Network topology dynamic optimization method and device for large-scale power supply and distribution equipment networking

Abstract

The application discloses a network topology dynamic optimization method and equipment for large-scale power supply and distribution equipment networking, which comprises the following steps: S101, real-time data acquisition and state sensing; S102, network topology modeling and performance evaluation; S103, dynamic risk assessment and optimization target generation; S104, topology optimization decision based on a feasibility maintaining genetic algorithm; and S105, optimal strategy verification and seamless switching. From the Pareto optimal solution set, an optimal network topology reconstruction scheme is selected according to a preset decision strategy. After the reconstruction scheme is verified at a control level, a device cascade and circuit breaking control instruction sequence is generated, and through a distributed collaborative control mechanism, relevant nodes are guided to complete the disturbance-free switching of the network topology within a predetermined time window. The application can avoid invalid solution generation and significantly improve the convergence efficiency of the large-scale network topology optimization algorithm.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power supply and distribution system intelligentization and networking, and particularly relates to a network topology dynamic optimization method and equipment for large-scale power supply and distribution equipment networking. BACKGROUND

[0002] With the rapid development of Internet of Things, intelligent sensing and communication technology, the intelligentization and networking level of power supply and distribution system is continuously improved. Large-scale power supply and distribution equipment access to the network and realize collaborative operation has become the mainstream trend in the industry, especially in some power supply and distribution systems with a large number of high-capacity impact loads, high power quality and power supply reliability, and complex load types. Building an efficient and reliable power supply and distribution equipment Internet of Things is the key infrastructure to realize intelligent power grid self-healing and optimal operation. Under this background, how to dynamically adjust the network topology structure according to the real-time running state to balance the load, eliminate bottlenecks and improve the overall reliability of the system has become the focus and difficulty of technical research in this field.

[0003] At present, many researches in the industry are committed to realizing the optimal operation of the power supply and distribution network through network reconstruction technology. The usual method first collects the real-time running state data of each node in the network through sensors; then builds a network topology model and calculates its overall performance indicators; then identifies risks according to the performance deviation and generates an optimization target set with maximizing reliability and minimizing worst load rate as the core; finally, the optimization problem is input into an improved multi-objective genetic algorithm model, which explores the solution space through operations such as crossover and mutation to generate a Pareto optimal solution set, and selects the optimal scheme to execute network topology switching. This method applies the powerful global search capability of genetic algorithm to the combinatorial optimization problem of network reconstruction, which theoretically provides a feasible technical path for the dynamic optimization of large-scale networks.

[0004] However, the above-mentioned prior art has a defect in the actual large-scale engineering application. Since the power supply and distribution network must strictly follow the radial operation to avoid ring networks and ensure the selectivity of protection, the feasible topology structure must meet strict physical and safety constraints. The crossover and mutation operations in the standard genetic algorithm are essentially random and blind, and when applied to large-scale networks, they will destroy these constraints with a very high probability, generating a large number of invalid topology schemes. These invalid schemes usually manifest as illegal topology structures with ring networks or islands. Since the algorithm must evaluate the fitness of each individual generated, and evaluating a network topology requires complex power flow calculation and reliability analysis, the computational overhead is extremely large. Therefore, the frequent and large generation of these invalid solutions will cause the valuable computing resources to be wasted unnecessarily, causing the convergence speed of the algorithm to decrease exponentially, and even falling into a stagnant state, unable to calculate a feasible optimization scheme within the time window of the system running state change.

[0005] The consequences caused by the technical problem are direct and serious. This means that when facing large-scale actual networks with hundreds of nodes, the optimization algorithm in the prior art will lose its practical value due to low efficiency, the system cannot respond to load fluctuations and fault risks in time, the dynamic optimization function is useless, and the operation and maintenance personnel are forced to return to a static and conservative operation mode relying on artificial experience, making it difficult to achieve the goal of building an intelligent self-healing power grid. Therefore, there is an urgent need in the art for an innovative method that can effectively avoid the generation of invalid solutions and significantly improve the convergence efficiency of large-scale network topology optimization algorithms to solve this deep technical problem existing in the prior art. SUMMARY

[0006] The large-scale power supply and distribution equipment networking network topology dynamic optimization method, device and storage medium provided by the application can at least solve one of the technical problems in the background art.

[0007] To achieve the above-mentioned purpose, the following technical solutions are adopted in the application:

[0008] A large-scale power supply and distribution equipment networking network topology dynamic optimization method, comprising the following steps,

[0009] Step S101, periodically collecting and uploading the running state data of each node in the network through the sensor nodes and communication modules deployed on the power supply and distribution equipment; the running state data at least includes current load rate, voltage stability index, equipment temperature, communication link delay and packet loss rate;

[0010] Step S102, based on the real-time running state data collected in step S101, a network topology model based on physical connection relationship and electrical coupling relationship is constructed; the current overall performance index of the network topology model is calculated, the overall performance index is a weighted comprehensive value, and the calculation depends on the load balancing degree of each node, the network average communication delay and the key path redundancy;

[0011] Step S103, according to the deviation of the current overall performance index and the preset performance threshold, the potential bottleneck nodes and high-risk links in the network are identified; based on the identification result, the topology optimization target set containing the objective function and the constraint condition is generated, with the joint optimization goal of maximizing the overall reliability of the network and minimizing the worst node load rate;

[0012] Step S104, input the topology optimization target set into an improved multi-objective optimization algorithm model, and perform iterative calculation to generate a Pareto optimal solution set; the improved multi-objective optimization algorithm model is a feasibility-preserved non-dominated sorting genetic algorithm, new network topology structures are explored through feasibility-guided crossover operation and feasibility-guided mutation operation, it is ensured that only feasible solutions meeting the radial operation constraint of the power grid are generated in the evolution process, and the joint optimization target is used as a fitness function to evaluate and screen the feasible solutions;

[0013] Step S105, from the Pareto optimal solution set, an optimal network topology reconfiguration scheme is selected according to a preset decision strategy; after the reconfiguration scheme is verified at the control level, a device cascade and circuit control instruction sequence is generated, and through a distributed collaborative control mechanism, the related nodes are guided to complete the disturbance-free switching of the network topology within a predetermined time window.

[0014] In another aspect, the application also discloses a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to make the processor execute the steps of the above method.

[0015] In another aspect, the application also discloses a computer device, which comprises a memory and a processor, and the memory stores a computer program, and the computer program is executed by the processor to make the processor execute the steps of the above method.

[0016] According to the above technical solution, the network topology dynamic optimization method and device of large-scale power supply and distribution equipment networking can avoid the generation of invalid solutions and significantly improve the convergence efficiency of the large-scale network topology optimization algorithm, and the specific beneficial effects are as follows:

[0017] 1. The application proposes an improved multi-objective genetic algorithm, i.e. a feasibility-preserved non-dominated sorting genetic algorithm, and the core technical means is to design a feasibility-guided crossover operation and a feasibility-guided mutation operation. The crossover operation generates a new solution by identifying common and isolable load areas in the parent chromosomes and exchanging the corresponding power supply tree gene fragments; the mutation operation generates a new solution by performing an atomic operation of "throwing and switching a pair of switches", i.e. after randomly closing a tie-in switch, immediately randomly disconnecting a original line switch in the loop network formed by the tie-in switch. The improvement ensures that each new topology scheme explored by the algorithm in the evolution process strictly meets the hard constraint of radial operation of the power grid, thereby completely avoiding the waste of huge computing resources caused by evaluating invalid solutions, and significantly improving the convergence speed and solution efficiency of the algorithm in large-scale network optimization problems.

[0018] 2, The application proposes a brand-new network topology performance quantitative evaluation and dynamic weight assignment method, the core technical means of which is to build a weighted network model integrating electrical and communication characteristics. The model assigns a dynamic weight value to each edge in the physical adjacency matrix, which is the weighted sum of the load rate difference of the two end nodes, communication link delay and packet loss rate ; and on this basis, three core performance indicators are innovatively defined: load balancing degree , network average communication delay and key path redundancy , which together constitute a vector for comprehensively evaluating the topology state. The method converts the abstract "network health degree" into a calculable and accurate mathematical index, providing reliable and quantifiable input basis for subsequent automated optimization decisions.

[0019] 3, The application proposes an adaptability evaluation acceleration method based on a deep learning agent model, the core technical means of which is to introduce a specially trained deep neural network to replace the computationally expensive traditional power flow calculation and simulation process to quickly predict the performance indicators of the topology scheme represented by the chromosome in the genetic algorithm. The network receives the binary coded chromosome, outputs the objective function prediction value through the structure containing the residual block , and adopts the transfer learning mechanism to pre-train the base model using historical data, and then fine-tune the incremental weights online through the current network data , so that the model quickly adapts to the current network state. The method converts the time-consuming high-fidelity simulation evaluation into millisecond-level network forward propagation, greatly reducing the computational cost of the adaptability evaluation link in the genetic algorithm, which is the key to realizing real-time dynamic optimization.

[0020] 4, The application proposes a full-closed-loop automatic optimization method from perception to execution, the overall technical means of which is to build a complete technical chain integrating real-time data acquisition, topology performance evaluation, dynamic risk identification, improved algorithm optimization, intelligent decision-making and automatic control. The method starts with the fusion perception of multi-source heterogeneous operating state data, forms the optimization target through quantitative modeling and risk evaluation, efficiently searches for the Pareto optimal solution set through the innovative feasibility preservation algorithm, and finally realizes the seamless issuance and execution of the optimal strategy relying on the software-defined network technology. The overall method realizes the leap from "monitoring" to "autonomy" of large-scale power supply and distribution networks, and can continuously and dynamically self-optimize its topology structure without human intervention, ultimately achieving the multi-objective comprehensive optimization of improving power supply reliability, balancing load and ensuring communication quality. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 is the flowchart of the application;

[0022] Figure 2This is a schematic diagram of the core operation of the feasibility-preserving genetic algorithm of the present invention;

[0023] Figure 3 This is a schematic diagram of the network topology performance quantification evaluation model of the present invention;

[0024] Figure 4 This is a schematic diagram illustrating the selection of Pareto optimal solutions in the multi-objective optimization decision-making process of this invention;

[0025] Figure 5 This is a schematic diagram of the abnormal data repair and seamless switching control of the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0027] like Figure 1 As shown in this embodiment, the network topology dynamic optimization method for large-scale power supply and distribution equipment networking includes the following steps:

[0028] Step S101, Real-time Data Acquisition and Status Awareness: Through sensor nodes and communication modules deployed on power supply and distribution equipment, the operating status data of each node in the network is periodically collected and uploaded; the operating status data includes at least current load rate, voltage stability index, equipment temperature, communication link delay and packet loss rate;

[0029] Step S102, Network Topology Modeling and Performance Evaluation: Based on the real-time operating status data collected in step S101, a network topology model based on physical connection relationships and electrical coupling relationships is constructed; the current overall performance index of the network topology model is calculated. The overall performance index is a weighted comprehensive value, and its calculation depends on the load balancing degree of each node, the average network communication latency, and the critical path redundancy.

[0030] Step S103, Dynamic Risk Assessment and Optimization Target Generation: Based on the deviation between the current overall performance index and the preset performance threshold, identify potential bottleneck nodes and high-risk links in the network; based on the identification results, generate a topology optimization target set containing objective functions and constraints with the joint optimization objective of maximizing the overall reliability of the network and minimizing the worst node load rate;

[0031] Step S104, topology optimization decision based on feasibility-preservation genetic algorithm: input the topology optimization target set into an improved multi-objective optimization algorithm model, and perform iterative calculation to generate a Pareto optimal solution set; the improved multi-objective optimization algorithm model is a feasibility-preservation non-dominated sorting genetic algorithm, which explores new network topology structures through feasibility-guided crossover operation and feasibility-guided mutation operation, ensures that only feasible solutions satisfying the radial operation constraint of the power grid are generated in the evolution process, and evaluates and screens the feasible solutions by taking the joint optimization target as a fitness function;

[0032] Step S105, optimal strategy verification and seamless switching: from the Pareto optimal solution set, an optimal network topology reconstruction scheme is selected according to a preset decision strategy; after the reconstruction scheme is verified at the control level, a device cascade and circuit control instruction sequence is generated, and through a distributed collaborative control mechanism, the relevant nodes are guided to complete the disturbance-free switching of the network topology within a predetermined time window.

[0033] The following is a specific description:

[0034] As shown in FIG. 1, in the network topology modeling and performance evaluation step S102, the system first receives the real-time running state data set Figure 3 from step S101, where represents the current load rate of network node i, represents the voltage stability index of node i, represents the device temperature of node i, represents the communication link delay of node pair , represents the communication link packet loss rate of node pair , is a set of network nodes, and the number of elements in the set is N; the system constructs a physical adjacency matrix according to the physical connection relationship between nodes, and constructs a communication adjacency matrix according to the communication link between nodes, thereby forming the basic structure of the network topology.

[0035] In order to quantify the influence of the load and communication state between nodes on the network performance, the system assigns a dynamic weight value to each edge in the physical adjacency matrix, where represents the difference between the current load rates of node i and node j, reflecting the degree of load imbalance, and are the delay and packet loss rate ​The normalized value is used to eliminate the influence of different units on the weights. α, β, and γ are preset weighting coefficients that adjust the contributions of load difference, latency, and packet loss to the edge weights, respectively. Based on the weighted adjacency matrix, the system calculates the network load balancing degree. ,in The average load factor of network nodes is used to quantify the dispersion of load distribution; the greater the dispersion, the more unbalanced the load. The average network communication latency is determined by... Calculation, where The total delay on the shortest path between node pairs (i, j) is calculated using a weighted adjacency matrix to solve for the shortest path, and the delay is then quantified as an overall network communication performance indicator. Critical path redundancy is determined through... Calculate, where M is the number of critical paths in the network. Let K be the set of backup links on the k-th critical path. This is the available capacity metric for link l, representing the redundancy capability of the critical path in the event of a single point of failure. Higher redundancy indicates stronger network resilience. Finally, this step will adjust the load balancing... Average network communication latency With critical path redundancy Combined into a performance index vector The results are then output to the subsequent topology optimization process for risk assessment and optimization target generation, ensuring balanced network load distribution, good communication performance, and sufficient critical path redundancy during the optimization process.

[0036] Step S103 Dynamic Risk Assessment and Optimization Target Generation: Taking the real-time operating performance of the system as input, it receives the overall performance index set transmitted by the monitoring module. ,in Indicates network reliability metrics, This indicates the worst node load rate in the entire network. Indicates the node voltage deviation. This represents the system's active power loss. The system first calculates the performance deviation vector. ,in A preset set of performance thresholds is used, with elements corresponding to the lower reliability limit, upper load rate limit, upper voltage deviation limit, and loss threshold, respectively. By analyzing the sign and magnitude of the deviations, critical nodes and high-risk links with performance degradation are identified. For each link... With nodes Calculate risk indicators ;in For link load rate, For voltage deviation, The risk index is a local reliability factor, where α1, α2, and α3 are weighting coefficients determined based on historical sensitivity analysis. Higher risk index values ​​indicate higher potential failure and congestion risks. The system uses the risk index distribution as a basis to filter out systems with risk levels exceeding a threshold. bottleneck node set With high-risk link set Based on this, a topology optimization objective set is constructed. ,in The objective function representing network reliability is... The objective function and constraint function represent the worst-case node load rate. It includes radial constraints, node voltage constraints, and line capacity constraints. The system calculates the feasible region of the current topology and forms a constraint set by fusing multi-source monitoring data, providing an optimization search space for subsequent feasibility-preserving genetic algorithms. The process of generating the optimization target set realizes joint modeling of risk and performance, enabling subsequent topology optimization to dynamically reduce the operating load in high-risk areas while maintaining power supply reliability and safety margin, forming an adaptive optimization input oriented to the real-time operating status, and outputting to the topology optimization decision module in step S104.

[0037] like Figure 2 As shown, in the topology optimization decision step of the feasibility-preserving genetic algorithm in step S104, the system receives the topology optimization target set O generated in step S103. ,in The function representing the overall network reliability index. The objective function represents the load rate of the worst-performing node in the network. The network operation constraints include radial power supply topology constraints, node capacity constraints, and line flow constraints. x is a decision vector representing a combination of switching states. Each chromosome is represented using binary encoding. ,in The m-th switch in the network corresponds to the open / closed state, where 1 represents closed power supply and 0 represents open power supply. During the initialization phase, Prim's minimum spanning tree algorithm is used to generate a population of feasible solutions satisfying a radial structure under the constraints of the physical adjacency matrix. ,in For population size, each individual Both satisfy the basic feasibility of the power grid topology. Crossover operations are implemented through subtree swapping, and the system starts from the parent chromosome. Identifying shared load areas Determine the switch index set corresponding to the subtree. Swap parent and child tree segments. To generate new candidate topologies while maintaining power supply radiation structure constraints. The mutation operation employs a switching pair strategy, where the system randomly selects a disconnected tie switch. is set to 1, the formed loop network is randomly selected from the original closed line switch is set to 0, ensuring the balance of loop closure and radial constraint conditions. In the fitness evaluation stage, the objective function vector is calculated for each individual , where represents the higher the network reliability, is used to convert the minimum worst load rate target into a co-directional optimization indicator with reliability, facilitating non-dominated sorting operations. The system selects the next generation population through non-dominated sorting and congestion calculation methods, retaining individuals that are optimal and evenly distributed in the Pareto sense. After iterative evolution, the algorithm outputs the Pareto optimal solution set , where represents the individual The solution set contains network topology schemes that achieve optimal compromise between reliability and worst load rate, providing a decision basis for subsequent optimal strategy verification and seamless switching.

[0038] In the subtree exchange operation, the system first determines the gene segment corresponding to the donor subtree , where is the set of switch indices from the power node to the complete feeder path in the load area , containing all switch state variables along the path. This segment exchange operation exchanges the corresponding gene segments in the parent chromosome, generating two new child chromosomes while ensuring that the topology structure maintains the integrity of the radial constraint, i.e., each load area is still directly connected to the power node through a unique path, avoiding the formation of redundant loops or isolated nodes. After the exchange is completed, the system confirms that each child chromosome is connected to the power node through topology connectivity checks, ensuring the integrity of the power supply path. Subsequently, the system performs power flow calculation on each child to verify the satisfaction of electrical constraints, including node voltage within the allowed range, line flow not exceeding the rated capacity, and key node load meeting design requirements. For individuals with voltage out-of-limit or line overload in the power flow calculation, the system punishes their fitness, reducing their objective function value and lowering their selection probability in the next generation, ensuring that feasible solutions that meet electrical constraints are prioritized in the evolution process, thereby optimizing network reliability and load balancing while maintaining the radial constraint of the topology.

[0039] In the switching pair mutation operation, the system first identifies the formed loop network C to ensure that the radial constraint of the network is maintained during the mutation process. Loop network identification uses a depth-first search algorithm, starting from the closed tie switch endpoints a, b to traverse the network connected branches, tracking node connection relationships along the path to obtain the set of all switch state variables in the loop network , where This represents the set of all switch indices involved on the path from node a to node b. For the closed tie switch itself, the entire set C represents the complete switching loop formed by the ring network. The mutation operation involves randomly selecting and disconnecting the tie switch. Set to closed circuit 1, and simultaneously randomly select an existing closed circuit switch in ring network C. Set to disconnect (0) to maintain network connectivity and readjust power supply paths. After mutation, the system first verifies whether the offspring chromosome still satisfies the radial topology constraint, ensuring that each load node is connected to the power node through only one path. Next, it performs equipment capacity verification and power flow calculation checks on the lines. If a line exceeds its rated capacity or experiences overload, the fitness of that offspring individual is downgraded, reducing its probability of being selected in subsequent non-dominated sorting selections. This ensures that the evolutionary process prioritizes feasible solutions that satisfy both radial and capacity constraints, achieving joint optimization of network reliability and load balancing.

[0040] like Figure 4 As shown, in the decision-making strategy, the system adopts a comprehensive evaluation model combining the entropy weight method and TOPSIS to evaluate the Pareto optimal solution set. The process involves sorting and selecting solutions, where n is the number of Pareto solutions and m is the number of optimization objectives. The objective function values ​​of each solution form a matrix. ,in This represents the evaluation value of the i-th solution on the j-th objective. The system first calculates the entropy value for each objective. ;in ; is the normalized contribution of the i-th solution to the j-th objective, and the entropy value. This reflects the dispersion of target information; the greater the dispersion, the more significant the target's impact on the overall evaluation. The weights of each target are calculated based on the entropy value. Targets with higher weights have a greater proportion in the final comprehensive evaluation. Subsequently, the system constructs a weighted normalized matrix. This is used to eliminate dimensional differences and introduce target weights, ensuring comparability of different targets in comprehensive evaluation. The ideal solution is determined based on the weighted normalization matrix. With negative ideal solution These represent the optimal and worst values ​​for each objective, respectively, and are used to calculate the Euclidean distance from each solution to the ideal and negative ideal solutions. ,in The smaller the value, the closer the solution is to the ideal solution. A larger value indicates that the solution is farthest from the negative ideal solution. The relative proximity of each solution is calculated based on the distance. This index comprehensively considers both the degree of closeness of the solution to the ideal solution and the degree of distance from the negative ideal solution; a higher value indicates better overall performance. The system selects the relative closeness... The largest solution is the final implementation plan. ; the scheme takes into account network reliability, load balancing and other optimization goals, realizes the determination of the optimal compromise solution in the Pareto solution set, and provides decision basis for seamless switching of network topology.

[0041] As shown in Figure 5 , in the data repair processing of step S105, the system first receives the historical running state data sequence and the abnormal data of the current time step , wherein represents the historical running state of each node in the network. The system expands the historical data into a sequence input within a time window τ, wherein each is a flattened state vector containing the load rate , voltage index , device temperature , link delay and packet loss rate of all nodes, which are used to capture the timing pattern of network operation. The system trains and predicts the historical sequence through a long short-term memory network (LSTM), learns the dynamic change rule under normal operation, and outputs a predicted value which represents a reasonable estimate of the state of each node at the current time step. To repair abnormal data, the system weights and fuses the original observation and the prediction result according to the confidence coefficient of anomaly detection , wherein a larger λ indicates a higher degree of trust in abnormal observation values, and a smaller λ indicates a greater reliance on predicted values to repair abnormalities. The fused repair data maintains temporal consistency and physical rationality. Finally, the system outputs complete and repaired running state data to the network topology modeling and performance evaluation step, providing accurate and reliable input for subsequent topology optimization and ensuring that optimization decisions can still be made stably under the interference of abnormal data.

[0042] In the seamless switching control of step S105, the system realizes precise execution of network topology reconstruction through the SDN controller, receives the final reconstruction scheme and the corresponding switch action instruction set , wherein S is a set of programmable circuit breakers in the network, represents a closed switch, represents an open switch. The system maps each switch action to a flow table rule sequence , wherein each rule contains a matching field for locating the corresponding switch and port, and an action field for specifying switch actions and execution timestamps, ensuring that actions are implemented in the predetermined order. The SDN controller issues rules to the corresponding switches in time slots through the OpenFlow protocol, enabling synchronized issuance and controllable execution of instructions. The execution sequence guarantee strategy sets the operation interval to ensure that the electrical operation safety interval requirement is met between consecutive switch actions, preventing current surges and voltage fluctuations from affecting network stability. During rule execution, the switch applies the action field in timestamp order to perform real-time breaker closing or opening operations. The system simultaneously collects execution status and outputs confirmation signals to the monitoring system to verify whether the topology switching is successful, ensuring load balancing, communication connectivity, and continuity of electrical constraints during network reconstruction, providing safe and reliable closed-loop control protection for dynamic optimization of large-scale power supply and distribution equipment.

[0043] In the multi-objective optimization algorithm model, the system introduces a deep neural network to assist in training to accelerate the evaluation of chromosome fitness by the genetic algorithm. The input layer of the network structure receives chromosome encoding , where M is the total number of controllable switches in the network, and each element corresponds to the opening or closing state of a switch. The hidden layer is composed of fully connected layers and residual blocks, which capture the complex mapping relationship between chromosome encoding and network performance objectives through nonlinear transformation, and the introduction of residual blocks alleviates the gradient vanishing problem in deep network training. The output layer generates the predicted value of the objective function , where correspond to the predicted values of network reliability and worst node load rate , respectively. The network training data comes from historical optimization process records , where is the true value of the objective function corresponding to the ith chromosome. The training process optimizes network parameters through a weighted mean squared error loss function: , where the weight is the inverse of the variance of each objective function, used to balance the influence of different objectives in the training process, is the historical variance of the objective . Through training, the network can quickly predict the objective function value of any chromosome x for fitness evaluation by the genetic algorithm, avoiding complete power flow calculation or complex network index calculation each time, thereby significantly accelerating algorithm convergence and improving the computational efficiency and real-time performance of multi-objective topology optimization, while ensuring that the predicted results are consistent with the true objective function values, providing a reliable reference for subsequent Pareto optimal solution generation.

[0044] During neural network training, the system uses a transfer learning mechanism, using historical network topology data as the source domain and current network topology data As the target domain, the adaptability of the prediction model to the new topology environment is improved. A domain adaptation layer is introduced in the deep neural network to align the feature distribution of the source domain and the target domain by minimizing the maximum mean difference , wherein is the feature mapping function mapped to the high-dimensional reproducing kernel Hilbert space H, and are the number of source domain and target domain samples respectively, and the source domain pre-training model features and target domain features are made as consistent as possible by minimizing MMD, thereby reducing the domain bias. The network weight is dynamically adjusted by the incremental update mechanism , wherein the base weight comes from the source domain pre-training model, providing global feature learning of historical network topology, and the incremental weight is obtained by online fine-tuning of the target domain data, used to capture the specific changes and dynamic characteristics of the current network topology. Through this mechanism, the system generates a prediction model that adapts to the current network environment can evaluate the objective function value and network performance indicators of the chromosome in real time, which is used to accelerate the fitness calculation of the multi-objective optimization algorithm, improve the efficiency of Pareto solution search, and ensure the prediction accuracy when the topology structure changes or new nodes are accessed, providing reliable real-time support for dynamic optimization and seamless switching of large-scale power supply and distribution equipment networking.

[0045] In summary, the network topology dynamic optimization method for large-scale power supply and distribution equipment networking of the embodiment of the application has the following beneficial effects:

[0046] (1) A feasible genetic algorithm is proposed to ensure that all topology schemes generated during the optimization process meet the radial operation constraints of the power grid, avoiding the consumption of invalid computing resources; (2) A dynamic weight assignment and multi-dimensional performance index evaluation system is designed to realize accurate quantitative perception of the network electrical and communication state; (3) The switching and subtree exchange operations are used to maintain the diversity and search ability of the genetic algorithm solution while ensuring feasibility; (4) A deep learning-based surrogate model is introduced for fitness evaluation, which converts complex power flow calculation into neural network forward propagation, greatly improving the optimization decision speed; (5) A full-closed-loop automation system from data acquisition to control execution is constructed to realize online autonomous optimization and seamless switching of network topology;

[0047] To be specific, the algorithm level of the embodiments of the present application can replace the original feasibility-preserving non-dominated sorting genetic algorithm with an improved particle swarm optimization algorithm adapted to radial constraints, through the design of a "radial-preserving particle update rule", that is, the particle moves only within the solution space that meets the uniqueness of the power supply path, and simultaneously eliminates the particle positions corresponding to the ring network or island, to equally avoid invalid solutions; in the data anomaly repair link, the original LSTM network scheme can be replaced by a hybrid repair method combining Kalman filtering with physical rules of the power supply system, through real-time state prediction and physical constraint correction of abnormal data to reduce the dependence on the amount of historical data; in the dynamic weight assignment of performance evaluation, the preset weighting coefficients alpha, beta and gamma can be replaced by an adaptive weight mechanism, which dynamically adjusts the contribution of each index according to the real-time network bottleneck, to enhance the real-time adaptability of performance evaluation; in the control execution level, the original SDN centralized controller scheme can be replaced by a distributed control architecture of edge computing nodes, which processes the topology optimization decision and switch instruction issuance by the edge nodes close to the power supply equipment, reduces the data transmission delay of the central node, and is particularly suitable for large-scale device networking scenarios; in the proxy model of fitness evaluation, the original deep neural network scheme can be replaced by a gradient boosting tree model, which has the characteristics of high small-sample training efficiency and fast inference speed, and can realize fast prediction of the objective function in scenarios with frequent network state fluctuations but limited historical samples, to ensure the real-time performance of the optimization decision.

[0048] The technical key points and protection points of the present application are that, in view of the core defects of the traditional multi-objective genetic algorithm in the prior art, such as the generation of invalid solutions of ring network / island, the waste of computing resources and slow convergence when applied to large-scale power supply and distribution network topology optimization, a feasibility-preserving non-dominated sorting genetic algorithm is innovatively designed, through "subtree exchange crossover operation" and "switching pair mutation operation", to ensure that only feasible solutions meeting the radial constraints are generated in the evolution process from the operation mechanism, to completely avoid the evaluation cost of invalid solutions; unlike the existing evaluation method that only focuses on electrical characteristics, a dynamic weighted performance evaluation system integrating electrical and communication is constructed, through dynamic weight The fusion node load rate difference, normalized communication delay and packet loss rate are differentiated, and three-dimensional core indexes of load balancing degree, network average communication delay and critical path redundancy are defined to realize comprehensive quantification of the network state; an adaptability evaluation acceleration mechanism based on deep learning is introduced, a deep neural network containing a residual block is used to replace traditional complex power flow calculation, and combined with transfer learning, the evaluation time is reduced to milliseconds; a full closed-loop automatic optimization process from perception to execution is created, through multi-source data acquisition, dynamic risk identification, improved algorithm optimization, and then relying on the SDN controller to generate flow table rules and issue switch instructions according to the safety interval, the topology is switched without disturbance, and an LSTM network is used to weight and integrate abnormal data to ensure input reliability, and the above technical design forms a complete innovative system covering algorithm mechanism, evaluation dimension, calculation efficiency and control closed loop, which is significantly different from the prior art and builds the core protection boundary.

[0049] Step 1: S101 real-time data acquisition and state perception

[0050] Acquisition device: Hall current sensor is deployed on each node to acquire current load rate, PT voltage sensor to acquire voltage stability index, and thermocouple temperature sensor to acquire equipment temperature, and industrial Ethernet module is deployed on each link to acquire communication delay / packet loss rate, with a collection period of 10s.

[0051] The initial collected data are shown in Table 1 (part of key nodes / links):

[0052] Table 1

[0053]

[0054] Step 2: S102 network topology modeling and performance evaluation

[0055] Modeling parameters:

[0056] Constructing physical adjacency matrix (marked nodes 23 and 56 are physically connected through link 18-19), and communication adjacency matrix (marked link 18-19 communication state);

[0057] Dynamic weight Calculation: take α=0.4 (load difference weight), β=0.3 (delay weight), and γ=0.3 (packet loss rate weight), and the of link 18-19 is

[0058] Performance index calculation:

[0059] Load balancing degree , wherein the average load rate is calculated as , the dispersion is high, and the load is unbalanced;

[0060] Average communication delay , exceeds the preset 50ms threshold;

[0061] Critical path redundancy , 8 critical paths, each with spare link capacity Take 0-1, calculated , weak fault tolerance.

[0062] Step 3: S103 dynamic risk assessment and optimization target generation

[0063] Risk identification: According to the results of S102, node 23, load rate 88% close to the upper limit of capacity is the bottleneck node, link 18-19 / 25-26, delay exceeds the threshold, packet loss rate is high, is a high-risk link;

[0064] Optimization target set: O={maxf1(x) (network reliability), minf2(x) (worst node load rate) | g1(x)≤0 (radial constraint), g2(x)≤90% (node capacity constraint)}, where, Quantified by "critical link fault load recovery rate", initial , node 23 load rate.

[0065] Step 4: S104 genetic algorithm optimization of feasibility maintenance type

[0066] Algorithm parameters:

[0067] Chromosome encoding: 40-bit binary (corresponding to 40 switches, 1=close, 0=off), population size 250;

[0068] Initialization: generate initial population satisfying radial with Prim algorithm , acyclic network / island;

[0069] Cross operation: identify common load area , load nodes 61-75, exchange parent And The corresponding switch index set Gene fragment, generate offspring;

[0070] Mutation operation: randomly select off tie-in switch , closed, use depth-first search to identify ring network , off To maintain radial;

[0071] Fitness evaluation: use deep neural network to predict , output Pareto optimal solution set after 200 iterations .

[0072] Optimization result: select the optimal scheme with entropy weight-TOPSIS, that is, select the maximum At this time: Reliability is improved, The node 23 load rate is reduced, , , .

[0073] Step 5: S105 optimal strategy verification and seamless switching

[0074] Verification and execution:

[0075] Control layer verification: the optimal scheme is subjected to power flow calculation, the node voltage is 0.94-1.02 p.u., which meets the requirements, and the line flow is less than or equal to 115%, which is not overrated;

[0076] Seamless switching: the SDN controller generates a switch action instruction set There are 12 instructions, which are mapped to flow table rules R={r1(match x 19 Port, action open, timestamp t1), (r2(match x 28 Port, action close, timestamp t1+50ms)),...};

[0077] Switching result: total time length 520ms, voltage fluctuation ±1.5% (meets the requirements of GB / T12325), no current impact, and load interruption 0 times.

[0078] III. Beneficial effect verification as shown in Table 2;

[0079] Table 2

[0080]

[0081] In another aspect, the application also discloses a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to make the processor execute steps of the above method.

[0082] In still another aspect, the application also discloses a computer device, which comprises a memory and a processor, and the memory stores a computer program, and the computer program is executed by the processor to make the processor execute steps of the above method.

[0083] It can be understood that the system, device and storage medium provided by the embodiments of the application correspond to the method provided by the embodiments of the application, and the explanation, examples and beneficial effects of related contents can refer to the corresponding part in the above method.

[0084] In the embodiments described above, all or some of the steps can be implemented by hardware, software, firmware or any combination thereof. When implemented by software, all or some of the steps can be implemented in the form of one or more computer programs or program elements. The computer programs reside (at least temporarily) in a memory of a computer during execution. The memory can be a RAM memory, a flash memory, a ROM memory, an EPROM memory, or any other suitable memory. The memory can be integral to or separate from the computer. The computer programs can be written in any suitable programming language, such as C, C++, Java, Visual Basic, etc. The computer programs can be written in assembly or machine language, if desired. The computer programs can be distributed over network coupled file servers, or can be distributed by any other suitable means.

[0085] It is to be understood that the terminology "first", "second", etc. used herein merely for the purpose of distinguishing one entity or action from another, and does not necessarily imply these entities or actions are mutually exclusive or are either temporal or spatial priorities of one another. Moreover, the terms "comprising", "including", or any other variation thereof, are intended to cover a non-exclusive inclusion, such that processes, methods, articles, or apparatuses that comprise a list of elements are not necessarily limited to those elements, but can include other elements not expressly listed or inherent to such processes, methods, articles, or apparatuses. Without more specificity, no element is implied to be essential, unless explicitly required.

[0086] Each of the embodiments described in the specification adopt a relevant manner for description, and the same or similar parts among the embodiments can be referred to each other. Each embodiment focuses on the difference from other embodiments. In particular, for the system embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the part of the method embodiments.

[0087] The above examples are only used to illustrate the technical solutions of the present application, and are not intended to limit the present application; although the present application has been described in detail with reference to the foregoing examples, those skilled in the art should understand that the technical solutions recorded in the foregoing examples can be modified, or some technical features can be replaced by equivalent features; and these modifications or replacements 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 application.

Claims

1. A network topology dynamic optimization method for large-scale power supply and distribution equipment networking, characterized in that, The method comprises the following steps of: S101, collecting and uploading the operation data of each node in the network, such as the current load rate, voltage stability index, device temperature, communication link delay and packet loss rate, through the sensor node and the communication module on the power supply and distribution equipment; S102, constructing a network topology model based on the data collected in S101; calculating the overall performance index of the current network, which depends on the load balancing degree of each node, the average communication delay of the network and the redundancy of the key path; S103, identifying the potential bottleneck nodes and high-risk links in the network and optimizing the target according to the deviation of the current overall performance index from the preset performance threshold; S104, inputting the topology optimization target set into an improved multi-objective optimization algorithm model to generate a Pareto optimal solution set; the improved multi-objective optimization algorithm model is a feasibility-preserving non-dominated sorting genetic algorithm, which explores new network topology structures through feasibility-guided crossover operation and mutation operation to ensure that only feasible solutions that meet the radial operation constraints of the power grid are generated, and the feasible solutions are screened according to the joint optimization target as the fitness function; S105, selecting an optimal network topology reconstruction scheme from the Pareto optimal solution set according to a preset decision strategy; after verifying the reconstruction scheme at the control level, generating a device cascade and circuit control instruction sequence, and guiding the relevant nodes to complete the disturbance-free switching of the network topology within a predetermined time window through a distributed collaborative control mechanism; Step S104 includes that the system receives the topology optimization target set generated from step S103 wherein represents a network overall reliability index function, represents a network worst node load rate target function, is a network operation constraint condition, including a power supply radial topology constraint, a node capacity constraint and a line flow constraint, and x is a decision vector representing a switch state combination; Each chromosome is represented using binary encoding wherein corresponding to the on-off state of the mth switch in the network, 1 represents closing the power supply, and 0 represents opening. The initialization stage generates a feasible solution population satisfying the radial structure under the constraint of the physical adjacency matrix by a Prim minimum spanning tree algorithm wherein is the population size, each individual satisfies the basic feasibility of the power grid topology; the crossover operation is implemented by exchanging sub-trees, the system identifies the common load area from the parent chromosomes , determines the switch index set corresponding to the sub-trees , and exchanges the parent sub-tree segments to generate a new candidate topology while maintaining the power supply radial structure constraint; The mutation operation employs a switching strategy, where the system randomly selects a disconnected handshake switch. Set to closed loop 1, in the formed ring network Randomly select the existing closed circuit switch Set to open (0) to ensure the balance between loop closure and radiation constraint conditions; In the fitness evaluation phase, the objective function vector is calculated for each individual wherein The greater the value, the higher the network reliability, for converting the minimization of the worst load rate objective into a co-directional optimization indicator with reliability, facilitating the non-dominated sorting operation; The system filters the next generation population through non-dominated sorting and crowding calculation method, retains the individuals optimal in Pareto sense and uniformly distributed, and outputs the Pareto optimal solution set after iterative evolution wherein represents the individual The solution set contains the network topology scheme optimal in compromise between reliability and worst load rate, and provides a decision basis for subsequent optimal strategy verification and seamless handover.

2. The method for network topology dynamic optimization of large-scale power supply and distribution equipment networking according to claim 1, characterized in that: The step S102 includes that the system first receives the real-time running state data set from the step S101 wherein represents the current load rate of the network node i, represents the voltage stability index of the node i, represents the device temperature of the node i, represents the communication link delay of the node pair , represents the communication link packet loss rate of the node pair , is a network node set, the number of elements in the set is N; The system constructs a physical adjacency matrix according to the physical connection relationship between nodes , constructs a communication adjacency matrix according to the communication link between nodes , and forms the infrastructure of the network topology In order to quantify the influence of the load and communication state between nodes on the network performance, the system assigns a dynamic weight value to each edge in the physical adjacency matrix: wherein, represents the difference of the current load rate between node i and node j, reflecting the degree of load imbalance, and are the normalized values of the delay and the packet loss rate respectively, used to eliminate the influence of different dimensions on the weight, and α, β, γ are preset weighting coefficients, respectively adjusting the contribution of load difference, delay and packet loss to the edge weight; Based on the weighted adjacency matrix, the system calculates the network load balancing degree Wherein The network node average load rate is used to quantify the discrete degree of load distribution, and the greater the discrete degree, the more unbalanced the load is. The network average communication delay is given by where is the total delay on the shortest path between node pair (i, j), and the shortest path is solved by the weighted adjacency matrix. The delay concentration is measured as an overall network communication performance indicator. Key path redundancy is calculated by where M is the number of key paths in the network, is the set of backup links on the kth key path, is the available capacity indicator of link / , which represents the redundancy capability of the key path when a single point of failure occurs. The higher the redundancy, the stronger the network's fault tolerance. Finally, the load balancing degree , the network average communication delay and the critical path redundancy are combined into a performance index vector and output to the subsequent topology optimization process for risk assessment and optimization target generation, ensuring that the network load distribution is balanced, the communication performance is good, and there is sufficient critical path redundancy guarantee during the optimization process.

3. The method of claim 2, wherein the method further comprises: determining a network topology of the power supply and distribution equipment network; and determining a network topology of the power supply and distribution equipment network. Step S104 includes, further including in the subtree exchange operation, the system first determines the gene segment corresponding to the electronic tree wherein is the switch index set of the complete feeder path from the power node to the load area all switch state variables along the path, the segment exchange operation exchanges the corresponding gene segment in the parent chromosome to generate two new child chromosomes, while ensuring that the topology maintains the integrity of the radial constraint, that is, each load area is still directly connected to the power node through a unique path, avoiding the formation of redundant loops or isolated nodes. After the exchange is completed, the system confirms that each child chromosome is connected to the power node through a topology connectivity check, ensuring the integrity of the power supply path. Subsequently, the system performs power flow calculation on each offspring to verify the satisfaction of electrical constraints, including that the voltage of each node is within the allowed range, the line flow does not exceed the rated capacity, and the load of the key node meets the design requirements; for individuals that exceed the voltage limit or overload in the power flow calculation, the system punishes their fitness, reduces their objective function value, and reduces their probability of being selected in the next generation, so as to ensure that feasible solutions that meet the electrical constraints are preferentially retained in the evolution process, thereby optimizing the network reliability and load balancing while ensuring the radial constraint of the topology.

4. The method of claim 3, wherein the method further comprises: Step S104 includes, in the switching of the variation operation, the system first identifies the formed ring network C to ensure that the network radial constraint is maintained in the variation process, the ring network identification adopts a depth-first search algorithm, starting from the closed tie switch endpoints a, b, traversing the network connected branches, tracking the node connection relationship along the path, and obtaining all switch state variable sets in the ring network wherein represents the set of all involved switch indexes on the path from node a to node b, is the closed tie switch itself, and the entire set C represents the complete switch loop formed by the ring network; The variation operation is performed by closing a randomly selected tie switch while randomly opening an existing closed circuit breaker in the ring network C to maintain network connectivity and re-adjust the power supply path; After mutation, the system first verifies whether the offspring chromosome still satisfies the radial topology constraint, ensures that each load node is connected to the power supply node through only one path, and then performs device capacity verification and performs power flow calculation on the line flow; if a line exceeds the rated capacity or is overloaded, the fitness of the offspring individual is degraded, and the probability of being selected in the subsequent non-dominated sorting selection is reduced, so as to ensure that feasible solutions that meet the radial constraint and capacity constraint are preferentially retained in the evolution process, and the joint optimization of network reliability and load balancing is realized.

5. The method of claim 4, wherein: Step S104 includes, in the decision strategy, the system adopts the comprehensive evaluation model combining entropy weight method with TOPSIS to the Pareto optimal solution set Sorting and selection, wherein n is the number of Pareto solutions, and m is the number of optimization objectives; the objective function value of each solution forms a matrix Wherein Indicates the evaluation value of the i-th solution on the j-th objective. The system first calculates the entropy value of each target Wherein The normalized contribution of the i-th solution on the j-th objective, entropy value Reflects the dispersion degree of the target information, the greater the dispersion degree indicates that the target has a more significant impact on the comprehensive evaluation; According to the entropy value, the weight of each target is calculated The higher the weight, the higher the proportion in the final comprehensive evaluation, then the system constructs a weighted normalized matrix: To eliminate dimensional differences and introduce target weights, ensure that different targets are comparable in comprehensive evaluation; determine the ideal solution based on the weighted normalized matrix And the negative ideal solution Respectively represent the optimal and worst values of each target, used to calculate the Euclidean distance of each solution to the ideal solution and the negative ideal solution wherein the smaller the value, the closer the solution is to the ideal solution, the larger the value, the further the solution is from the negative ideal solution; calculating the relative closeness of each solution based on distance The index comprehensively considers the closeness of the solution to the ideal solution and the distance from the negative ideal solution. The larger the value is, the better the comprehensive performance is. The system selects the solution with the maximum relative closeness as the final implementation scheme .

6. The method of claim 5, wherein: Step S105 includes that in the data repair processing, the system first receives the historical running state data sequence of each node in the network and the abnormal data of the current time step , wherein represents the historical running state of each node in the network; The system unrolls the historical data into a sequence input within a time window τ where each is the flattened state vector containing the load rate of all nodes , voltage indicators , device temperatures , link delays and packet loss rates to capture the timing patterns of the network operation; The system trains and predicts the historical sequence through a long short-term memory network (LSTM), learns the dynamic change law under normal operation, and outputs a prediction value represents a reasonable estimate of the state of each node at the current time step; To repair abnormal data, the system weights the original observation and the prediction result according to the confidence coefficient of the abnormality detection To repair abnormal data, the system weights the original observation and the prediction result according to the confidence coefficient of the abnormality detection ; The larger the λ, the higher the trust degree of abnormal observation value, the smaller the value, the more dependent on the predicted value to repair the anomaly, and the fused repaired data maintains the time sequence consistency and physical rationality; Finally, the system outputs the complete and repaired running state data To the network topology modeling and performance evaluation step, provide accurate and reliable input for subsequent topology optimization, ensure that the optimization decision can still be stable under the interference of abnormal data.

7. The method of claim 6, wherein the method further comprises: Step S105 includes, in seamless switching control, the system implements accurate execution of network topology reconstruction through the SDN controller, and receives a final reconstruction scheme A corresponding switch action instruction set Wherein S is a set of programmable circuit breakers in the network, Indicates closing the switch, Indicates opening the switch; The system maps each switch action as a sequence of flow table rules Where each rule Contains a match field For locating the corresponding switch and port, and an action field For specifying the switch action and an execution timestamp, ensuring that actions are implemented in a predetermined order; The SDN controller issues rules to corresponding switches by OpenFlow protocol in time periods, so as to realize synchronous issuing and controllable execution of instructions; and the execution sequence guarantee strategy is realized by setting operation intervals The electrical operation safety interval requirement between continuous switch actions is guaranteed, and the influence of current impact and voltage fluctuation on network stability is prevented. In the process of rule execution, the switch applies the action field in the time stamp order, executes the circuit breaker closing or opening operation in real time, the system simultaneously collects the execution state and outputs the confirmation signal to the monitoring system, which is used to verify whether the topology switching is successful, ensures the load balance, communication connectivity and continuity of electrical constraints in the whole network reconstruction process, and provides safe and reliable closed-loop control guarantee for dynamic optimization of large-scale power supply and distribution equipment.

8. The method for network topology dynamic optimization of large-scale power supply and distribution equipment networking according to claim 1, characterized in that: In the multi-objective optimization algorithm model, a deep neural network is introduced for auxiliary training to accelerate the evaluation of chromosome fitness by the genetic algorithm; the input layer of the network structure receives the chromosome code , where M is the total number of controllable switches in the network, and each element corresponds to the on-off state of a switch; the hidden layer is composed of a combination of fully connected layers and residual blocks, which capture the complex mapping relationship between the chromosome code and the network performance target through nonlinear transformation, and the introduction of the residual block alleviates the gradient vanishing problem in deep network training; the output layer generates the predicted value of the objective function , where correspond to the predicted values of the network reliability and the worst node load rate , respectively; The network training data comes from historical optimization process records wherein is the true value of the objective function for the ith chromosome; the training process optimizes the network parameters through a weighted mean squared error loss function: where the weights are the inverse of the variance of each objective function, used to balance the influence of different objectives during training, is the historical variance of objective . In the process of neural network training, the system adopts a transfer learning mechanism to use historical network topology data as a source domain, and current network topology data as a target domain, to improve the adaptability of the prediction model to new topology environments; the system introduces a domain adaptation layer in the deep neural network to minimize the maximum mean difference aligning the feature distributions of the source domain and the target domain, wherein is a feature mapping function mapping to a high-dimensional reproducing kernel Hilbert space H, and are the number of source domain and target domain samples respectively, and by minimizing MMD, the source domain pre-training model features and the target domain features are made as consistent as possible, thereby reducing the domain bias. Network weights are dynamically adjusted by an incremental update mechanism wherein the base weights from a source domain pre-trained model, providing global feature learning of historical network topology, incremental weights obtained by online fine-tuning from target domain data, used to capture specific changes and dynamic characteristics of the current network topology. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-8 when the computer program is executed by the processor. The computer program, when executed by the processor, causes the processor to perform the steps of the method of any one of claims 1 to 8.

Images