of the system by constructing scenario iteration models and delay compensation models. The scene iteration model of the digital twin mapping module controls the update magnitude of the simulation results with an iteration step size of 0.3. It optimizes the simulation results by combining the difference between the actual observed voltage and the simulation results and the gradient of the simulation results. The actual observed voltage is collected by a voltage sensor (sampling frequency of 200 ms / time). The gradient of the simulation results reflects the sensitivity of the simulation results to the input vector. The model iterates the simulation results every 5 minutes, reducing the deviation between the simulation results and the actual observed voltage to within ±1%, significantly improving the simulation accuracy. The communication delay compensation model of the distributed feeder automation 5G collaborative control module introduces a reference delay (set to 50 ms). The adjustment commands are compensated according to the ratio of the actual communication delay to the reference delay. The actual communication delay is monitored in real time through the 5G link. The compensated commands can offset the impact of the delay on the execution effect. This model ensures that the execution error of the compensated commands is controlled within ±3%, and the command issuance success rate remains 100%. It effectively solves the problems of insufficient simulation accuracy and the impact of communication delay on the control effect in the existing technology, ensuring the stable operation of the system in complex communication environments.

[0041] Preferably, the multi-scale feature fusion model of the spatiotemporal coupled voltage sensitivity dynamic graph convolution calculation module is as follows: ,in, Let i be the multi-scale fusion feature of node i at time t. For scale quantity, Let be the weight at the s-th scale. For graph convolution operations at scale s, ... Here is the distribution network topology at time t; the edge probability calculation model of the source-load coordination probability prediction module is: ,in, for The edge probability density of photovoltaic power output at any given moment Let be the joint probability density.

[0042] Specifically, the feature fusion module of the spatiotemporally coupled voltage sensitivity dynamic graph convolution calculation module and the edge probability calculation module of the source-load co-probability prediction module improve the system's feature extraction capability and prediction comprehensiveness by constructing a multi-scale feature fusion model and an edge probability calculation model. The multi-scale feature fusion model of the spatiotemporally coupled voltage sensitivity dynamic graph convolution calculation module sets three scales, each corresponding to a different size graph convolution kernel (2×2, 3×3, and 4×4 respectively). The weights are determined through training data (the sum of the weights for each scale is 1, typically set to 0.2, 0.5, and 0.3 respectively). This model fuses node features extracted from different scales, reducing the dimensionality of the fused feature vector by 30% while retaining over 95% of the key information, improving the efficiency and accuracy of subsequent sensitivity calculations, and shortening the sensitivity coefficient calculation time to within 200 milliseconds; the source-load co-probability prediction module... The edge probability calculation model of the prediction module obtains the edge probability density of photovoltaic output by integrating the joint probability density. The integration range is the possible value range of load power (0 to 800 kW). This model can output the probability distribution of photovoltaic output separately, including mean, variance and quantile. The deviation between the prediction result and the actual photovoltaic output is controlled within ±4%. It provides more detailed prediction data for the digital twin mapping module, so that the voltage simulation can more accurately reflect the impact of photovoltaic output changes on voltage. It overcomes the shortcomings of existing technologies such as incomplete feature extraction and single prediction dimension, and further improves the system voltage control capability.

[0043] Preferably, the source-load coordinated probability prediction module includes: a data preprocessing unit, which receives the sensitivity matrix output by the spatiotemporally coupled voltage sensitivity dynamic graph convolution calculation module, identifies outliers in the sensitivity matrix, fills in missing data using interpolation methods, and aligns the processed sensitivity matrix with historical photovoltaic power output and load data in time to form a dataset with a unified time dimension; a feature extraction unit, which performs feature engineering on the dataset, extracts the intraday variation features of photovoltaic power output, the seasonal fluctuation features of load power, and the spatiotemporal correlation features of the sensitivity matrix, uses principal component analysis to reduce the dimensionality of the extracted features, retains the main feature information, and generates feature vectors; a probability model construction unit, which constructs a Gaussian mixture probability model based on the feature vectors, determines the number of model components and initial parameters, and uses the expectation-maximization algorithm to iteratively optimize the model parameters so that the similarity between the probability distribution output by the model and the historical data distribution meets a preset threshold; and a prediction result output unit, which inputs the time features of a future preset period into the optimized probability model, calculates the probability distribution parameters of photovoltaic power output and load power at each time, including mean, variance, and quantiles, and encapsulates the probability distribution results into a standard data format for transmission to the digital twin mapping module.

[0044] Specifically, the source-load coordinated probability prediction module comprises four units to clarify the implementation process of each stage and ensure the accuracy and usability of the prediction data. The data preprocessing unit receives the sensitivity matrix output from the spatiotemporally coupled voltage sensitivity dynamic graph convolution calculation module, uses the 3σ criterion to identify outliers (data exceeding the mean ± 3 standard deviations is considered outlier), fills in missing data using linear interpolation (the interpolation interval is consistent with the original data sampling interval, 200 milliseconds), and then aligns the processed sensitivity matrix with historical photovoltaic output and load data from the past three months (sampling interval 200 milliseconds) by timestamp to form a dataset with a unified time dimension. Data integrity must reach over 99.5%. The feature extraction unit performs feature engineering on the dataset, extracting features such as the peak-valley difference of intraday photovoltaic output, the periodic coefficient of seasonal load power fluctuations, and the spatiotemporal correlation of the sensitivity matrix. Principal component analysis is used to retain features with a cumulative contribution rate of up to 95%. Principal components of more than 90% reduce the original feature dimension by 40%, generating feature vectors with uniform dimensions. The probability model building unit builds a Gaussian mixture probability model based on the feature vectors, setting the number of components to 5. The model parameters are optimized iteratively using the expectation-maximization algorithm (the number of iterations does not exceed 50, and the convergence threshold is set to 1e-5), so that the KL divergence between the model output probability distribution and the historical data distribution is less than 0.05. The prediction result output unit inputs the time features of the next 1 hour and the time interval of 5 minutes into the optimization model, calculates the mean, variance and 90th quantile of the source load power at each time, and encapsulates them in JSON format for transmission to the digital twin mapping module. The transmission rate is not less than 100Mbps, ensuring that the data is transmitted within 1 second, providing high-quality data support for subsequent voltage simulation.

[0045] Preferably, the digital twin mapping module includes: a twin construction unit, which acquires the geographical wiring diagram, equipment parameters, and topology data of the distribution network, builds a digital twin framework consistent with the actual distribution network in a virtual environment, maps the physical parameters of photovoltaic inverters, transformers, and feeder equipment to the twin model, and establishes a one-to-one correspondence between the equipment model and the actual equipment; and a data mapping unit, which receives the probability distribution results output by the source-load coordination probability prediction module, maps the photovoltaic output and load power data in the probability distribution results to the corresponding equipment models and node positions in the digital twin according to the time series and spatial node coordinates, and performs spatial mapping between the data and the twin. Inter-parameter matching; Scenario simulation unit, based on the mapped data, calls the circuit simulation engine in the digital twin to simulate the voltage calculation process of each node in the distribution network under different combinations of photovoltaic output and load power, considering the influence of line impedance and equipment loss on voltage, and outputs the time-domain waveform data of node voltage under each scenario; Result analysis unit, performs feature extraction on the voltage waveform data output by the scenario simulation, calculates the maximum value, minimum value and fluctuation range of the voltage of each node, compares it with the allowable range of distribution network voltage, identifies voltage over-limit nodes and over-limit periods, and integrates the voltage simulation results with the over-limit information before transmitting it to the distributed feeder automation 5G collaborative control module.

[0046] Specifically, the digital twin mapping module is divided into four units to achieve full-process control from twin construction to simulation result output, thereby improving the realism and effectiveness of voltage simulation. The digital twin construction unit acquires the geographical wiring diagram of the distribution network (scale 1:5000), equipment parameters (line resistance range 0.1-0.5Ω / km, reactance range 0.3-0.8Ω / km, transformer ratio 10kV / 0.4kV, capacity 500-2000kVA), and topology data. It then builds a digital twin framework on the Unity3D platform, controlling the geometric dimension error between the equipment model and the actual equipment within ±3%, establishing a one-to-one mapping relationship to ensure consistency between the twin and the actual distribution network structure. The data mapping unit receives the probability distribution results output by the source-load collaborative probability prediction module and maps the photovoltaic output (0-1000kW) and load power (0-800kW) data of each node to the corresponding node and equipment model in the twin, with a mapping delay not exceeding 500 milliseconds and a data matching accuracy of 100%. The scene simulation unit... The system calls the PSCAD / EMTDC simulation engine to simulate voltage distribution under different scenarios based on mapped data. The simulation step size is set to 10 milliseconds. It considers the impact of line impedance loss (calculated according to actual line parameters) and equipment efficiency (transformer efficiency 95%-98%, inverter efficiency 96%-99%) on voltage. It outputs the time-domain waveform of node voltage under each scenario (duration 1 hour, sampling interval 200 milliseconds). The result analysis unit extracts features from the waveform data, calculates the maximum, minimum and fluctuation range of each node voltage (the fluctuation range must be less than ±5% of the rated voltage), compares it with the allowable range of distribution network voltage (0.95-1.05 times the rated voltage), identifies over-limit nodes and over-limit periods (over-limit duration exceeding 10 seconds is judged to require adjustment), integrates the voltage simulation results and over-limit information into an Excel report, and transmits it to the distributed feeder automation 5G collaborative control module. The report generation time is no more than 2 seconds.

[0047] Preferably, the distributed feeder automation 5G collaborative control module includes: a status acquisition unit, which establishes a communication connection with each feeder terminal unit in the distribution network through a 5G communication module, acquires switch status, current, voltage, and equipment operating status data uploaded by the terminal units according to a preset sampling frequency, verifies the acquired data, removes invalid data, and generates a real-time status dataset; and an instruction generation unit, which receives the voltage simulation results and over-limit information output by the digital twin mapping module, combines them with the real-time status dataset, uses a fuzzy control algorithm to calculate the voltage regulation amount of each over-limit node, and determines the reactive power regulation range of the photovoltaic inverter and the transformer tap based on the regulation amount. The system switches between different voltage levels and adjusts the capacity of the reactive power compensation device to generate an initial voltage regulation command. The command optimization unit performs a feasibility analysis on the initial voltage regulation command, considering the equipment's adjustment range, adjustment speed, and the synergistic effect between different devices. It uses a particle swarm optimization algorithm to optimize the command, avoiding frequent equipment actions and adjustment conflicts, and generating the optimal voltage regulation command. The command issuing unit distributes the optimal voltage regulation command to the corresponding feeder terminal unit according to the equipment address code, issues the command in real time via the 5G communication link, records the command issuance time and content, and receives command execution feedback information from the terminal unit, forming a closed-loop command execution mechanism.

[0048] Specifically, the distributed feeder automation 5G collaborative control module includes four units to realize closed-loop management of commands from acquisition, generation, optimization to issuance, ensuring the timeliness and coordination of control commands. The status acquisition unit establishes connections with more than 30 feeder terminal units via a 5G communication module (using the NR band, downlink rate 100-200Mbps, uplink rate 50-100Mbps). It acquires data such as switch status (open / closed, status recognition accuracy 100%), current (0-1000A), voltage (0.3-10kV), and equipment operating temperature (-20-80℃) at a sampling frequency of 200 milliseconds. Invalid data is removed using a CRC check algorithm (the proportion of failed checks must be less than 0.5%), generating a real-time status dataset. The instruction generation unit receives the voltage simulation results and over-limit information output by the digital twin mapping module. Combining this with the real-time status dataset, it uses a fuzzy control algorithm (fuzzy subsets divided into 5 levels, membership function using a triangular function) to calculate the voltage regulation of the over-limit nodes (reactive power regulation range 0-500kvar, tap changer adjustment levels 1-17, reactive power compensation capacity 0-200kvar), generating a preliminary regulation. The instruction optimization unit performs a feasibility analysis on the initial instructions, considering the equipment adjustment range (inverter reactive power adjustment not exceeding 50% of rated capacity, transformer tap changer switching interval not less than 30 seconds), adjustment speed (reactive power compensation device response time 200-500 milliseconds), and inter-equipment coordination. It uses a particle swarm optimization algorithm (50 particles, 30 iterations, inertia weight 0.5-0.9) to optimize the instructions, avoiding frequent equipment actions (action interval less than 1 minute is considered frequent) and adjustment conflicts (adjustment target deviation between different devices exceeding 10% is considered conflict), generating the optimal instructions. The instruction distribution unit allocates the optimal instructions to the corresponding terminal units according to the equipment address encoding (using 16-bit binary encoding), and distributes them in real time via the 5G link (distribution delay 20-50 milliseconds). Simultaneously, it records the instruction distribution time (accurate to milliseconds) and content, and receives instruction execution feedback information from the terminal units (feedback success rate 100%), forming a closed-loop instruction execution system to ensure the effective implementation of adjustment instructions.

[0049] The spatiotemporal coupled voltage sensitivity dynamic graph convolution model is the core model in this invention for accurately calculating the voltage sensitivity of the distribution network. Essentially, it combines the spatiotemporal characteristics of the distribution network with a graph convolution algorithm to construct a computational model that reflects the voltage correlation between different nodes at different times. To implement this model, real-time data with a 200-millisecond sampling interval is first acquired through a photovoltaic output and load data acquisition module. After standardization by the data processing module, an initial topology map containing no fewer than 50 nodes is constructed based on the geographical wiring diagram of the distribution network and equipment parameters. Then, a 10-minute time window and a 3×3 convolution kernel size are set. Each node and its no more than 8 adjacent nodes are used as the analysis unit. The dynamic graph convolution algorithm calculates the sensitivity coefficient of node voltage to photovoltaic output and load power at each time point. Simultaneously, a topological inertia coefficient (set to 0.7) is introduced to update the topology map and sensitivity matrix every 10 minutes, ensuring that the calculation adapts to the dynamic changes in the distribution network topology, and that the accuracy of the sensitivity coefficient calculation is controlled within ±0.001 kV / kW. The purpose of this model is to provide accurate sensitivity parameters for the source-load co-probability prediction module, reflecting the degree of influence of photovoltaic and load changes on voltage in the spatiotemporal dimension. It overcomes the shortcomings of traditional sensitivity calculations that ignore topological dynamic changes and spatiotemporal coupling characteristics, improves the comprehensiveness and accuracy of voltage influencing factor analysis, lays a data foundation for subsequent source-load prediction and voltage regulation command generation, and ensures the scientific nature of distribution network voltage analysis under high photovoltaic penetration.

[0050] The source-load co-probability prediction and digital twin mapping model is the core model combination in this invention to realize the linkage between source-load prediction and voltage simulation. The former is used to output the source-load power probability distribution, and the latter is used to construct a virtual simulation environment to simulate voltage distribution. The implementation of the source-load collaborative probability prediction model requires receiving the sensitivity matrix output by the spatiotemporally coupled voltage sensitivity dynamic graph convolution model, combining historical data from the past 3 months with a sampling interval of 200 milliseconds, setting a Gaussian mixture model with 5 components, optimizing parameters through the expectation-maximization algorithm (iterations not exceeding 50 times, convergence threshold 1e-5), and outputting the mean, variance, and quantile of the source-load power of each node every 5 minutes in the next hour, with the prediction deviation controlled within ±5%. The implementation of the digital twin mapping model requires building a twin with a geometric error of ±3% on the Unity3D platform based on the 1:5000 scale geographical wiring diagram of the distribution network and equipment parameters (line resistance 0.1-0.5Ω / km, reactance 0.3-0.8Ω / km, etc.), mapping the source-load prediction data to the twin at 5-minute time nodes, calling the PSCAD / EMTDC simulation engine (10 millisecond step size) to simulate voltage distribution, and outputting a 1-hour time domain waveform. The purpose of this model combination is to provide predictive data and simulation results for the distributed feeder automation 5G collaborative control platform, solve the problems of disconnect between prediction and simulation and low simulation accuracy in traditional technologies, realize the visualization simulation of source load changes and voltage distribution, identify voltage over-limit risks in advance, provide accurate scenario basis for control command generation, and support the foresight and accuracy of voltage regulation in photovoltaic high penetration distribution networks.

[0051] The distributed feeder automation 5G collaborative control platform is the core platform in this invention for realizing the closed loop of voltage regulation command generation, issuance, and execution, integrating status acquisition, command processing, and 5G communication functions. The platform requires connection to more than 30 feeder terminal units via a 5G NR band module (downlink 100-200Mbps, uplink 50-100Mbps). It collects data such as switch status, current (0-1000A), and voltage (0.3-10kV) at a frequency of 200 milliseconds, and generates a real-time dataset after CRC verification (invalid data <0.5%). In the command generation stage, a 5-level fuzzy control algorithm is used to calculate the regulation amount (reactive power 0-500kvar, tap changer 1-17, etc.), and then a particle swarm optimization algorithm (50 particles, 30 iterations) is used to consider the equipment regulation range (inverter reactive power ≤ 50% of rated capacity) and coordination to optimize command generation. Finally, commands are issued using 16-bit binary encoding, with 5G latency controlled within 20-50 milliseconds, while receiving feedback ensures a 100% success rate. The platform's function is to transform the voltage simulation results output by the digital twin mapping model into actual adjustment actions, achieving multi-point voltage collaborative control. It overcomes the shortcomings of traditional control platforms, such as high communication latency and poor coordination, leveraging the low latency characteristics of 5G to ensure rapid command response. Through multi-unit collaboration, it avoids equipment action conflicts, ensuring timely and accurate voltage adjustment in distribution networks with high photovoltaic penetration, and maintaining the safe and stable operation of the power grid.

[0052] like Figure 2As shown, a multi-point voltage integrated control system for a high-penetration photovoltaic distribution network is implemented through a multi-point voltage integrated control method for a high-penetration photovoltaic distribution network. The system includes: Step 1: Collecting real-time data on the output power of each photovoltaic inverter, the load power consumption of each node, and the bus voltage within the distribution network through a photovoltaic output and load data acquisition module. The collected data is then transmitted to a data processing module for standardization to obtain standardized data. Step 2: A spatiotemporal coupled voltage sensitivity dynamic graph convolution calculation module receives the standardized data, constructs a distribution network topology graph, and calculates the sensitivity matrix of each node voltage to photovoltaic output and load power at different times based on graph convolution operators. This sensitivity matrix is ​​then transmitted to a source-load collaborative probability prediction module. Step 3: The source-load collaborative probability prediction module receives the sensitivity matrix, combines it with historical photovoltaic output and load data to establish a probability prediction model, and outputs the photovoltaic output and load power of each node within a preset future time period. Step 4: The probability distribution results of the load power are sent to the digital twin mapping module; Step 5: The digital twin mapping module receives the probability distribution results, constructs a digital twin of the distribution network, maps the probability distribution results to the digital twin model, simulates the voltage distribution of the distribution network under different scenarios, and outputs the voltage simulation results to the distributed feeder automation 5G collaborative control module; Step 6: The distributed feeder automation 5G collaborative control module receives the voltage simulation results, obtains the real-time status information of each feeder terminal unit through the 5G communication link, generates and optimizes the voltage regulation command based on the voltage simulation results and terminal status information, and transmits it to the voltage control execution module; Step 7: The voltage control execution module receives the voltage regulation command, controls the reactive power output of the photovoltaic inverter, the tap changer switching of the on-load tap changer, and the switching of the static var compensator, and simultaneously collects the voltage data of each node after regulation and feeds it back to the data processing module to form a control closed loop.

[0053] A multi-point voltage integrated control system and method for photovoltaic high-penetration distribution networks is proposed. This system and method achieve comprehensive perception, accurate prediction, efficient simulation and coordinated control of the voltage of photovoltaic high-penetration distribution networks through data exchange and technical linkage among various modules. It completely solves the problems of incomplete voltage analysis, lagging control response and insufficient regulation accuracy in the background technology, and ensures the safe and stable operation of the distribution network.

[0054] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," "link," and "fix" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0055] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various equivalent changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A multi-point voltage integrated control system for photovoltaic high-penetration distribution networks, characterized in that, include: The photovoltaic output and load data acquisition module collects real-time data on the output power of each photovoltaic inverter, the power consumption of each node load, and the bus voltage within the distribution network, and transmits the collected data to the data processing module. The spatiotemporal coupled voltage sensitivity dynamic graph convolution calculation module receives standardized data output from the data processing module, constructs a distribution network topology graph structure, calculates the sensitivity matrix of each node voltage to photovoltaic output and load power at different times based on the graph convolution operator, and transmits the sensitivity matrix to the source-load collaborative probability prediction module. The source-load collaborative probability prediction module receives the sensitivity matrix, combines it with historical photovoltaic power output and load data, establishes a probability prediction model, and outputs the probability distribution results of photovoltaic power output and load power of each node within a preset future time period to the digital twin mapping module. The digital twin mapping module receives probability distribution results, constructs a digital twin of the distribution network, maps the probability distribution results to the digital twin model, simulates the voltage distribution of the distribution network under different photovoltaic output and load power scenarios, and outputs the voltage simulation results to the distributed feeder automation 5G collaborative control module. The distributed feeder automation 5G collaborative control module receives the voltage simulation results, obtains the real-time status information of each feeder terminal unit through the 5G communication link, generates voltage regulation commands based on the voltage simulation results and terminal status information, and transmits them to the voltage control execution module. The voltage control execution module receives the voltage regulation commands and controls the reactive power output of the photovoltaic inverter, the tap changer switching of the on-load tap changer, and the switching of the static var compensator to perform multi-point voltage regulation of the distribution network.

2. The multi-point voltage integrated control system for photovoltaic high-penetration distribution networks according to claim 1, characterized in that, The model expression used in the spatiotemporal coupled voltage sensitivity dynamic graph convolution calculation module is as follows: ,in, Let be the voltage sensitivity coefficient of node i to node j at time t. It is the Sigmoid activation function. Let i be the set of adjacent nodes of node i. These are the adjacency matrix elements of the distribution network topology graph. The graph convolution weight matrix, This provides the combined photovoltaic output and load power data for node k at time t. The bias vector is used; the probability distribution model adopted by the source-load collaborative probability prediction module is: ,in, for Photovoltaic power output at all times With load power The joint probability density, Let covariance matrix be the variance matrix. It is the mean vector. It is a random vector.

3. The multi-point voltage integrated control system for photovoltaic high-penetration distribution networks according to claim 1, characterized in that, The expression for the distribution network voltage simulation model in the digital twin mapping module is as follows: ,in, for The analog voltage at time node n, Let be the reference voltage at node n. This refers to the number of photovoltaic access nodes. for The deviation of photovoltaic output from the reference value at time node i. For the number of load nodes, for The deviation of the load power at time node j from the reference value. The nodes at time t are respectively The voltage sensitivity coefficient for node n; the instruction generation model for the distributed feeder automation 5G collaborative control module is as follows: ,in, This represents the voltage adjustment command for node n at time t. For adjustment coefficients, The voltage reference value for node n. For the number of historical moments, for The analog voltage at time node n, Let be the simulated voltage at node n at time t.

4. The multi-point voltage integrated control system for photovoltaic high-penetration distribution networks according to claim 1, characterized in that, The dynamic topology update model of the spatiotemporal coupled voltage sensitivity dynamic graph convolution calculation module is as follows: ,in, Let be the adjacency matrix elements of nodes i and k at time t. For topological inertia coefficient, The nodes at time t are respectively The combined data of photovoltaic output and load power The maximum value of the fused data difference between nodes i and k. As an indicator variable, it takes the value 1 when nodes i and k are directly connected, and 0 otherwise; the error correction model for the source-load collaborative probability prediction module is as follows: ,in, for Predict the mean vector at any given time. For correction factor, for The vector of actual photovoltaic output versus load power at any given time. for Predict the photovoltaic output and load power vector in real time. for Predict the mean vector at any given time.

5. The multi-point voltage integrated control system for photovoltaic high-penetration distribution networks according to claim 1, characterized in that, The scenario iteration model for the digital twin mapping module is as follows: ,in, for Real-time digital twin simulation results The simulation results are at time t. The iteration step size, The voltage data of the distribution network actually observed at time t. The simulation result at time t is given by the input vector. The gradient; the communication delay compensation model for the distributed feeder automation 5G collaborative control module is: ,in, This is the voltage regulation command after compensation. Let be the 5G communication delay of node n at time t. As a reference delay, This is the original voltage adjustment command for node n at time t.

6. The multi-point voltage integrated control system for photovoltaic high-penetration distribution networks according to claim 1, characterized in that, The multi-scale feature fusion model of the spatiotemporal coupled voltage sensitivity dynamic graph convolution calculation module is as follows: ,in, Let i be the multi-scale fusion feature of node i at time t. For scale quantity, Let be the weight at the s-th scale. For graph convolution operations at scale s, ... Here is the distribution network topology at time t; the edge probability calculation model of the source-load coordination probability prediction module is: ,in, for The edge probability density of photovoltaic power output at any given moment Let be the joint probability density.

7. The multi-point voltage integrated control system for photovoltaic high-penetration distribution networks according to claim 1, characterized in that, The source-load coordinated probability prediction module includes: a data preprocessing unit, which receives the sensitivity matrix output by the spatiotemporally coupled voltage sensitivity dynamic graph convolution calculation module, identifies outliers in the sensitivity matrix, fills in missing data using interpolation methods, and aligns the processed sensitivity matrix with historical photovoltaic power output and load data in time to form a dataset with a unified time dimension; a feature extraction unit, which performs feature engineering on the dataset, extracts intraday variation features of photovoltaic power output, seasonal fluctuation features of load power, and spatiotemporal correlation features of the sensitivity matrix, uses principal component analysis to reduce the dimensionality of the extracted features, retains the main feature information, and generates feature vectors; a probability model construction unit, which constructs a Gaussian mixture probability model based on the feature vectors, determines the number of model components and initial parameters, and uses the expectation-maximization algorithm to iteratively optimize the model parameters so that the similarity between the probability distribution output by the model and the historical data distribution meets a preset threshold; and a prediction result output unit, which inputs the time features of a future preset period into the optimized probability model, calculates the probability distribution parameters of photovoltaic power output and load power at each time, including mean, variance, and quantiles, and encapsulates the probability distribution results into a standard data format for transmission to the digital twin mapping module.

8. The multi-point voltage integrated control system for photovoltaic high-penetration distribution networks according to claim 1, characterized in that, The digital twin mapping module includes: a twin construction unit, which acquires the geographical wiring diagram, equipment parameters, and topology data of the distribution network, builds a digital twin framework consistent with the actual distribution network in a virtual environment, maps the physical parameters of photovoltaic inverters, transformers, and feeder equipment to the twin model, and establishes a one-to-one correspondence between the equipment model and the actual equipment; and a data mapping unit, which receives the probability distribution results output by the source-load coordination probability prediction module, maps the photovoltaic output and load power data in the probability distribution results to the corresponding equipment models and node positions in the digital twin according to the time series and spatial node coordinates, and performs spatial matching between the data and the twin. The system includes a scenario simulation unit, which, based on the mapped data, calls the circuit simulation engine in the digital twin to simulate the voltage calculation process of each node in the distribution network under different combinations of photovoltaic output and load power. It considers the impact of line impedance and equipment loss on voltage and outputs the time-domain waveform data of node voltage under each scenario. The result analysis unit extracts features from the voltage waveform data output by the scenario simulation, calculates the maximum, minimum and fluctuation range of the voltage at each node, compares it with the allowable voltage range of the distribution network, identifies voltage over-limit nodes and over-limit periods, and integrates the voltage simulation results with the over-limit information before transmitting them to the distributed feeder automation 5G collaborative control module.

9. The multi-point voltage integrated control system for photovoltaic high-penetration distribution networks according to claim 1, characterized in that, The distributed feeder automation 5G collaborative control module includes: a status acquisition unit, which establishes communication connections with each feeder terminal unit in the distribution network via a 5G communication module, acquires switch status, current, voltage, and equipment operating status data uploaded by the terminal units according to a preset sampling frequency, verifies the acquired data, removes invalid data, and generates a real-time status dataset; and an instruction generation unit, which receives voltage simulation results and over-limit information output by the digital twin mapping module, combines them with the real-time status dataset, uses a fuzzy control algorithm to calculate the voltage regulation amount of each over-limit node, and determines the reactive power regulation amplitude of the photovoltaic inverter and the transformer tap changer based on the regulation amount. The system switches gears and adjusts the capacity of the reactive power compensation device to generate an initial voltage regulation command. The command optimization unit performs a feasibility analysis on the initial voltage regulation command, considering the equipment's adjustment range, adjustment speed, and the synergistic effect between different devices. It uses a particle swarm optimization algorithm to optimize the command, avoiding frequent equipment actions and adjustment conflicts, and generating the optimal voltage regulation command. The command issuing unit distributes the optimal voltage regulation command to the corresponding feeder terminal unit according to the equipment address code, issues the command in real time via the 5G communication link, records the command issuance time and content, and receives command execution feedback information from the terminal unit, forming a closed-loop command execution mechanism.

10. The photovoltaic high-penetration distribution network multi-point voltage integrated control system according to any one of claims 1-9, characterized in that, This system is implemented through a multi-point voltage integrated control method for photovoltaic high-penetration distribution networks, including: Step 1: Collect real-time data on the output power of each photovoltaic inverter, the power consumption of each node load, and the bus voltage in the distribution network through the photovoltaic output and load data acquisition module. Transmit the collected data to the data processing module for standardization processing to obtain standardized data. Step 2: The spatiotemporal coupled voltage sensitivity dynamic graph convolution calculation module receives standardized data, constructs a distribution network topology graph structure, calculates the sensitivity matrix of each node voltage to photovoltaic output and load power at different times based on the graph convolution operator, and transmits the sensitivity matrix to the source-load collaborative probability prediction module. Step 3: The source-load collaborative probability prediction module receives the sensitivity matrix, combines it with historical photovoltaic output and load data to establish a probability prediction model, and outputs the probability distribution results of photovoltaic output and load power of each node within a future preset time period to the digital twin mapping module. Step 4: The digital twin mapping module receives the probability distribution results, constructs a digital twin of the distribution network, maps the probability distribution results to the digital twin model, simulates the voltage distribution of the distribution network under different scenarios, and outputs the voltage simulation results to the distributed feeder automation 5G collaborative control module. Step 5: The distributed feeder automation 5G collaborative control module receives the voltage simulation results, obtains the real-time status information of each feeder terminal unit through the 5G communication link, generates and optimizes the voltage regulation command based on the voltage simulation results and terminal status information, and transmits it to the voltage control execution module; Step 6: The voltage control execution module receives the voltage regulation command, controls the reactive power output of the photovoltaic inverter, the tap changer switching of the on-load tap changer, and the switching of the static var compensator. At the same time, it collects the voltage data of each node after regulation and feeds it back to the data processing module to form a control closed loop.

Images