Control method and system of multi-all intelligent connection distributed power acquisition security terminal

Abstract

The application discloses a control method and system of a multi-in-one intelligent distributed power supply acquisition safety terminal, relates to the technical field of intelligent control of power distribution networks, and converts the topology structure of a power distribution network line into a graph structure to obtain a topology graph structure; key time nodes in historical operation data are identified, time sequence attention weights are calculated to weight the historical voltage sequence, and weighted historical data are obtained; the topology graph structure and the weighted historical data are input into a graph neural network for feature aggregation to obtain node embedding representation; a target line is identified according to a comprehensive risk value, a multi-objective optimization function is constructed based on the node embedding representation for solving, and an initial access capacity scheme is obtained; the initial access capacity scheme is substituted into a digital twin model for simulation, an anti-risk coefficient is calculated, and the anti-risk coefficient is corrected in combination with a load threshold value and a reverse power threshold value to obtain an optimal access capacity scheme. The accuracy of distributed power supply access capacity evaluation and the optimization level of the carrying capacity of a power distribution system are significantly improved.

Description

Technical Field

[0001] This application relates to the field of intelligent control technology for power distribution networks, and in particular to a control method and system for a multi-functional intelligent distributed power source acquisition and security terminal. Background Technology

[0002] With the accelerated development of new urban areas and the agglomeration of electricity load, power distribution networks need to achieve localized balance between power supply and demand by connecting distributed generation sources. In existing technologies, the assessment methods for the capacity of distributed generation sources mainly adopt static calculation methods based on empirical formulas. According to the line parameters, load data and voltage constraints of the distribution network, the connectable capacity of each line is determined through a simplified electrical calculation model. Typical methods include capacity calculation based on voltage drop formulas, capacity limitation based on line thermal stability constraints, and access assessment based on short-circuit capacity. These methods provide preliminary technical basis for the grid connection of distributed generation sources in practical applications.

[0003] However, existing assessment methods have significant shortcomings. First, they do not fully utilize the distribution network topology. Existing methods simplify the distribution network into a single-layer tree structure or perform independent calculations on local line segments, ignoring the coupling relationships between nodes and the impact of global topology characteristics on capacity allocation. This results in a lack of a global optimization perspective in capacity assessment. Second, they use a single approach to processing historical operating data. Existing methods assign the same weight to all historical data or only consider average statistical characteristics, failing to identify the special impact of key operating conditions such as peak load periods and extreme weather on capacity assessment. This makes it difficult for the assessment results to reflect the true carrying capacity of the distribution network under different operating scenarios. Third, they lack the ability to coordinate and optimize multiple objectives during capacity allocation. Existing methods typically optimize only a single objective, such as voltage quality or line load rate, lacking a comprehensive consideration of multiple mutually constraining objectives such as voltage quality, line load rate, and line losses. This makes it difficult to achieve the overall optimal performance of the distribution network.

[0004] Existing methods lack dynamic prediction of future operational risks and adaptive adjustment mechanisms for capacity plans. Based on static data at the current moment, they fail to anticipate potential risks such as voltage exceedances and equipment aging that may occur in the distribution network over future operating cycles after distributed generation (DG) integration. When these risks manifest in actual operation, the determined capacity plan is difficult to adjust dynamically, potentially causing the distribution network to deviate from its safety boundaries. Furthermore, existing methods often use fixed safety margin coefficients when determining capacity, failing to differentiate based on the historical performance, equipment status, and load characteristics of different branch lines. This is particularly problematic for lines with stable operation and good equipment condition. Fixed safety margins lead to overly conservative capacity configurations that waste the system's carrying capacity potential. For lines with high operational risks and severely aged equipment, fixed safety margins may not provide sufficient protection, causing safety hazards. Furthermore, existing methods lack a systematic constraint mechanism for dealing with the reverse power generated when distributed generation output exceeds local load. Simply limiting capacity limits is insufficient to accurately control the impact of reverse power on upstream equipment and grid stability. Therefore, there is an urgent need to develop a distributed generation access capacity assessment method that can fully explore the characteristics of the distribution network topology, identify key operating conditions, achieve multi-objective collaborative optimization, dynamically predict operational risks, and adaptively adjust capacity schemes. Summary of the Invention

[0005] This application provides a control method and system for a multi-in-one intelligent distributed power acquisition and safety terminal. By constructing a graph neural network capacity optimization model that integrates topology features and key operating conditions, and combining it with a risk resistance coefficient correction mechanism based on digital twin dynamic simulation, this method solves the problems of insufficient utilization of global topology information, insufficient multi-objective collaborative optimization capability, and lack of dynamic risk prediction and adaptive correction in capacity schemes in existing technologies. This improves the accuracy of distributed power access capacity assessment and the load-bearing capacity optimization level of the power distribution system.

[0006] In a first aspect, this application provides a control method for an all-in-one intelligent distributed power acquisition security terminal, the control method of which includes:

[0007] Step S1: Convert the distribution network line topology into a graph structure to obtain the topology graph structure;

[0008] Step S2: Identify key time nodes in historical operation data, and weight the historical voltage sequence according to the time series attention weights calculated based on the key time nodes to obtain weighted historical data;

[0009] Step S3: Input the topological graph structure and the weighted historical data into the graph neural network and perform feature aggregation to obtain the node embedding representation;

[0010] Step S4: Identify the target line based on the comprehensive risk value, construct a multi-objective optimization function based on the node embedding representation, and solve it to obtain the initial access capacity scheme;

[0011] Step S5: Substitute the initial access capacity scheme into the digital twin model for simulation, calculate the risk resistance coefficient, and make corrections based on the load threshold and reverse power threshold to obtain the optimal access capacity scheme.

[0012] Secondly, this application provides a control system for an all-in-one intelligent distributed power acquisition and security terminal, the control system of which includes:

[0013] The conversion module is used to convert the distribution network line topology into a graph structure to obtain a topology graph structure.

[0014] The weighting module is used to identify key time nodes in historical operating data, and to weight the historical voltage sequence based on the time-series attention weights calculated from the key time nodes to obtain weighted historical data.

[0015] The input module is used to input the topological graph structure and the weighted historical data into the graph neural network for feature aggregation to obtain node embedding representations;

[0016] The solution module is used to identify the target line based on the comprehensive risk value, construct a multi-objective optimization function based on the node embedding representation, and solve it to obtain the initial access capacity scheme.

[0017] The simulation module is used to substitute the initial access capacity scheme into the digital twin model for simulation, calculate the risk resistance coefficient, and make corrections in combination with the load threshold and reverse power threshold to obtain the optimal access capacity scheme.

[0018] Thirdly, a control device for an all-in-one intelligent distributed power acquisition security terminal is provided, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor invokes the instructions in the memory to cause the control device of the all-in-one intelligent distributed power acquisition security terminal to execute the above-described control method of the all-in-one intelligent distributed power acquisition security terminal.

[0019] Fourthly, a computer-readable storage medium is provided, wherein instructions are stored in the computer-readable storage medium, which, when executed on a computer, cause the computer to execute the control method of the all-in-one intelligent distributed power acquisition security terminal described above.

[0020] This application has at least the following substantial technical effects:

[0021] The technical solution provided in this application overcomes the limitations of existing technologies that simplify the distribution network to independent calculations of local line segments and equally weight all historical data by converting the distribution network line topology into a graph structure and identifying key time nodes in historical operating data for time-series attention weighting. The construction of the graph structure preserves the complete connection relationships between nodes and the electrical parameter information of the edges in the distribution network, enabling subsequent analysis to perform capacity optimization based on a global topology perspective rather than relying solely on local line characteristics. The calculation of time-series attention weights identifies peak load days and extreme weather days as key time nodes, assigning higher weights to operating data under these special conditions. This allows the weighted historical data to more accurately reflect the voltage characteristics and load fluctuation patterns of the distribution network under key operating scenarios. Compared to the simple average statistics of historical data in existing technologies, the time-series attention mechanism in this application highlights operating conditions that have a significant impact on capacity assessment, avoiding the dilution of key features by a large amount of routine operating data, thereby improving the sensitivity and accuracy of capacity assessment to actual operating risks. This application inputs the topology graph structure and weighted historical data into a graph neural network to obtain node embedding representations through feature aggregation. This solves the problem that existing technologies cannot effectively integrate distribution network topology information and historical operating modes. Through the message passing and feature aggregation mechanism of multi-layer graph convolution, the graph neural network enables the embedding representation of each node to not only include the node's own electrical characteristics and operating status, but also integrate the topological location information and operating characteristics of its surrounding multi-hop neighbor nodes. This deep feature fusion capability is not available in traditional static calculation methods based on empirical formulas. The node embedding representation provides a rich feature space for subsequent multi-objective optimization, enabling capacity allocation decisions to comprehensively consider the topological importance of nodes in the distribution network, historical operating stability, and the influence relationship of neighboring nodes.

[0022] This application constructs a multi-objective optimization function based on node embedding representation, which includes voltage quality, line load rate, and line loss. It then solves for the initial access capacity scheme using a multi-objective evolutionary algorithm. This overcomes the shortcomings of existing technologies that optimize only a single objective, leading to difficulties in achieving optimal overall distribution network performance. The construction of the multi-objective optimization function simultaneously considers three mutually constraining optimization objectives: minimizing the maximum voltage deviation to ensure voltage quality, minimizing the maximum line load rate to avoid line overload, and minimizing total line loss to improve transmission efficiency. A Pareto optimal solution set is obtained through a non-dominated sorting genetic algorithm, providing decision-makers with multiple non-dominated capacity allocation schemes. The initial access capacity scheme selected from the Pareto optimal solution set based on a normalized comprehensive score achieves a reasonable trade-off among the three optimization objectives, ensuring both voltage quality and operational safety of the distribution network while fully exploring the carrying capacity potential and transmission efficiency of the lines. This application incorporates the initial access capacity scheme into a digital twin model for simulation and calculates the risk resistance coefficient for capacity correction. This addresses the problem of existing technologies lacking the ability to predict future operational risks and dynamically correct based on current static data when assessing capacity. The digital twin model generates predicted load and distributed power output by setting simulation parameters for future time spans, based on historical load curve patterns and meteorological data-driven models. Monte Carlo simulations are run to statistically analyze the number of voltage overruns and the maximum voltage deviation of each branch line. The calculation of the risk resistance coefficient integrates two dimensions: voltage overrun risk and equipment aging risk. The mechanism of correcting the initial access capacity based on the risk resistance coefficient enables adaptive adjustment of the capacity scheme. For lines with low risk resistance coefficients, the access capacity is increased to fully utilize the carrying potential; for lines with high risk resistance coefficients, the access capacity is reduced to ensure operational safety. Combined with the constraint verification of load thresholds and reverse power thresholds, it further ensures that no reverse power flow exceeding safety limits will occur after the distributed power source is connected. The final optimal access capacity scheme considers both the global performance of multi-objective optimization and has undergone risk verification and adaptive correction through dynamic simulation, significantly improving the scientificity, accuracy, and safety of distributed power source access capacity configuration. Attached Figure Description

[0023] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a schematic diagram of one embodiment of the control method for the all-in-one intelligent distributed power acquisition security terminal in this application.

[0025] Figure 2This is a schematic diagram illustrating the relationship between the risk resistance coefficient and the modified access capacity in an embodiment of this application.

[0026] Figure 3 This is a schematic diagram of one embodiment of the control system of the all-in-one intelligent distributed power acquisition and security terminal in this application.

[0027] Figure 4 This is a schematic block diagram of the control device of the multi-in-one intelligent distributed power acquisition and security terminal in this embodiment of the invention. Detailed Implementation

[0028] This application provides a control method and system for an all-in-one intelligent distributed power acquisition security terminal. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0029] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of the control method for the all-in-one intelligent distributed power acquisition security terminal in this application includes:

[0030] Step S1: Convert the distribution network line topology into a graph structure to obtain the topology graph structure;

[0031] Step S2: Identify key time nodes in historical operation data, and weight the historical voltage sequence according to the time series attention weights calculated based on the key time nodes to obtain weighted historical data;

[0032] Step S3: Input the topological graph structure and weighted historical data into the graph neural network to perform feature aggregation and obtain the node embedding representation;

[0033] Step S4: Identify the target line based on the comprehensive risk value, construct a multi-objective optimization function based on the node embedding representation, and solve it to obtain the initial access capacity scheme;

[0034] Step S5: Substitute the initial access capacity scheme into the digital twin model for simulation, calculate the risk resistance coefficient, and make corrections based on the load threshold and reverse power threshold to obtain the optimal access capacity scheme.

[0035] It is understood that the executing entity of this application can be the control system of an all-in-one intelligent distributed power acquisition and security terminal, or it can be a terminal or a server; the specific implementation is not limited here. This application's embodiment uses a server as an example for illustration.

[0036] Specifically, in the process of converting the distribution network line topology into a graph structure, the line connection relationship and equipment distribution information of the distribution network are read. The substation busbars, branch access points and load points in the distribution network are abstracted into a set of nodes in the graph structure, and the lines connecting these nodes are abstracted into a set of edges in the graph structure. Each edge carries the resistance value, reactance value and physical length of the line segment as weight parameters, forming a complete topological graph structure mathematical expression of the network connection characteristics and electrical parameter characteristics of the distribution network.

[0037] When identifying key time nodes in historical operating data, voltage and power measurement records within a preset time period are extracted from the historical database. The difference between the daily load peak and load trough is calculated to obtain the load peak-to-trough difference. The load peak-to-trough difference is compared with a threshold of the historical average plus a standard deviation multiple. When the load peak-to-trough difference exceeds the threshold, the day is marked as a load peak day. At the same time, temperature records from meteorological data are obtained. When the highest temperature of the day exceeds the high temperature threshold or the lowest temperature is below the low temperature threshold, the day is marked as an extreme weather day. Load peak days and extreme weather days are identified as key time nodes. Higher time-series attention weight values ​​are assigned to key time nodes, while lower time-series attention weight values ​​are assigned to ordinary time nodes. The voltage value of each time point in the historical voltage sequence is multiplied by the corresponding time-series attention weight to obtain a weighted voltage value. The weighted voltage values ​​of all time points are arranged in chronological order to form weighted historical data.

[0038] When the topology graph structure and weighted historical data are input into the graph neural network, the graph neural network receives the initial features of the node set and the weight information of the edge set. The first graph convolutional layer calculates the message passing value between each node and its neighboring nodes. The message passing value is calculated based on the features of the neighboring nodes and the weights of the connecting edges. Each node collects the message passing values ​​from all neighboring nodes, normalizes them according to the node degree, and then accumulates them. The accumulated result is used to obtain the node features of the first layer through an activation function. Subsequent graph convolutional layers repeat the message passing and feature aggregation operations to fuse the node features with information from more distant neighboring nodes layer by layer. After multiple layers of graph convolution, the feature representation of each node includes the node's location information in the distribution network topology, the electrical characteristics of the surrounding neighboring nodes, and the historical operating status. The final layer features of all nodes are processed by global pooling to obtain the node embedding representation.

[0039] When identifying target lines based on the comprehensive risk value, the voltage deviation of the trunk line is calculated by dividing the difference between the real-time voltage and the rated voltage by the rated voltage. The line load pressure is calculated by dividing the current transmission power by the line's rated capacity. The frequency of voltage exceedances is recorded by counting the number of times the voltage exceeds the allowable range. The comprehensive risk value is obtained by multiplying the voltage deviation, line load pressure, and frequency of voltage exceedances by their respective weighting coefficients and then summing them. The comprehensive risk value is compared with a preset risk threshold. If the comprehensive risk value exceeds the threshold, the trunk line is determined as the target line. The target line is then divided into several branch lines and constructed based on node embedding representation. A multi-objective optimization function is constructed, which includes a voltage quality objective function that minimizes the maximum voltage deviation of each node, a line load rate objective function that minimizes the maximum load rate of each line, and a line loss objective function that minimizes the total power loss of all lines. A multi-objective evolutionary algorithm is used to set up an initial population where each individual represents an access capacity allocation scheme. A new generation of population is generated through crossover and mutation operations. The three objective function values ​​of each individual are calculated, and non-dominated sorting is performed to identify non-dominated solutions of the Pareto front. After multiple generations of evolution, the scheme with the highest comprehensive score is selected from the Pareto optimal solution set as the initial access capacity scheme.

[0040] When the initial access capacity scheme is substituted into the digital twin model for simulation, the digital twin model establishes a virtual distribution network environment based on the distribution network topology and electrical parameters. The simulation time span is set to cover a future period and divided into multiple time steps. For each time step, a predicted load value is generated based on historical load curve patterns. A distributed generation power output prediction value is generated based on a generation model driven by weather forecast data. The predicted load and distributed generation power output are substituted into the power flow calculation equations to solve for the voltage at each node and the power flow of each line. Multiple Monte Carlo simulations are run for each branch line. The fluctuation range of load and output is randomly sampled in each simulation. The number of times the voltage exceeds the limit for that branch line during the simulation is counted and divided by the total number of simulations to obtain the probability of exceeding the limit. The maximum voltage deviation amplitude of that branch line in all simulations is recorded. A first risk coefficient is calculated based on the probability of exceeding the limit and the maximum voltage deviation amplitude. Based on the branch line... The second risk coefficient is calculated based on the service life of the equipment. The first and second risk coefficients are averaged to obtain the risk resistance coefficient. The modified access capacity is obtained by multiplying the initial access capacity by the correction function of the risk resistance coefficient. The load threshold is calculated as the sum of the rated power of all load equipment on the branch line multiplied by a preset proportional coefficient. When the modified access capacity is less than the load threshold, the modified access capacity is directly used as the optimal access capacity of the branch line. When the modified access capacity is greater than or equal to the load threshold, the reverse power under the most unfavorable operating condition is calculated as the modified access capacity minus the historical minimum load. The reverse power is compared with the reverse power threshold, which is obtained by multiplying the transformer capacity by the allowable reverse power proportional coefficient. When the reverse power exceeds the reverse power threshold, the optimal access capacity of the branch line is recalculated based on the historical minimum load plus the reverse power threshold. After all branch lines are modified, the optimal access capacity scheme is obtained.

[0041] In one specific embodiment, step S1 includes:

[0042] Read the physical topology data of the distribution network, which includes the coordinates of the starting node of the main line, the location of the connection point of the branch line, and the distribution information of the load equipment;

[0043] Real-time acquisition of operational data from various nodes of the distribution network, including bus voltage, line current, active power, and reactive power;

[0044] The physical topology data is mapped to a graph structure. The node set of the graph structure includes bus nodes, branch access point nodes, and load point nodes. The weight of each edge in the edge set of the graph structure includes line resistance, line reactance, and line length, thus obtaining the topology graph structure.

[0045] Specifically, when reading the physical topology data of the distribution network, the starting node coordinates of the main lines are extracted from the database of the distribution network management system. The spatial location information of each main line from the substation outgoing end to the end is recorded using latitude and longitude coordinates or plane rectangular coordinate system. The access point location of the branch lines is extracted. The specific location point where each branch line connects to the main line is recorded, including the line length from the access point to the starting point of the main line and the spatial coordinates of the access point. The distribution information of the load equipment is extracted. The location point of each load equipment is connected to which branch line, including the load equipment type, rated power and installation location coordinates. Real-time acquisition of operational data from various nodes in the distribution network is achieved through measuring devices deployed at key locations within the network. Voltage transformers are installed at substation busbars to measure real-time busbar voltage values, and current transformers are installed on each line segment to measure real-time current flowing through that line. Active power is calculated from the voltage and current measurements; this is equal to the cosine of voltage multiplied by current multiplied by the power factor, reflecting the actual power transmitted by the line. Reactive power is calculated from the voltage multiplied by current multiplied by the power factor, reflecting the reactive component transmitted by the line. The acquisition frequency is set to a certain number of times per second to ensure the capture of dynamic changes in the distribution network's operational status. When mapping physical topology data to a graph structure, a node set is established, abstracting the substation busbars in the distribution network as busbar nodes as the starting point of the power supply of the distribution network. The connection positions of the main lines and branch lines are abstracted as branch access point nodes as key positions of topology bifurcation. The connection positions of load equipment are abstracted as load point nodes as terminal positions of power consumption. An edge set is established, abstracting the line segment connecting two nodes as an edge. Each edge stores the resistance value of the line segment, reflecting the resistance of the conductor material and cross-sectional area to current transmission; stores the line reactance value, reflecting the influence of the electromagnetic field around the conductor on current transmission; and stores the line length value, reflecting the physical distance between the two nodes. These three parameters together constitute the weight attribute of the edge, affecting the voltage drop and power loss calculation of the line segment. After the node set and edge set are constructed, a topology graph structure is formed, which fully expresses the network connection relationship and electrical characteristics of the distribution network.

[0046] In one specific embodiment, each node in the node set stores the node type code, historical average voltage, historical maximum load power, node geographical location code, and node equipment service life, while each edge in the edge set stores the line impedance modulus, line rated capacity, line current load rate, and line historical fault count.

[0047] Specifically, each node in the node set establishes an independent attribute storage space to store five types of feature data. The node type encoding adopts an integer encoding method, encoding the bus node as value 1, the branch access point node as value 2, and the load point node as value 3 to distinguish the functional roles of different types of nodes in the distribution network. The historical average voltage is obtained by extracting the voltage measurement values ​​of all time points in the past preset time period of the node and calculating the arithmetic mean to reflect the typical voltage level of the node. The historical maximum load power is recorded by querying the power peaks of all load equipment connected to the node during the historical operation period to record the maximum power demand that the node has ever undertaken. The node geographical location encoding converts the latitude and longitude coordinates or plane coordinates of the node into numerical vectors to facilitate the expression of the spatial distribution characteristics of the node in subsequent calculations. The working years of the node equipment are calculated from the commissioning date of the corresponding equipment of the node to the current date to obtain the number of years, reflecting the aging degree and reliability level of the equipment. Each edge in the edge set stores four types of feature data describing the electrical characteristics and operating status of the line segment. The line impedance magnitude is calculated by taking the square root of the line resistance and line reactance, and the magnitude of the impedance represents the total degree of obstruction to current transmission by the line segment. The line rated capacity is determined by the material, cross-sectional area, and heat dissipation conditions of the conductor. The maximum current or maximum transmission power allowed to pass through the line segment for a long period of time must not exceed this capacity to ensure the safe operation of the line. The current load rate of the line is calculated by dividing the current actual transmission power of the line segment by the line rated capacity, and the percentage value reflects the load level of the line. The historical fault count of the line is the cumulative value of the number of protection actions, equipment trips, and fault alarms that have occurred during the past operation of the line segment, reflecting the historical reliability performance of the line.

[0048] In one specific embodiment, step S2 includes:

[0049] Extract historical operating data sequences within a preset time period. The historical operating data sequences include the voltage time series of each node and the power flow time series of each edge.

[0050] Calculate the daily load peak-to-valley difference. When the load peak-to-valley difference exceeds the sum of the products of the historical average peak-to-valley difference and the standard deviation, mark that day as a load peak day.

[0051] Historical temperature data is obtained, and when the highest temperature of a day exceeds the first temperature threshold or falls below the second temperature threshold, that day is marked as an extreme weather day. Peak load days and extreme weather days are identified as key time nodes.

[0052] The timing attention weights are calculated based on key time nodes, and the historical voltage sequences are weighted to obtain weighted historical data.

[0053] Specifically, when extracting historical operation data sequences within a preset time period, complete operation records for the past 90 days are read from the historical database of the distribution network monitoring system. For each node in the graph structure, the voltage measurement values ​​of that node at each 15-minute time interval are extracted and arranged in chronological order to form the voltage time sequence of that node. For each edge in the graph structure, the active power transmission values ​​of that edge at each 15-minute time interval are extracted and arranged in chronological order to form the power flow time sequence of that edge. A total of 8,640 time points are collected over 90 days, and each node and each edge has time sequence data of the same length. When calculating the daily load peak-valley difference, each natural day in the 90-day period is processed separately. Power measurements at all time points of that day are extracted to find the maximum and minimum power values. Subtracting the minimum power value from the maximum power value yields the load peak-valley difference for that day, reflecting the severity of load fluctuations. The daily load peak-valley difference data over the 90 days are compiled into a sequence containing 90 values. The arithmetic mean of this sequence is calculated to obtain the historical average peak-valley difference. The standard deviation of this sequence is calculated to reflect the dispersion of the peak-valley difference data. The standard deviation is multiplied by a coefficient of 1.5 and added to the historical average peak-valley difference to obtain the discrimination threshold. The daily load peak-valley difference value is checked day by day. When the load peak-valley difference of a certain day exceeds the discrimination threshold, that day is marked as a load peak day, indicating that the load fluctuations on that day are abnormally severe. When acquiring historical temperature data, the system reads daily temperature records for the past 90 days from the meteorological data interface, including the daily maximum and minimum temperatures. A first temperature threshold of 35 degrees Celsius is set as the high-temperature criterion, and a second temperature threshold of -5 degrees Celsius is set as the low-temperature criterion. Each day is checked, and if the maximum temperature of a certain day exceeds 35 degrees Celsius or the minimum temperature of a certain day is lower than -5 degrees Celsius, that day is marked as an extreme weather day, indicating that the climate conditions on that day are abnormal. All dates marked as peak load days and all dates marked as extreme weather days are summarized to form a set of key time nodes. The set of key time nodes contains the index positions of all time points on these special dates.When calculating the temporal attention weights based on key time nodes, a weight value is assigned to each time point in the historical operational data sequence. It is checked whether the time point belongs to the set of key time nodes. If the date of the time point is only marked as a peak load day, the weight is increased by 0.3 from the baseline weight of 1.0 to obtain a weight of 1.3. If the date of the time point is only marked as an extreme weather day, the weight is increased by 0.5 from the baseline weight to obtain a weight of 1.5. If the date of the time point is marked as both a peak load day and an extreme weather day, the weight is increased by 0.8 from the baseline weight to obtain a weight of 1.8. If the date of the time point is neither a peak load day nor an extreme weather day, the baseline weight of 1.0 is maintained. The historical voltage sequence is then weighted by multiplying the voltage measurement value of each node at each time point by the corresponding temporal attention weight to obtain the weighted voltage value. The weighted voltage values ​​of all nodes at all time points are arranged in the original time order and node order to form weighted historical data. The voltage information of key time nodes in the weighted historical data receives a larger numerical weight and has a more significant impact on subsequent feature learning.

[0054] In one specific embodiment, the timing attention weight is calculated as follows: a first weight increment is added for peak load days, a second weight increment is added for extreme weather days, a third weight increment is added for dates that simultaneously meet the conditions of peak load day and extreme weather day, and a baseline weight value is used for ordinary dates.

[0055] Specifically, the calculation of temporal attention weights employs a hierarchical incremental mechanism, assigning differentiated weight values ​​to different types of time points. For peak load days, when adding the first weight increment, the baseline weight value for all time points on that day is increased by 0.3 as the first weight increment, with the baseline weight value set to 1.0. The weight of the time point on the peak load day is calculated as the baseline weight value plus the first weight increment, equaling 1.3. For extreme weather days, when adding the second weight increment, the baseline weight value for all time points on that day is increased by 0.5 as the second weight increment. The weight of the time point on the extreme weather day is calculated as the baseline weight value plus the second weight increment, equaling 1.5. For dates that simultaneously meet both peak load and extreme weather conditions, when adding the third weight increment, the baseline weight value for all time points on that day is increased by 0.8 as the third weight increment. The weight of the time point on a date that simultaneously meets both conditions is calculated as the baseline weight value plus the third weight increment, equaling 1.8. The value of the third weight increment is greater than the simple sum of the first and second weight increments, reflecting the additional emphasis under the dual special conditions. When using the baseline weight value for ordinary dates, all time points on dates that are neither marked as peak load days nor extreme weather days directly use the baseline weight value of 1.0 without any incremental adjustment. The weight increment is set according to the monotonically increasing principle to ensure that time points with more special conditions receive higher weight values. The first weight increment of 0.3 is less than the second weight increment of 0.5, reflecting that the impact of extreme weather on the distribution network operation is greater than that of simple peak load. The third weight increment of 0.8 is greater than the sum of the first and second weight increments of 0.8, ensuring that the weight under dual conditions is significantly higher than that under single conditions.

[0056] In one specific embodiment, step S3 includes:

[0057] The topological graph structure and weighted historical data are input into the graph neural network model, which contains multiple graph convolutional layers.

[0058] The initial feature vectors of each node are processed by message passing and feature aggregation through the first layer of graph convolutional layer to obtain the node features of the first layer.

[0059] The receptive field of each node is expanded layer by layer by subsequent graph convolutional layers, so that the features of each node are fused with the information of its neighboring nodes to obtain multi-layer node features.

[0060] Global pooling is performed on the multi-layer node features to obtain the node embedding representation.

[0061] Specifically, when the topology graph structure and weighted historical data are input into the graph neural network model, the topology graph structure provides the connection relationship between the set of nodes and the set of edges, the weighted historical data provides the time series features of each node, and the graph neural network model uses a deep architecture with 4 layers of graph convolutional layers to extract the topology features and operation mode of the distribution network layer by layer. When the initial feature vectors of each node are processed by message passing and feature aggregation in the first layer of graph convolutional layer, the initial feature vector of each node is formed by concatenating the node type code, historical average voltage, historical maximum load power, node geographical location code and the working years of the node equipment stored in the node to form a fixed-dimensional vector. In the message passing process, each node sends a message to all its neighboring nodes. The content of the message is calculated based on the node's initial feature vector and the weight parameters of the connecting edges. The weight parameters of the connecting edges include line resistance, line reactance and line length. In the feature aggregation process, each node receives messages from all neighboring nodes and normalizes and scales the multiple message vectors received according to the degree of the sending node and the degree of the receiving node. The normalized message vectors are summed and accumulated. The accumulated result is added to the node's own initial feature vector and then processed by the activation function to obtain the first layer of node features. The activation function uses the modified linear unit function to set negative values ​​to zero and retain positive values. As the receptive field of each node is expanded layer by layer through subsequent graph convolutional layers, the second graph convolutional layer takes the node features of the first layer as input and repeats message passing and feature aggregation operations. In the second layer, the receptive field of each node is expanded to all neighboring nodes within a 2-hop range of that node. The third graph convolutional layer takes the node features of the second layer as input and continues to pass and aggregate. In the third layer, the receptive field of each node is expanded to all neighboring nodes within a 3-hop range of that node. The fourth graph convolutional layer takes the node features of the third layer as input and completes the final pass and aggregation. In the fourth layer, the receptive field of each node is expanded to all neighboring nodes within a 4-hop range of that node. After 4 layers of graph convolution, the feature representation of each node is fused with the topological information and operational state information of all nodes within a 4-hop range of its surrounding neighbors to obtain multi-layer node features. When performing global pooling on multi-layer node features, a graph attention pooling mechanism is used to calculate the importance contribution of each node to the entire graph structure. The features of each node in the fourth layer are input into the multilayer perceptron network to obtain the importance score of that node. After performing an exponential function transformation on the importance scores of all nodes, the attention coefficient of each node is obtained. The fourth-layer features of each node are multiplied by the attention coefficient of that node and then summed over all nodes to obtain the global feature representation at the graph level. At the same time, the fourth-layer features of each node are retained as the node embedding representation of that node. The node embedding representation includes the location features of the node in the distribution network topology, the electrical characteristics of the surrounding neighboring nodes, and the weighted historical operating mode.

[0062] In one specific embodiment, the message passing process includes: calculating the message passing value based on the node's initial feature vector and edge weights; the feature aggregation process includes: normalizing the message passing values ​​of neighboring nodes according to the node degree and then summing them, and obtaining the aggregated node features through an activation function.

[0063] Specifically, when the message passing process calculates the message passing value based on the initial feature vector of a node and the edge weights, for each edge connecting node i and node j in the graph structure, during the process of node j sending a message to node i, the initial feature vector of node j is first read, which includes the node type code, historical average voltage, historical maximum load power, node geographical location code, and the working years of the node equipment to form a feature vector of dimension D. The edge weights connecting node i and node j are read, which include the line resistance, line reactance, and line length of the line segment. The initial feature vector of node j is multiplied by the weight matrix to obtain the transformed feature vector. The dimension of the weight matrix is ​​D multiplied by D and obtained through training. The transformed feature vector is multiplied by the edge weights to obtain the message passing value sent by node j to node i. The edge weights, as scaling factors, adjust the strength of message passing and reflect the influence of the line electrical characteristics on information propagation. When feature aggregation processing normalizes and accumulates the message passing values ​​of neighboring nodes according to the node degree, node i receives message passing values ​​from all neighboring nodes to form a message set. The degree of node i is calculated as the number of edges directly connected to node i. For each message passing value received by node i, the message passing value is divided by the square root of the degree of the sending node and then divided by the square root of the degree of the receiving node i to complete the normalization process. The purpose of normalization is to balance the message passing intensity between nodes with different degrees and avoid the high-degree nodes dominating the feature aggregation process. All normalized message passing values ​​are vector-summed to obtain the aggregated message vector. The aggregated message vector is added to the initial feature vector of node i to obtain the feature vector before the update. The feature vector before the update is added with a bias term and then input into the activation function. The activation function uses a modified linear unit function to process each element of the feature vector before the update. When the element value is greater than 0, the original value is maintained; when the element value is less than or equal to 0, 0 is output. The output of the activation function is the aggregated node feature, which is used as the output feature of the node in the first graph convolutional layer.

[0064] In one specific embodiment, the global pooling process employs a graph attention pooling mechanism, which calculates the attention coefficients of each node using a multilayer perceptron, and then performs a weighted summation of the features of each node based on the attention coefficients to obtain a graph-level feature representation.

[0065] Specifically, the node features output from the fourth graph convolutional layer are input into a multilayer perceptron network. This network contains two fully connected layers. The first fully connected layer maps the node features from the original dimension to the hidden layer dimension, which is half the original dimension. The output of the first fully connected layer is processed by a modified linear unit activation function. The second fully connected layer maps the hidden layer features to a scalar value, representing the importance score of the node to the entire power distribution network structure. The importance scores calculated for all nodes form a score vector. When calculating the attention coefficient of each node, an exponential function transformation is applied to each importance score in the score vector. This transformation maps the importance score to a positive real number space. The sum of the exponential transformation values ​​of all nodes is used as the normalized denominator. The attention coefficient of each node is equal to its exponential transformation value divided by the normalized denominator. The attention coefficient ranges from 0 to 1. The sum of the attention coefficients of all nodes equals 1, satisfying the normalization constraint of the probability distribution. Nodes with higher importance scores receive larger attention coefficients, and nodes with lower importance scores receive smaller attention coefficients. When performing a weighted summation of the features of each node based on the attention coefficient, the fourth-layer feature vector of each node is multiplied by the attention coefficient of that node to obtain a weighted feature vector. The weighted feature vectors of all nodes are then summed along their dimensions to obtain a global feature vector of fixed dimensions. This global feature vector, as a graph-level feature representation, integrates the topological information and operating status of all nodes. Nodes with larger attention coefficients contribute more significantly to the global feature representation, while nodes with smaller attention coefficients contribute relatively less. This graph-level feature representation is used for subsequent risk assessment and capacity optimization decisions.

[0066] In one specific embodiment, step S4 includes:

[0067] The comprehensive risk value is calculated based on the degree of voltage deviation of the main line, the line load pressure, and the frequency of voltage over-limit.

[0068] The comprehensive risk value is compared with the preset risk threshold. When the comprehensive risk value is greater than the preset risk threshold, the main line is marked as a high-risk line and identified as the target line.

[0069] The target line is divided into several branch lines, and a multi-objective optimization function containing voltage quality objective, line load rate objective and line loss objective is constructed based on node embedding representation;

[0070] A multi-objective evolutionary algorithm is used to solve the multi-objective optimization function to obtain the initial access capacity scheme.

[0071] Specifically, when calculating the comprehensive risk value based on the voltage deviation degree, line load pressure, and voltage over-limit frequency of the main line, the voltage deviation degree is calculated by dividing the absolute value of the difference between the current real-time voltage and the rated voltage of the main line by the rated voltage to obtain the normalized voltage deviation ratio. The line load pressure is calculated by dividing the current transmission power of the main line by the rated capacity of the line to obtain the line load rate. The voltage over-limit frequency is calculated by dividing the number of times the voltage of the main line exceeds the allowable range within the statistical monitoring period by the total number of samples within the monitoring period to obtain the over-limit frequency. A weighting coefficient of 0.4 is assigned to the voltage deviation degree to reflect the importance of voltage quality to the safety of the distribution network. A weighting coefficient of 0.35 is assigned to the line load pressure to reflect the impact of the line carrying capacity on system stability. A weighting coefficient of 0.25 is assigned to the voltage over-limit frequency to reflect the reference value of historical over-limit behavior for risk assessment. The comprehensive risk value is obtained by multiplying the voltage deviation degree by the weighting coefficient of 0.4, the line load pressure by the weighting coefficient of 0.35, and the voltage over-limit frequency by the weighting coefficient of 0.25. When comparing the comprehensive risk value with the preset risk threshold, the preset risk threshold is set to 0.6 as the criterion for distinguishing between normally operating lines and high-risk lines. When the comprehensive risk value of a certain trunk line is greater than 0.6, it is determined that the operating risk of the line exceeds the acceptable level. The trunk line is marked as a high-risk line that needs to be connected to a distributed power source for voltage support and power compensation. The high-risk line is then identified as the target line for subsequent capacity optimization. When dividing the target line into several branch lines, all branch access point nodes extending from the target trunk line are identified according to the topology diagram. The line segment between each branch access point node and its downstream load point node constitutes a branch line. The target line is divided into several branch lines with the same number of branch access point nodes. When constructing a multi-objective optimization function based on node embedding representation, the voltage quality objective function is designed to minimize the maximum voltage deviation of all nodes. The voltage value of each node under a given access capacity scheme is obtained through power flow calculation. The absolute value of the deviation between the voltage value of each node and the rated voltage is taken, and the maximum deviation value is used as the function value of the voltage quality objective. The line load rate objective function is designed to minimize the maximum load rate of all lines. The transmission power of each line under a given access capacity scheme is obtained through power flow calculation. The transmission power of each line is divided by the rated capacity of the line to obtain the load rate. The maximum load rate is then used as the function value of the line load rate objective. The line loss objective function is designed to minimize the sum of power losses of all lines. The power loss of each line is equal to the square of the current flowing through the line multiplied by the resistance value of the line. The sum of the power losses of all lines is used to obtain the function value of the line loss objective.When using a multi-objective evolutionary algorithm to solve the multi-objective optimization function, the algorithm adopts a non-dominated sorting genetic algorithm framework. The decision variables are defined as the access capacity vector composed of the access capacity of each branch line. A population size of 100 is set to simultaneously maintain 100 different access capacity schemes. During population initialization, the decision variables of each individual are randomly assigned values ​​to satisfy the upper and lower limits of the access capacity constraints. The upper limit of the access capacity for a single branch line is set as the remaining capacity of that branch line multiplied by a safety margin coefficient of 0.85. The remaining capacity equals the line's rated capacity minus the current load power. The total upper limit of the access capacity is set as the available capacity of the substation where the target line is located. During the evolutionary process, crossover and mutation operations are performed on the current population to generate offspring. The crossover operation randomly selects two parent individuals with a probability of 0.9 to exchange some decision variables to generate two offspring individuals. Mutation... The operation adds random perturbations to the decision variables of offspring individuals with a probability of 0.1, calculates the three objective function values ​​of each individual to form an objective vector, and performs non-dominated sorting on all individuals to identify the Pareto front. Individuals on the Pareto front represent non-dominated solutions that cannot be improved simultaneously on the three objectives. Non-dominated solutions on the Pareto front are retained, while dominated inferior solutions are eliminated. After repeating the evolutionary process for 500 generations, the final Pareto optimal solution set is obtained. The scheme with the best overall performance is selected from the Pareto optimal solution set. The overall performance is calculated by normalizing the three objective function values ​​to the interval between 0 and 1 and then calculating a weighted sum score. The weighted sum score is equal to the normalized voltage quality objective multiplied by 0.4, the normalized line load rate objective multiplied by 0.35, and the normalized line loss objective multiplied by 0.25. The access capacity vector corresponding to the individual with the smallest weighted sum score is selected as the initial access capacity scheme.

[0072] In one specific embodiment, the comprehensive risk value is obtained by multiplying the voltage deviation score, the line load pressure score, and the voltage over-limit frequency score by their respective risk weight coefficients and then summing them.

[0073] Specifically, the comprehensive risk value is obtained by summing the scores for voltage deviation severity, line load pressure, and voltage over-limit frequency, each multiplied by its corresponding risk weight coefficient. The voltage deviation severity score is calculated by dividing the absolute value of the difference between the current real-time voltage and the rated voltage of the main line by the rated voltage to obtain the voltage deviation ratio. This ratio is then mapped to a scoring range of 0 to 1. A score of 0 indicates that the voltage fully complies with the standard, while a score of 1 indicates that the voltage deviation has reached the maximum tolerance level. The line load pressure score... The line load rate is calculated by dividing the current transmission power of the main line by its rated capacity. This load rate is used directly as the score. A load rate of 0 indicates no load, while a load rate of 1 indicates full load. The voltage over-limit frequency score is calculated by dividing the number of voltage over-limit occurrences on the main line within the monitoring period by the total number of samples taken within the period. This frequency is used as the score. A frequency of 0 indicates no over-limit occurrences, while a frequency close to 1 indicates frequent over-limit occurrences. Risk weighting coefficients are set based on the impact of each risk indicator on the safe operation of the distribution network. The risk weighting coefficient for voltage deviation is set at 0.4 because voltage quality directly affects the normal operation of electrical equipment and the stability of the distribution network. The risk weighting coefficient for line load pressure is set at 0.35 because line overload can lead to equipment overheating and insulation aging, shortening equipment lifespan. The risk weighting coefficient for voltage over-limit frequency is set at 0.25 because historical over-limit behavior reflects the vulnerability and potential fault risks of the line. The calculation process for the comprehensive risk value is as follows: multiply the voltage deviation score by a risk weighting coefficient of 0.4 to obtain the voltage deviation risk contribution value; multiply the line load pressure score by a risk weighting coefficient of 0.35 to obtain the load pressure risk contribution value; multiply the voltage over-limit frequency score by a risk weighting coefficient of 0.25 to obtain the over-limit frequency risk contribution value; and sum the three risk contribution values ​​to obtain the comprehensive risk value. The comprehensive risk value ranges from 0 to 1. The closer the value is to 1, the higher the comprehensive operational risk of the main line; the closer the value is to 0, the safer the operation of the main line.

[0074] In one specific embodiment, the multi-objective evolutionary algorithm is a non-dominated sorting genetic algorithm, the decision variable is the access capacity vector of each branch line, and the constraints include the upper limit of the access capacity of a single branch line and the upper limit of the total access capacity.

[0075] Specifically, when the multi-objective evolutionary algorithm is a non-dominated sorting genetic algorithm, it uses the basic framework of a genetic algorithm combined with a non-dominated sorting mechanism to handle the conflict relationships between multiple optimization objectives. When the decision variable is the access capacity vector of each branch line, the target line is divided into several branch lines numbered from 1 to n. Each branch line corresponds to a decision variable representing the access capacity of distributed power sources on that branch line. The n decision variables form the access capacity vector, which constitutes the solution space of the algorithm. Each individual in the population represents a complete access capacity allocation scheme, and the individual's gene encoding is the component of the access capacity vector. When the constraints include the upper limit of the access capacity of a single branch line, for the i-th branch line, the remaining capacity is calculated as equal to the line's rated capacity minus the current load power. The remaining capacity is multiplied by a safety margin coefficient of 0.85 to obtain the upper limit of the access capacity of that branch line. The purpose of setting the safety margin coefficient is to reserve a margin between the access capacity and line safety to avoid line overload. The access capacity decision variable of the i-th branch line must be less than or equal to the upper limit of the access capacity of that branch line, and the access capacity decision variable must be greater than or equal to 0, indicating that negative values ​​are not allowed. When the constraints include the upper limit of the total access capacity, the total access capacity is obtained by summing the access capacity decision variables of all branch lines. The total access capacity is limited by the available capacity of the substation where the target line is located. The available capacity of the substation is equal to the total capacity of the substation minus the currently occupied capacity. The sum of the total access capacity decision variables must be less than or equal to the available capacity of the substation to ensure that the substation will not be overloaded after the distributed power source is connected. During the execution of the non-dominated sorting genetic algorithm, when initializing the population, the access capacity vector of each individual is randomly assigned a value and checked to see if it meets the two constraints: the upper limit of access capacity for a single branch line and the upper limit of total access capacity. If the constraints are not met, the value is randomly reassigned until the constraints are met. The evolutionary operations include selection operations, which select individuals with higher fitness from the current population as parents; crossover operations, which partially exchange the access capacity vectors of two parent individuals to generate offspring individuals; and mutation operations, which add random perturbations to the access capacity vectors of offspring individuals. After the crossover and mutation operations, the offspring individuals are checked to see if they meet the constraints. If the constraints are violated, a repair operation is performed to adjust the out-of-bounds access capacity values ​​to within the constraints. After calculating the three objective function values ​​of each individual in the population, non-dominated sorting is performed. Non-dominated sorting divides individuals into multiple frontier levels. The first frontier contains non-dominated solutions that are not dominated by any other individual, the second frontier contains solutions dominated only by individuals in the first frontier, and so on. Based on the frontier level and crowding distance, excellent individuals are selected to enter the next generation of the population. The evolutionary process is repeated until the maximum number of generations or the convergence condition is reached.

[0076] In one specific embodiment, the voltage quality objective is to minimize the maximum voltage deviation, the line load rate objective is to minimize the maximum line load rate, and the line loss objective is to minimize the total line loss. The initial access capacity scheme is selected from the Pareto optimal solution set based on the normalized comprehensive score.

[0077] Specifically, when the voltage quality objective is to minimize the maximum voltage deviation, for a given access capacity scheme, power flow calculation is performed to obtain the voltage values ​​of all nodes in the distribution network. The power flow calculation solves for the voltage of each node based on the power balance equation and the voltage drop equation. The voltage deviation of each node is calculated as the absolute value of the difference between the node's voltage value and the rated voltage divided by the rated voltage to obtain the normalized voltage deviation ratio. The maximum voltage deviation ratio is found by traversing all nodes and used as the voltage quality objective function value of the access capacity scheme. The maximum voltage deviation reflects the degree of deviation of the node with the worst voltage quality in the distribution network. The optimization objective of minimizing the maximum voltage deviation ensures that the voltage of all nodes is as close as possible to the rated voltage. When the line load rate objective is to minimize the maximum line load rate, for a given access capacity scheme, the transmission power of all lines in the distribution network is obtained through power flow calculation. The load rate of each line is calculated by dividing the transmission power of that line by its rated capacity. The maximum load rate is found by iterating through all lines and used as the line load rate objective function value for this access capacity scheme. The maximum line load rate reflects the load-bearing pressure of the heaviest line in the distribution network. Minimizing the maximum line load rate avoids overloading of individual lines while improving the overall balance of line utilization. When the line loss objective is to minimize the total line loss, for a given access capacity scheme, the power loss of each line is calculated. The line power loss is equal to the square of the current flowing through the line multiplied by the resistance of the line. The current value is calculated by dividing the transmission power of the line by the line voltage. The total line loss is obtained by summing the power losses of all lines in the distribution network and used as the line loss objective function value for this access capacity scheme. Minimizing the total line loss reduces energy loss in the distribution network and improves power transmission efficiency. When selecting an initial access capacity scheme from the Pareto optimal solution set based on the normalized comprehensive score, the Pareto optimal solution set contains multiple non-dominated solutions. Each non-dominated solution corresponds to a set of access capacity vectors and three objective function values. Normalization of the three objective function values ​​maps target values ​​with different dimensions to a unified 0-1 interval. The normalized value of the voltage quality target is equal to the maximum voltage deviation of the scheme divided by the maximum maximum voltage deviation among all schemes. The normalized value of the line load rate target is equal to the maximum line load rate of the scheme divided by the maximum maximum line load rate among all schemes. The normalized value of the line loss target is equal to the total line loss of the scheme divided by the maximum total line loss among all schemes. The comprehensive score is calculated as the normalized voltage quality target multiplied by a weighting factor of 0.4, the normalized line load rate target multiplied by a weighting factor of 0.35, and the normalized line loss target multiplied by a weighting factor of 0.25. The weighting factors reflect the relative importance of each target in practical applications. The scheme with the smallest comprehensive score is found by traversing the Pareto optimal solution set; the access capacity vector corresponding to this scheme is the initial access capacity scheme.

[0078] In one specific embodiment, step S5 includes:

[0079] Substitute the initial access capacity scheme into the digital twin model and set the simulation time span and time step;

[0080] For each time step, a predicted load is generated based on the historical load curve pattern, and a distributed power generation output prediction is generated based on the meteorological data-driven model.

[0081] Monte Carlo simulations were performed on each branch line to count the number of voltage over-limits and the maximum voltage deviation of each branch line, and the risk resistance coefficient was calculated.

[0082] The initial access capacity scheme is modified based on the risk resistance coefficient, and the optimal access capacity scheme is obtained by combining the load threshold and reverse power threshold for constraint verification.

[0083] Specifically, the digital twin model constructs a virtual operating environment for the distribution network based on the topology diagram. The access capacity values ​​of each branch line in the initial access capacity scheme are set as the upper limit of the output of the corresponding distributed generation in the digital twin model. The simulation time span is set to cover a complete monthly operating cycle over the next 30 days, and the time step is set to 15 minutes, dividing the simulation time span into 2880 discrete time steps, each corresponding to an operational snapshot of the distribution network. When generating the predicted load for each time step based on the historical load curve pattern, the average load value for the same period over the past 90 days is extracted from the historical database as the benchmark load. The fluctuation characteristics of the historical load data are analyzed to calculate the standard deviation. A random disturbance following a normal distribution is added to the predicted load for that time step. The mean of the random disturbance is 0, and the standard deviation is 0.05 times the standard deviation of the historical load. The predicted load equals the benchmark load plus the uncertainty fluctuation of the simulated load due to the random disturbance. When generating the distributed generation output prediction based on the meteorological data-driven model, for photovoltaic power, the data for that time step is obtained from the meteorological data interface. The solar irradiance prediction value is used to calculate the output of a photovoltaic power source. The output is calculated by multiplying the photovoltaic rated capacity by the photovoltaic conversion efficiency (0.18), then multiplying by the predicted solar irradiance value, and dividing by the standard irradiance of 1000 watts per square meter. For wind power sources, the wind speed prediction value for that time step is obtained from the meteorological data interface. When the wind speed is less than the cut-in wind speed (3 m / s), the wind power output is 0. When the wind speed is between 3 and 12 m / s, the wind power output is calculated based on the cube of the wind speed. When the wind speed is greater than 12 m / s but less than 25 m / s, the wind power output equals the rated capacity. When the wind speed is greater than 25 m / s, the wind power output is 0 to protect equipment safety.When performing Monte Carlo simulations on each branch line, 1000 simulation experiments were run independently for each branch line. Each simulation experiment randomly sampled the predicted load and distributed generation output at 2880 time steps. Different random disturbances were superimposed on the predicted values ​​to simulate various uncertainties in actual operation. The node voltage of the branch line at each time step was calculated for each simulation experiment. The number of time steps in which the branch line experienced voltage exceedances in a single simulation was counted. The criterion for voltage exceedance was that the deviation of the node voltage from the rated voltage exceeded the allowable deviation range. The number of simulations in 1000 simulations in which voltage exceedances occurred was taken as the number of simulations in which voltage exceedances occurred. The number of times the voltage exceeds the limit for a branch line is recorded. The maximum voltage deviation amplitude of the branch line in 1000 simulations is equal to the maximum voltage deviation in all simulations across all time steps. When calculating the risk resistance coefficient, the first risk coefficient is calculated first, which is equal to the number of times the voltage exceeds the limit divided by 1000, multiplied by the maximum voltage deviation amplitude, and divided by the rated voltage. The service life of the equipment for the branch line is read. When the service life is greater than 15 years, the risk of the equipment entering the aging period increases. The second risk coefficient is equal to 1 plus 0.05 multiplied by the service life minus 15, and the difference is taken as a non-negative number. The risk resistance coefficient is equal to the first risk coefficient plus the second risk coefficient, divided by 2 to obtain the comprehensive risk assessment value. When adjusting the initial access capacity scheme based on the risk resistance coefficient, the adjusted access capacity equals the initial access capacity multiplied by 2 minus the adjustment factor of the risk resistance coefficient. When the risk resistance coefficient is close to 0, the adjustment factor is close to 2, making the adjusted access capacity close to twice the initial access capacity. When the risk resistance coefficient is close to 1, the adjustment factor is close to 1, making the adjusted access capacity close to the initial access capacity. When the risk resistance coefficient is greater than 1, the adjustment factor is less than 1, making the adjusted access capacity less than the initial access capacity, thus reducing the access volume of high-risk lines. When combining load threshold for constraint verification, the load threshold is calculated as the sum of the rated power of all load devices on the branch line multiplied by 0.7. When the adjusted access capacity is less than the load threshold, it is determined that the output of the distributed power source is insufficient to generate reverse power flow, and the adjusted access capacity is directly taken as the optimal access capacity for the branch line. When the modified access capacity is greater than or equal to the load threshold, a reverse power constraint verification is required. When performing constraint verification in conjunction with the reverse power threshold, the reverse power threshold is calculated to be equal to the transformer capacity multiplied by 0.3 to ensure that the reverse power does not exceed 30% of the transformer capacity. The reverse power under the most unfavorable operating condition is calculated to be equal to the modified access capacity minus the historical minimum load. The historical minimum load is taken as the 5th percentile of the historical load data of the branch line. When the reverse power exceeds the reverse power threshold, the optimal access capacity of the branch line is recalculated based on the historical minimum load plus the reverse power threshold. When the reverse power does not exceed the reverse power threshold, the modified access capacity is kept as the optimal access capacity. After all branch lines have completed the modification and constraint verification, the optimal access capacity of each branch line is summarized to form a complete optimal access capacity scheme.

[0084] In one specific embodiment, the calculation of the risk resistance coefficient includes: calculating a first risk coefficient based on the ratio of the number of voltage over-limits to the total number of simulations and the ratio of the maximum voltage deviation to the rated voltage; calculating a second risk coefficient based on the service life of the branch line; and averaging the first risk coefficient and the second risk coefficient to obtain the risk resistance coefficient.

[0085] Specifically, when calculating the first risk coefficient based on the ratio of the number of voltage overruns to the total number of simulations and the ratio of the maximum voltage deviation to the rated voltage, the number of voltage overruns is statistically obtained from 1000 Monte Carlo simulations. The statistical standard is that if at least one voltage overrun occurs in 2880 time steps of a simulation experiment, that simulation is counted as an overrun. Dividing the number of voltage overruns by the total number of simulations (1000) yields the overrun probability, reflecting the likelihood of a voltage overrun occurring on the branch line. The maximum voltage deviation is recorded from all time steps of the 1000 simulations, showing the most severe voltage deviation. Dividing the maximum voltage deviation by the rated voltage yields the normalized deviation ratio, reflecting the severity of the voltage deviation. The first risk coefficient equals the overrun probability multiplied by the deviation ratio. When both the overrun probability and the deviation ratio are large, a higher first risk coefficient indicates that the branch line faces a higher voltage risk after initial capacity is connected. When both the overrun probability and the deviation ratio are small, a lower first risk coefficient indicates that the branch line has a stronger risk resistance capability. When calculating the second risk coefficient based on the service life of a branch line, the commissioning date of the key equipment on that branch line is read from the equipment management system. The service life is calculated by measuring the time span from the commissioning date to the current date. A threshold of 15 years is set as the age at which the equipment enters its aging period. When the service life is less than or equal to 15 years, the equipment is in normal operation and the second risk coefficient is 1, with no additional risk weight added. When the service life is greater than 15 years, the equipment enters its aging period and its reliability decreases. The second risk coefficient is equal to 1 plus 0.05 multiplied by the difference between the service life and 15. For every additional year beyond 15 years, the second risk coefficient increases by 0.05. When the service life is 20 years, the second risk coefficient is equal to 1 plus 0.05 multiplied by 5, which equals 1.25. When the service life is 30 years, the second risk coefficient is equal to 1 plus 0.05 multiplied by 15, which equals 1.75. The second risk coefficient increases linearly with the service life, reflecting the cumulative impact of equipment aging on line risk. When averaging the first risk coefficient and the second risk coefficient to obtain the risk resistance coefficient, the first risk coefficient and the second risk coefficient are added together and then divided by 2 to complete the arithmetic average calculation. The risk resistance coefficient comprehensively considers two dimensions: the risk of voltage exceeding the limit and the risk of equipment aging after the branch line is connected to the initial capacity. The larger the risk resistance coefficient value, the higher the overall risk of the branch line, and the more necessary it is to reduce the access capacity. The smaller the risk resistance coefficient value, the stronger the risk resistance capability of the branch line, and the more permissible it is to increase the access capacity.

[0086] In one specific embodiment, the initial access capacity scheme is modified according to the risk resistance coefficient, including: multiplying the initial access capacity of each branch line by the correction function of the corresponding risk resistance coefficient to obtain the modified access capacity.

[0087] Specifically, when modifying the initial access capacity scheme based on the risk resistance coefficient, the modified access capacity is obtained by multiplying the initial access capacity of each branch line by the corresponding risk resistance coefficient correction function. The correction function is designed as 2 minus the risk resistance coefficient. This function maps the risk resistance coefficient to a correction factor to achieve dynamic adjustment of the access capacity. When the risk resistance coefficient of a branch line is 0, the result of the correction function is 2 minus 0, which equals 2. The modified access capacity equals the initial access capacity multiplied by 2, indicating that the risk of this branch line is extremely low, allowing the access capacity to be expanded to twice the initial capacity to fully utilize the line's carrying capacity potential. When the risk resistance coefficient is 0.5, the result of the correction function is 2 minus 0.5, which equals 1.5. The modified access capacity equals the initial access capacity multiplied by 1.5, indicating that the risk of this branch line is relatively low, allowing for a moderate increase in access capacity. When the risk resistance coefficient is 1, the calculation result of the correction function is 2 minus 1, which equals 1. The corrected access capacity equals the initial access capacity multiplied by 1, keeping the initial capacity unchanged, indicating that the risk of this branch line is in a balanced state. When the risk resistance coefficient is 1.5, the calculation result of the correction function is 2 minus 1.5, which equals 0.5. The corrected access capacity equals the initial access capacity multiplied by 0.5, indicating that the risk of this branch line is high and the access capacity needs to be reduced to half of the initial capacity. When the risk resistance coefficient is 2 or greater, the calculation result of the correction function is 0 or a negative value. In this case, setting the corrected access capacity to 0 indicates that the risk of this branch line is too high and it is not suitable for connecting distributed power sources. The correction function adopts a linear decreasing relationship to ensure that the risk resistance coefficient and the corrected access capacity have an inverse trend. The higher the risk resistance coefficient, the smaller the corresponding correction factor, resulting in a lower corrected access capacity, achieving a dynamic balance between risk and capacity. The lower the risk resistance coefficient, the larger the corresponding correction factor, resulting in a higher corrected access capacity, fully exploring the carrying capacity of low-risk lines. For each branch line in the target line, the correction operation is performed independently. The access capacity value of the branch line in the initial access capacity scheme is read, and the risk resistance coefficient value calculated by the branch line during the simulation is read. The risk resistance coefficient is substituted into the correction function to calculate the correction factor. The initial access capacity is multiplied by the correction factor to obtain the corrected access capacity of the branch line. After all branch lines have completed the correction calculation, a corrected access capacity vector is formed. The corrected access capacity vector is used as the input data for constraint verification and enters the next stage of processing.

[0088] Figure 2 This is a schematic diagram illustrating the relationship between the risk resistance coefficient and the modified access capacity in an embodiment of this application. Figure 2The diagram illustrates the functional relationship between the initial access capacity and the risk resistance coefficient. The horizontal axis represents the risk resistance coefficient, ranging from 0 to 2.5, while the vertical axis represents the modified access capacity in megawatts (MW). The modified access capacity is calculated by multiplying the initial access capacity by 2 and subtracting the correction factor of the risk resistance coefficient. When the risk resistance coefficient is 0, the modified access capacity reaches its maximum value of 10 MW, indicating that the risk of this branch line is extremely low and the access capacity can be expanded. When the risk resistance coefficient is 1, the modified access capacity remains at its initial value of 5 MW, indicating that the risk is balanced. When the risk resistance coefficient is 2 or higher, the modified access capacity drops to 0 MW, indicating that the risk is too high and it is not suitable for connecting distributed power sources. The curves in the diagram show a linear decreasing trend, reflecting the inverse relationship between the risk resistance coefficient and the modified access capacity. This correction mechanism realizes dynamic capacity adjustment based on risk assessment. For low-risk lines, the access capacity is increased to fully tap the carrying potential, while for high-risk lines, the access capacity is reduced to ensure the safe and stable operation of the distribution network.

[0089] In one specific embodiment, the load threshold is a preset ratio of the sum of the rated power of all load devices on the branch line. When the corrected access capacity is less than the load threshold, the corrected access capacity is taken as the optimal access capacity of the branch line.

[0090] Specifically, when the load threshold is a preset proportion of the sum of the rated power of all load devices on a branch line, all load devices connected to each branch line are counted. The rated power parameter of each load device is read from the equipment ledger. Rated power represents the electrical power consumed by the device under normal operating conditions. The rated power of all load devices on the branch line is summed to obtain the total rated power of the load devices. The preset proportion is set to 0.7, which means that the load threshold is 70% of the total rated power of the load devices. The load threshold is equal to the total rated power of the load devices multiplied by 0.7. The preset proportion of 0.7 is selected because load devices will not operate at full load at the same time in actual operation. Under typical conditions, the simultaneity coefficient is between 0.6 and 0.8. Selecting 0.7 as the calculation coefficient for the load threshold takes into account the diversity of loads and retains a certain safety margin. When the corrected access capacity is less than the load threshold, it means that the maximum output of the distributed generation on the branch line is less than the typical power consumption of the local load. In this case, the power generated by the distributed generation is completely absorbed by the local load and no excess power is fed back to the upper-level grid. There is no risk of reverse power flow, and the power flow direction of the distribution network remains the normal direction from the upper-level grid to the load. The corrected access capacity is directly used as the optimal access capacity of the branch line without the need for reverse power constraint verification. The process of determining the optimal access capacity of the branch line is completed, and the next branch line is processed. The branch line is classified by the determination of the load threshold. For low access branch lines with corrected access capacity less than the load threshold, the processing flow is simplified and the calculation efficiency is improved. For high access branch lines with corrected access capacity greater than or equal to the load threshold, reverse power constraint verification is triggered to ensure the safe operation of the distribution network.

[0091] In one specific embodiment, when the corrected access capacity is greater than or equal to the load threshold, the reverse power under the most unfavorable operating condition is calculated. The reverse power is the difference between the corrected access capacity and the historical minimum load. When the reverse power exceeds the reverse power threshold, the optimal access capacity of the branch line is calculated based on the historical minimum load and the reverse power threshold.

[0092] Specifically, when the corrected access capacity is greater than or equal to the load threshold, it indicates that the maximum output of the distributed power source on this branch line may exceed the typical power consumption of the local load, posing a potential risk of reverse power flow. Further calculation of the reverse power under the most unfavorable operating condition is needed for verification. The most unfavorable operating condition is defined as the distributed power source operating at maximum output while the load is at minimum consumption. In this case, the electrical energy generated by the distributed power source significantly exceeds the local load demand, causing excess electrical energy to be fed back to the upstream grid, forming reverse power. Load power measurement records for the past 90 days of this branch line are extracted from the historical operation database. All measurement records are sorted from smallest to largest value, and the load value corresponding to the 5th percentile is taken as the historical minimum load. The 5th percentile indicates that the load is lower than this value only 5% of the time, reflecting the extreme low point of the load. The reverse power equals the corrected access capacity minus the historical minimum load. The corrected access capacity represents the maximum possible output of the distributed power source, and the historical minimum load represents the minimum possible consumption of the local load. The difference between the two is the excess power that cannot be absorbed by the local load under the most unfavorable operating condition and needs to be fed back to the upstream grid. When the reverse power exceeds the reverse power threshold, which is determined based on the capacity of the upstream transformer, the rated capacity of the transformer connected to the branch line is read. The reverse power threshold is equal to the rated capacity of the transformer multiplied by 0.3. The coefficient of 0.3 ensures that the reverse power does not exceed 30% of the transformer capacity. Excessive reverse power can lead to problems such as transformer reverse overload, voltage rise, and malfunction of protection devices. When the calculated reverse power value is greater than the reverse power threshold, it is determined that the corrected access capacity is too high and poses a safety hazard. The optimal access capacity of the branch line is calculated based on the historical minimum load and the reverse power threshold. The optimal access capacity is equal to the historical minimum load plus the reverse power threshold. This calculation method ensures that even under the most unfavorable operating conditions, the reverse power is exactly equal to the reverse power threshold and will not exceed the safety limit. The historical minimum load is absorbed by the local load, and the reverse power threshold is fed back to the upstream power grid without exceeding the reverse carrying capacity of the transformer. Through this calculation method, the corrected access capacity is reduced to the maximum allowable value that meets the reverse power constraint. When the calculated reverse power value is less than or equal to the reverse power threshold, it is determined that the corrected access capacity meets the reverse power constraint condition, and the corrected access capacity is maintained as the optimal access capacity of the branch line without further reduction.

[0093] The control method of the all-in-one intelligent distributed power acquisition and security terminal in the embodiments of this application has been described above. The control system of the all-in-one intelligent distributed power acquisition and security terminal in the embodiments of this application is described below. Please refer to... Figure 3 One embodiment of the control system for the all-in-one intelligent distributed power acquisition and security terminal in this application includes:

[0094] The conversion module is used to convert the distribution network line topology into a graph structure to obtain a topology graph structure.

[0095] The weighting module is used to identify key time nodes in historical operating data, and to weight the historical voltage sequence based on the time-series attention weights calculated from the key time nodes to obtain weighted historical data.

[0096] The input module is used to input the topological graph structure and the weighted historical data into the graph neural network for feature aggregation to obtain node embedding representations;

[0097] The solution module is used to identify the target line based on the comprehensive risk value, construct a multi-objective optimization function based on the node embedding representation, and solve it to obtain the initial access capacity scheme.

[0098] The simulation module is used to substitute the initial access capacity scheme into the digital twin model for simulation, calculate the risk resistance coefficient, and make corrections in combination with the load threshold and reverse power threshold to obtain the optimal access capacity scheme.

[0099] above Figure 3 The control system of the multi-in-one intelligent distributed power acquisition and security terminal in this embodiment of the invention is described in detail from the perspective of modular functional entities. The control device of the multi-in-one intelligent distributed power acquisition and security terminal in this embodiment of the invention is described in detail from the perspective of hardware processing.

[0100] Reference Figure 4 This invention also provides a control device for an all-in-one intelligent distributed power acquisition and security terminal. This control device can be a server, and its internal structure can be as follows: Figure 4 As shown. The control device of this all-in-one intelligent distributed power acquisition and safety terminal can be installed inside a switch cabinet. The control device includes a processor, memory, display screen, input device, network interface, and database connected via a system bus. The processor, designed as a computer, provides computing and control capabilities. The memory of the control device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of the control device stores the data corresponding to this embodiment. The network interface of the control device is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements the above-described method.

[0101] Those skilled in the art will understand that Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the control device of the all-in-one intelligent distributed power acquisition security terminal to which the present invention is applied.

[0102] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when the instructions are executed on a computer, cause the computer to perform the steps of the control method of the all-in-one intelligent distributed power acquisition security terminal.

[0103] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0104] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a control device (which may be a personal computer, server, or network device, etc.) of an all-in-one intelligent distributed power acquisition security terminal to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0105] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A control method for an all-in-one intelligent distributed power acquisition and safety terminal, characterized in that, The method includes: Step S1: Convert the distribution network line topology into a graph structure to obtain the topology graph structure; Step S2: Identify key time nodes in historical operation data, and weight the historical voltage sequence according to the time series attention weights calculated based on the key time nodes to obtain weighted historical data; Step S3: Input the topology graph structure and the weighted historical data into the graph neural network for feature aggregation to obtain the node embedding representation; the node embedding representation includes the location characteristics of the current node in the distribution network topology, the electrical characteristics of the surrounding neighboring nodes, and the weighted historical operating mode; Step S4: Identify the target line based on the comprehensive risk value, construct a multi-objective optimization function based on the node embedding representation, and solve it to obtain the initial access capacity scheme; Step S5: Substitute the initial access capacity scheme into the digital twin model for simulation, perform Monte Carlo simulation on each branch line, count the number of voltage over-limits and the maximum voltage deviation of each branch line, and calculate the risk resistance coefficient; multiply the initial access capacity of each branch line by the correction function of the corresponding risk resistance coefficient to obtain the corrected access capacity, and combine it with the load threshold and reverse power threshold for constraint verification to obtain the optimal access capacity scheme. Step S4 includes: The comprehensive risk value is calculated based on the voltage deviation of the main line, the line load pressure, and the frequency of voltage over-limit. The main routes whose comprehensive risk value is greater than the preset risk threshold are designated as the target routes; The target line is divided into several branch lines, and a multi-objective optimization function containing voltage quality target, line load rate target and line loss target is constructed based on the node embedding representation; The initial access capacity scheme is obtained by solving the multi-objective optimization function using a multi-objective evolutionary algorithm.

2. The control method for the all-in-one intelligent distributed power acquisition and security terminal according to claim 1, characterized in that, Step S1 includes: Read the physical topology data of the distribution network, which includes the coordinates of the starting node of the main line, the location of the access point of the branch line, and the distribution information of the load equipment; Real-time acquisition of operational data from each node of the distribution network, including bus voltage, line current, active power, and reactive power; The physical topology data is mapped to a graph structure, the node set of the graph structure includes bus nodes, branch access point nodes and load point nodes, and the weight of each edge in the edge set of the graph structure includes line resistance, line reactance and line length, thus obtaining the topology graph structure.

3. The control method for the all-in-one intelligent distributed power acquisition and security terminal according to claim 2, characterized in that, Each node in the node set stores the node type code, historical average voltage, historical maximum load power, node geographical location code, and node equipment service life. Each edge in the edge set stores the line impedance modulus, line rated capacity, line current load rate, and line historical fault count.

4. The control method for the all-in-one intelligent distributed power acquisition and security terminal according to claim 1, characterized in that, Step S2 includes: Extract historical operating data sequences within a preset time period, the historical operating data sequences including voltage time sequences of each node and power flow time sequences of each edge; Calculate the daily load peak-valley difference. When the load peak-valley difference exceeds the sum of the products of the historical average peak-valley difference and the standard deviation, mark that day as a load peak day. Historical temperature data is acquired, and when the highest temperature of a day exceeds a first temperature threshold or falls below a second temperature threshold, that day is marked as an extreme weather day. The peak load day and the extreme weather day are identified as the key time nodes. The time-series attention weights are calculated based on the key time nodes, and the historical voltage sequences are weighted to obtain the weighted historical data.

5. The control method for the all-in-one intelligent distributed power acquisition and security terminal according to claim 4, characterized in that, The calculation method for the time-series attention weight is as follows: for peak load days, a first weight increment is added; for extreme weather days, a second weight increment is added; for dates that simultaneously meet the conditions of peak load day and extreme weather day, a third weight increment is added; and for ordinary dates, a baseline weight value is used.

6. The control method for the all-in-one intelligent distributed power acquisition and security terminal according to claim 1, characterized in that, Step S3 includes: The topological graph structure and the weighted historical data are input into a graph neural network model, which contains multiple graph convolutional layers. The initial feature vectors of each node are processed by message passing and feature aggregation through the first layer of graph convolutional layer to obtain the node features of the first layer. The receptive field of each node is expanded layer by layer by subsequent graph convolutional layers, so that the features of each node are fused with the information of its neighboring nodes to obtain multi-layer node features. The multi-layer node features are subjected to global pooling to obtain the node embedding representation.

7. The control method for the all-in-one intelligent distributed power acquisition and security terminal according to claim 6, characterized in that, The message passing process includes: calculating the message passing value based on the node's initial feature vector and edge weights; the feature aggregation process includes: normalizing the message passing values ​​of neighboring nodes according to the node degree and then summing them, and obtaining the aggregated node features through an activation function.

8. The control method for the all-in-one intelligent distributed power acquisition and security terminal according to claim 6, characterized in that, The global pooling process employs a graph attention pooling mechanism, which calculates the attention coefficients of each node using a multilayer perceptron, and then performs a weighted summation of the features of each node based on these attention coefficients to obtain a graph-level feature representation.

9. The control method for the all-in-one intelligent distributed power acquisition and security terminal according to claim 1, characterized in that, The comprehensive risk value is obtained by multiplying the voltage deviation score, line load pressure score, and voltage over-limit frequency score by their respective risk weight coefficients and then summing them.

10. The control method for the all-in-one intelligent distributed power acquisition and security terminal according to claim 1, characterized in that, The multi-objective evolutionary algorithm is a non-dominated sorting genetic algorithm. The decision variable is the access capacity vector of each branch line, and the constraints include the upper limit of the access capacity of a single branch line and the upper limit of the total access capacity.

11. The control method for the multi-functional intelligent distributed power acquisition and security terminal according to claim 10, characterized in that, The voltage quality objective is to minimize the maximum voltage deviation, the line load rate objective is to minimize the maximum line load rate, and the line loss objective is to minimize the total line loss. The initial access capacity scheme is selected from the Pareto optimal solution set based on the normalized comprehensive score.

12. The control method for the all-in-one intelligent distributed power acquisition and security terminal according to claim 1, characterized in that, The calculation of the risk resistance coefficient includes: calculating a first risk coefficient based on the ratio of the number of voltage overruns to the total number of simulations and the ratio of the maximum voltage deviation to the rated voltage; calculating a second risk coefficient based on the service life of the branch line; and averaging the first risk coefficient and the second risk coefficient to obtain the risk resistance coefficient.

13. The control method for the all-in-one intelligent distributed power acquisition and security terminal according to claim 1, characterized in that, The load threshold is a preset ratio of the sum of the rated power of all load devices on the branch line. When the corrected access capacity is less than the load threshold, the corrected access capacity is taken as the optimal access capacity of the branch line.

14. The control method for the all-in-one intelligent distributed power acquisition security terminal according to claim 13, characterized in that, When the corrected access capacity is greater than or equal to the load threshold, the reverse power under the most unfavorable operating condition is calculated. The reverse power is the difference between the corrected access capacity and the historical minimum load. When the reverse power exceeds the reverse power threshold, the optimal access capacity of the branch line is calculated based on the historical minimum load and the reverse power threshold.

15. A control system for an all-in-one intelligent distributed power acquisition and security terminal, used to implement the control method for the all-in-one intelligent distributed power acquisition and security terminal as described in any one of claims 1-14, characterized in that, include: The conversion module is used to convert the distribution network line topology into a graph structure to obtain a topology graph structure. The weighting module is used to identify key time nodes in historical operating data, and to weight the historical voltage sequence based on the time-series attention weights calculated from the key time nodes to obtain weighted historical data. The input module is used to input the topological graph structure and the weighted historical data into the graph neural network for feature aggregation to obtain node embedding representations; The solution module is used to identify the target line based on the comprehensive risk value, construct a multi-objective optimization function based on the node embedding representation, and solve it to obtain the initial access capacity scheme. The simulation module is used to substitute the initial access capacity scheme into the digital twin model for simulation, calculate the risk resistance coefficient, and make corrections in combination with the load threshold and reverse power threshold to obtain the optimal access capacity scheme.

Images