Edge computing unit configuration method and device for power distribution network fault self-recovery

Abstract

The invention provides an edge computing unit configuration method and device for power distribution network fault self-recovery, and relates to the technical field of power distribution network fault self-recovery. The method comprises the steps that each candidate deployment node during deployment of an edge calculation unit in the power distribution network is used as a game participant, and a potential function is constructed based on the self-recovery benefit and the operation cost of each game participant and a cooperative benefit coefficient between adjacent game participants so as to construct a target function; the method comprises the following steps: taking the deployment number and deployment positions of edge calculation units in a power distribution network during deployment as a particle, carrying out fault self-recovery responsibility area division on the power distribution network according to the deployment number corresponding to each particle, carrying out optimization on the particles according to a responsibility area division result and a target function, and obtaining an optimal particle which enables the target function to be minimum, and determining an edge calculation unit configuration scheme under power distribution network fault self-recovery. The rapid and accurate self-recovery requirement of a modern intelligent power distribution network can be met, and the self-recovery efficiency is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power distribution network fault self-recovery, and particularly relates to an edge computing unit configuration method and device for power distribution network fault self-recovery. BACKGROUND

[0002] Under the background of energy internet, the power distribution network is experiencing a transition from a traditional unidirectional power supply network to an intelligent and bidirectional interactive complex system. With this transition, the power distribution network is facing unprecedented challenges, especially in terms of fault handling. The access of a large number of intelligent sensing devices and distributed energy sources makes the power distribution network structure more complex, with frequent faults and various types, posing a serious challenge to the traditional fault handling mechanism.

[0003] The traditional centralized fault handling mode has obvious deficiencies in self-recovery: first, the self-recovery decision lags. With the expansion of intelligent power devices, the amount of fault data generated grows geometrically, while all information under the traditional architecture needs to be transmitted to the remote master station for processing before the recovery instruction is issued, causing serious self-recovery decision delay. In addition, the self-recovery capability is fragmented. The master station needs to handle massive data and various types of faults from each region, lacking a deep understanding of the local network characteristics, resulting in the universalization and extensive self-recovery strategy. Finally, the self-recovery path is single. Under the traditional mode, the self-recovery path is planned by the master station, lacking flexibility and diversity, and once the preset path is unavailable, the entire self-recovery process will fail.

[0004] As a distributed computing paradigm, edge computing provides a new technical path for solving the above problems by processing data and decision-making at the network edge. By configuring edge computing units, the computing capability can be lowered to the vicinity of the data source, realizing the localized processing and rapid response of data, thereby greatly reducing the self-recovery decision delay and communication burden, creating favorable conditions for power distribution network fault self-recovery.

[0005] However, the inventors found that although the prior art shortens the fault response time through the edge computing architecture, in the research on the configuration of the edge computing unit for the power distribution network fault self-recovery, the edge computing unit is usually regarded as a decision node running independently, and this discrete processing method directly leads to the lack of cross-regional coordination mechanism. When the fault occurs at the boundary of the power supply area or needs multi-area collaborative recovery, each edge unit often falls into a "local optimal trap" due to the lack of an effective interaction model, and cannot form a globally optimal self-recovery scheme. More seriously, the traditional regional division method such as spectral clustering has obvious limitations in dealing with the dynamic topology characteristics of the power distribution network - the division results based on static network characteristics are difficult to adapt to dynamic factors such as load fluctuation and communication time delay difference, which makes the responsibility area boundary frequently appear decision conflicts during fault propagation. At the same time, the current mainstream deterministic optimization algorithm exposes inherent defects when facing complex and variable fault scenarios: the preset fixed parameter system cannot effectively respond to the uncertainty brought by network topology reconstruction and distributed power access, resulting in a serious lack of generalization ability of the self-recovery strategy. That is, the current method has significant defects in collaborative decision-making mechanism, regional division accuracy and dynamic optimization of strategy, and these problems together cause the existing system to have large fluctuations in self-recovery success rate and poor flexibility of recovery path when dealing with large-scale complex faults, which cannot meet the strict requirements of modern smart power distribution networks for fast and accurate self-recovery, and becomes a technical bottleneck restricting the improvement of self-recovery efficiency. SUMMARY

[0006] Embodiments of the present application provide a method and device for configuring edge computing units for power distribution network fault self-recovery to solve the problems in the process of configuring edge computing units in terms of collaborative decision-making mechanism, regional division accuracy and dynamic optimization of strategy.

[0007] In a first aspect, embodiments of the present application provide a method for configuring edge computing units for power distribution network fault self-recovery, comprising:

[0008] Each candidate deployment node of the edge computing unit in the power distribution network is taken as a game participant, a potential function is constructed based on the self-recovery benefit, operation cost and coordination benefit coefficient between adjacent game participants of each game participant, and a target function is constructed according to the potential function;

[0009] The deployment quantity and deployment position of the edge computing unit in the power distribution network are taken as a particle, the responsibility area is divided according to the deployment quantity corresponding to each particle for the fault self-recovery of the power distribution network, and the responsibility area division result corresponding to each particle is obtained;

[0010] According to the responsibility area division result and the target function, the particles are optimized to obtain an optimal particle that minimizes the target function, and the deployment quantity, the deployment position and the responsibility area division result corresponding to the optimal particle are determined as the edge computing unit configuration scheme under the power distribution network fault self-recovery.

[0011] In a second aspect, an embodiment of the present application provides an edge computing unit configuration device for power distribution network fault self-recovery, comprising:

[0012] A first processing module is configured to take each candidate deployment node of the edge computing unit in the power distribution network as a game participant, construct a potential function based on the self-recovery benefit, the operation cost and the synergy benefit coefficient between adjacent game participants of each game participant, and construct a target function according to the potential function;

[0013] A second processing module is configured to take the deployment quantity and the deployment position of the edge computing unit in the power distribution network as a particle, divide the responsibility area of the power distribution network according to the deployment quantity corresponding to each particle for fault self-recovery, and obtain the responsibility area division result corresponding to each particle;

[0014] A third processing module is configured to optimize the particle according to the responsibility area division result and the target function, obtain an optimal particle that minimizes the target function, and determine the deployment quantity, the deployment position and the responsibility area division result corresponding to the optimal particle as the edge computing unit configuration scheme under the power distribution network fault self-recovery.

[0015] In a third aspect, an embodiment of the present application provides an electronic device, comprising a memory and a processor, the memory stores a computer program, and the processor implements the method in the first aspect or any possible implementation manner of the first aspect when executing the computer program.

[0016] In the embodiment of the present application, each candidate deployment node in the deployment of the edge computing unit in the power distribution network is taken as a game participant, a potential function is constructed based on the self-recovery benefit, operation cost and synergy benefit coefficient between adjacent game participants of each game participant, a target function is constructed according to the potential function, the deployment quantity and deployment position in the deployment of the edge computing unit in the power distribution network are taken as a particle, the responsibility area division for the power distribution network is performed according to the deployment quantity corresponding to each particle, the responsibility area division result corresponding to each particle is obtained, then the particle is optimized according to the responsibility area division result and the target function, the optimal particle that makes the target function minimum is obtained, and the deployment quantity, deployment position and responsibility area division result corresponding to the optimal particle are determined as the edge computing unit configuration scheme under the power distribution network fault self-recovery. The dynamic game model between the edge computing units can be first established through the potential game theory, the decision-making behavior of each edge computing unit is mapped to a unified potential function space, the distributed decision-making process converges to the globally optimal Nash equilibrium state, and the cross-region synergy problem is fundamentally solved. On this basis, the responsibility area division for the power distribution network is performed according to the deployment quantity corresponding to each particle, which is helpful to realize the adaptive division of the responsibility area boundary, so that the responsibility area boundary can dynamically adapt to the network topology change and fault propagation characteristics, and the load balancing of the edge computing unit is significantly improved, then the particle is optimized according to the responsibility area division result and the target function, the self-recovery strategy can be autonomously adjusted according to the real-time network state, the reliability of the recovery process is ensured while the resource redundancy is reduced, thereby helping to meet the fast and accurate self-recovery requirements of the modern smart power distribution network and improve the self-recovery efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 is an implementation flowchart of the edge computing unit configuration method for power distribution network fault self-recovery provided by the embodiment of the present application;

[0018] Figure 2 is an implementation flowchart of the responsibility area division for fault self-recovery provided by the embodiment of the present application;

[0019] Figure 3 is an implementation flowchart of the particle optimization provided by the embodiment of the present application;

[0020] Figure 4 is an implementation flowchart of the edge computing unit configuration method for power distribution network fault self-recovery provided by another embodiment of the present application;

[0021] Figure 5 is a structural schematic diagram of the edge computing unit configuration device for power distribution network fault self-recovery provided by the embodiment of the present application;

[0022] Figure 6 is a schematic diagram of an electronic device provided by the embodiment of the present application. Detailed Implementation

[0023] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0024] See Figure 1 The document illustrates a flowchart of the edge computing unit configuration method for self-recovery of distribution network faults provided in an embodiment of the present invention, which is described in detail below:

[0025] Step 101: Take each candidate deployment node of the edge computing unit in the distribution network as a game player, construct a potential function based on the self-recovery benefit, operating cost and the synergy benefit coefficient with neighboring game players of each game player, and construct the objective function based on the potential function.

[0026] Optionally, the potential function can be constructed based on each player's self-recovery benefit, operating cost, and cooperative benefit coefficient with neighboring players. This can include:

[0027] The payoff function for each player is constructed based on their self-recovery benefit, operating cost, and synergistic benefit coefficient with neighboring players.

[0028] Based on the theory of potential games, a potential function is constructed according to the payoff function of each game participant.

[0029] Optionally, the objective function constructed based on the potential function may include:

[0030] Based on different deployment schemes for edge computing units in the distribution network, the self-recovery time, power restoration range, and self-recovery reliability of the distribution network under the corresponding deployment schemes are calculated.

[0031] The self-recovery time, power restoration range, and self-recovery reliability are all normalized to obtain normalized self-recovery time, normalized power restoration range, and normalized self-recovery reliability.

[0032] The objective function is constructed based on the normalized self-recovery time, normalized power supply recovery range, normalized self-recovery reliability, and potential function.

[0033] For example, calculating the self-recovery time of the distribution network under different deployment schemes for edge computing units in the distribution network can include:

[0034] Based on different deployment schemes for edge computing units in the distribution network, the fault detection time, fault location time, isolation decision time, control command communication and issuance time, and power restoration execution time of the distribution network under the corresponding deployment scheme are obtained.

[0035] The self-recovery time of the distribution network under the corresponding deployment scheme is obtained by summing the fault detection time, fault location time, isolation decision time, control command communication and issuance time, and power restoration execution time.

[0036] In this embodiment of the invention, potential game theory is first introduced to model the deployment decision process of potential distributed edge computing units (ECUs) in the distribution network as a game process. Each candidate deployment node in the distribution network is regarded as a player in the game, and its strategy is whether to deploy an ECU, using a binary decision variable s. i This can be represented (for example, s can be used) i =1 indicates deployment, s i =0 indicates no deployment). Based on this, a payoff function u is defined for each game participant i. i (S), this function measures the decision of player i to deploy an ECU (s) given a combination of global deployment strategies S. i =1) The overall benefits that can be obtained. This benefit function can mainly consider three factors: ECU i The self-healing benefit R provided by itself i (S), this benefit is related to its area of ​​responsibility (determined by a given global deployment strategy combination S), processing capacity, and load within the area; and to other deployed neighboring ECUs. j The additional benefits brought by collaboration (belonging to set N(i)) are quantified as the collaboration benefit coefficient C. ij ; and the cost or resource consumption required to deploy and run the ECU at node i. i (That is, operating costs). For example, the revenue function can take the following form:

[0037]

[0038] Where S = (s1, s2, ..., s n This represents the overall strategy combination that includes deployment decisions for all n candidate nodes, i.e., a complete ECU deployment scheme. i and s j Let N(i) be the deployment decision variables for nodes i and j, respectively. N(i) represents the variables related to the ECU. i A set of neighboring ECUs that have cooperative relationships. (C) ij For ECU i and ECU j The synergistic benefit coefficient between them. i The cost of deploying the ECU at node i. α, β, and γ are preset weighting coefficients used to adjust the relative importance of each component's revenue or cost in the total revenue.

[0039] On this basis, a global potential function Φ(S) is constructed for the system according to the potential game theory. The potential function reflects the overall benefit state of the system and needs to satisfy the following specific condition: for any game participant i, when it changes its strategy (from s i to s′ i ) alone, resulting in the overall strategy combination changing from S=(s i , S -i ) to S′=(s′ i , S -i ) (where S -i represents the unchanged strategies of all other game participants except i), the change amount u i (S′)-u i (S) of the benefit of game participant i itself must be exactly equal to the change amount Φ(S′)-Φ(S) of the global potential function. Based on the above-defined benefit function u i (S) and assuming that the synergy benefit is symmetric (i.e., C ij =C ji ), the exact potential function satisfying the above condition can be constructed as follows:

[0040]

[0041] where n is the total number of candidate nodes. The strategy combination S * corresponding to the global maximum of the potential function is the pure strategy Nash equilibrium point of the system, which represents a stable and overall optimal ECU configuration scheme of the system (including the total self-recovery benefit, the total synergy benefit, and deducting the total cost). Therefore, the objective function can be constructed based on the potential function, and the optimization goal is to find the strategy combination S that maximizes Φ(S).

[0042] On this basis, key performance evaluation indicators are defined to comprehensively evaluate the influence of the ECU configuration scheme on the fault self-recovery performance.

[0043] First, the self-recovery time T rec is the core of measuring recovery efficiency, which is composed of five parts: fault detection time T det , fault location time T loc , isolation decision time T dec , control instruction communication and issuance time T com , and power supply recovery execution time T res .

[0044] T rec =T det +T loc +T dec +T com +T res ;

[0045] wherein, Tdet Decided by the monitoring device;T loc Related to the positioning algorithm efficiency, ECU computing capability and fault complexity;T dec Depends on the decision algorithm complexity, ECU performance and power grid state;T com Influenced by the network structure, communication protocol and ECU deployment location;T res Mainly decided by the switching device action time. In actual evaluation, the specific value of T loc , T dec , T res usually needs to be obtained according to the used algorithm, hardware specification, network characteristics and fault scene, through power grid simulation, establishment of analytical model estimation or according to field test and historical data statistical analysis.

[0046] For a system with k ECUs deployed, the control instruction communication and issuing time T com (k) can be further analyzed, for example, considering the transmission, propagation and queuing delay and other factors. A possible estimation model is:

[0047]

[0048] In the formula, P is the average data packet size, v n is the communication network transmission rate, N t is the total number of monitoring, control terminal devices (such as FTU, smart meter, controllable switch, etc.) in the network that need to communicate with the ECU (note that N t is different from the number of ECU deployments k), is the average communication path length from the terminal to its ECU, v m is the signal propagation speed, h is the average routing hop number, λ(k) is the data arrival rate (which can be related to k, N t ), and μ is the routing service efficiency. This formula shows that the communication delay is affected by multiple factors, especially the deployment number k and the deployment location of the ECU, which will affect the average communication distance and network load, and then affect T com (k).

[0049] Secondly, the power supply recovery range η load (k, S) is defined as the proportion of the load that can be restored after fault isolation to the total load:

[0050]

[0051] In the formula, P rec (k, S) is the recoverable load power under the deployment of k ECUs according to the strategy S, which depends on the coordination capability and configuration location of the ECU, and can be represented as the independent recovery power P i own and the coordinated recovery power The sum of the above. total is the total load of the system.

[0052] Finally, the self-recovery reliability ψ rel (k, S), which reflects the success rate of the self-recovery process, is defined as the ratio of the number of successful self-recovery N success (k, S) to the total number of failures N total :

[0053]

[0054] The self-recovery reliability is particularly related to factors such as ECU performance, network state, and algorithm robustness.

[0055] When constructing the objective function, in order to make the key performance evaluation indicators of different dimensions comparable, first, normalize each key performance evaluation indicator:

[0056]

[0057] where T rec,min and T rec,max are the minimum and maximum values of the self-recovery time, respectively; η load,min and η load,max are the minimum and maximum values of the power supply recovery range, respectively; ψ rel,min and ψ rel,max are the minimum and maximum values of the self-recovery reliability, respectively.

[0058] The determination of these minimum and maximum values can usually be based on theoretical boundary analysis, historical data statistics, simulation evaluation results covering multiple scenarios, or system design performance requirements. For example, T rec,min can take the estimated value of the fastest response of the system, T rec,max can take the maximum tolerated recovery time of the system; η load,min can take 0, η load,max can take 1 (or the maximum value after considering constraints); ψ rel,min can take 0, ψ rel,max can take 1.

[0059] After normalization, combined with the potential function in the potential game model, the objective function F(k, S) is constructed as the comprehensive optimization objective function:

[0060] min F(k, S) = ω1T′ rec - ω2η′ load - ω3ψ′ rel - ω4Φ(S);

[0061] which expands to:

[0062] wherein ω1, ω2, ω3 and ω4 are weight coefficients, satisfying The comprehensive optimization objective function aims to minimize the self-recovery time while maximizing the power supply recovery range, self-recovery reliability and potential function value (i.e. the overall self-recovery benefit of the system).

[0063] Embodiments of the present application aim at the problem of "decision island" caused by the lack of effective coordination among edge computing units. A distributed decision optimization model is constructed by introducing potential game theory, which maps the local decision behavior of each edge computing unit to a unified potential function space. The model designs a complex potential function including self-recovery benefit, coordination gain coefficient and resource consumption (i.e. operation cost), so that the discretely distributed edge computing units naturally tend to the global optimal state in the process of pursuing individual benefit maximization, successfully solving the Nash equilibrium problem of cross-regional coordination. Moreover, embodiments of the present application break through the limitations of traditional single-index optimization and construct a comprehensive evaluation system covering multiple dimensions such as self-recovery time, power supply recovery range, self-recovery reliability, etc. By establishing a quantitative mapping model between edge computing unit configuration parameters and self-recovery performance indicators, a closed-loop feedback mechanism of "coordinated decision-making - regional division - strategy optimization" is realized. This three-dimensional coupled intelligent configuration method not only significantly improves the recovery speed of single-point faults, but also shows strong strategy adaptive ability when dealing with complex scenarios such as cascading failures, providing a new technical paradigm for building an intelligent self-recovery system with autonomous evolution ability.

[0064] In step 102, the deployment quantity and deployment location of the edge computing unit in the power distribution network are taken as a particle, and the power distribution network is divided into a self-recovery responsibility area according to the deployment quantity corresponding to each particle, to obtain a responsibility area division result corresponding to each particle.

[0065] Optionally, the power distribution network is divided into a self-recovery responsibility area according to the deployment quantity corresponding to each particle, to obtain a responsibility area division result corresponding to each particle, which can include:

[0066] The topological structure of the power distribution network is modeled as a graph structure, an adjacency matrix and a degree matrix corresponding to the graph structure are calculated, and a graph Laplacian matrix is constructed according to the adjacency matrix and the degree matrix.

[0067] The graph Laplacian matrix is spectrally decomposed to obtain eigenvalues and eigenvectors of the graph Laplacian matrix.

[0068] Based on a preset wavelet kernel function and each preset analysis scale, the graph wavelet operator under each preset analysis scale is obtained in combination with the eigenvalues and the eigenvectors.

[0069] Calculate the response intensity of each graph wavelet operator after it is applied to a unit signal centered on each node in the graph structure, and record it as the target feature value of each node at each preset analysis scale.

[0070] Construct a feature matrix based on each target feature value.

[0071] The number of deployments corresponding to each particle is used as the number of clusters. All nodes in the distribution network are clustered according to the feature matrix and the number of clusters, and the clustering results are used as the responsibility area division results for each particle.

[0072] In this embodiment of the invention, to achieve self-recovering responsibility area division that is sensitive to the distribution network topology and can reflect connection characteristics at different scales, graph wavelet transform (GWT) technology is used for in-depth analysis. For example... Figure 2 As shown, the specific implementation process can be as follows:

[0073] 1. Distribution Network Graph Structure Representation and Laplace Matrix Construction: First, the distribution network topology to be analyzed is abstracted into a mathematical graph model G = (V, E, W). Here, V represents the set of nodes in the distribution network (e.g., substations, switching stations, load nodes, etc.), E represents the set of connections between nodes (i.e., power lines), and W is an optional set of edge weights, which can be set based on line impedance, length, or simply connectivity (weight I). Based on this graph model, its corresponding adjacency matrix A and degree matrix D are calculated. Then, the graph Laplace matrix L = DA is constructed for subsequent spectral analysis, or its normalized form (e.g., L...) can be used. sym =ID -1 / 2 AD -1 / 2 To adapt to different network characteristics.

[0074] 2. Spectral Decomposition of the Graph Laplacian Matrix: Perform spectral decomposition (i.e., eigenvalue decomposition) on the graph Laplacian matrix L constructed in the previous step. This process calculates all eigenvalues ​​λ of matrix L. i (forming a diagonal matrix Λ) and its corresponding eigenvector u i (Columns forming an orthogonal matrix U) satisfy L=UΛU T This decomposition reveals the inherent frequency characteristics of the graph structure and forms the basis for graph wavelet analysis.

[0075] 3. Graph wavelet transform parameter setting and operator definition: To perform graph wavelet transform, a suitable wavelet kernel function g needs to be selected in advance. t The (·) function operates on the spectral domain (i.e., eigenvalues) of the graph and defines the filtering method for the signal at different scales. Commonly used wavelet kernel functions include the hot kernel g. t (λ)=e -tλ Or Mexican hat small wave gt (λ)=λe -λ2 / t And so on. Simultaneously, a series of discrete analytical scales t∈{t1,t2,…,t} are selected. m This is used to capture structural information of the power distribution network at different granularities or levels. For each selected scale t... j Combining the spectral decomposition results and the selected wavelet kernel function, the graph wavelet operator at this scale is defined as follows: This operator forms the core tool for performing multi-scale analysis on graphs.

[0076] 4. Node Multi-Scale Feature Extraction and Feature Matrix Construction: Utilizing the defined graph wavelet operator, feature extraction and feature matrix construction are performed for each node v in the distribution network. i Extraction can characterize it at various scales t j Numerical characteristics f of the importance of the substructure i,j The specific method for feature extraction is: computational graph wavelet operator. Acting on node v i The unit signal δ centered i The response intensity after that is used as node v i The target eigenvalue at this scale reflects the energy or centrality of a node at this structural scale. Using this method, for each node v... i At all selected scales t1, t2, ..., t m The following obtains a set of eigenvalues ​​{f i,1 ,f i,2 ,…,f i,m By concatenating these cross-scale eigenvalues, the node v is formed. i Multiscale eigenvector x i ={f i,1 ,f i,2 ,…,f i,m Finally, the feature vectors x of all n nodes are... i Stack them row by row to form the final feature matrix X∈R for subsequent processing. n×d , where d is the total feature dimension after concatenation.

[0077] 5. Clustering and Preliminary Division of Responsibility Regions Based on Feature Matrix: Using the feature matrix X obtained in the previous step as input, a clustering algorithm (e.g., k-means algorithm) is used to group all nodes in the distribution network. The clustering algorithm can be based on the similarity of nodes in a multi-scale feature space (i.e., feature vector x). iThe n nodes are divided into k clusters according to the distance between the nodes). Wherein, the cluster number k here is not preset in this step, but is dynamically determined by the optimization algorithm in the subsequent step according to the currently evaluated candidate solution (that is, the deployment number corresponding to each particle in the current population) in the search process. That is, the clustering process is performed according to the k value given by each particle of the current iteration of the subsequent optimization algorithm. Each cluster C l (l = 1, 2, …, k) constitutes an initial self-recovery responsibility area. In each preliminary divided area C l , the node with the highest score in the area can be selected as the initial ECU control center of the area by calculating the comprehensive centrality score of each node in the area The set of initial area division C l and control center position constitutes the basis for evaluating the performance of the configuration scheme, and its advantages and disadvantages will be reflected by calculating the objective function F(k, S), and will be implicitly optimized in the iteration process of the subsequent step, in order to meet the requirements of communication delay, load balancing and other aspects.

[0078] The embodiment of the application innovatively introduces the time-frequency analysis capability of wavelet transform into the power distribution network topology analysis in view of the single scale and poor adaptability of the traditional spectral clustering method in responsibility area division. By converting the network topology relationship into a multi-dimensional graph signal, using the multi-scale decomposition characteristics of wavelet transform, a dynamic feature graph spectrum analysis model is constructed. The model can simultaneously capture the macro-level structure and micro-node characteristics of the network topology, and realize dynamic adjustment of the responsibility area boundary with load fluctuation and fault propagation path change. Especially in the processing of dynamic reconfiguration of topology caused by distributed power access, it shows the adaptability that traditional methods cannot achieve, and lays a precise spatial reference for subsequent self-recovery strategy optimization.

[0079] Step 103, optimizing the particles according to the responsibility area division result and the objective function, obtaining the optimal particle that minimizes the objective function, and determining the deployment number, deployment position and responsibility area division result corresponding to the optimal particle as the edge computing unit configuration scheme under the fault self-recovery of the power distribution network.

[0080] Optionally, before the fault self-recovery responsibility area division of the power distribution network according to the deployment number corresponding to each particle is performed, the responsibility area division result corresponding to each particle can also be obtained.

[0081] Initializing the population based on chaotic mapping to obtain an initial population, the initial population including a plurality of particles.

[0082] Optionally, the optimization of the particles according to the responsibility area division result and the objective function to obtain the optimal particle that minimizes the objective function can include:

[0083] According to the responsibility region division result corresponding to each particle in the current population and the target function, the fitness value of each particle in the current population is calculated.

[0084] According to the fitness value of each particle in the current population, the individual optimal particle and the global optimal particle of the current population are determined, and the adaptive inertia weight and the adaptive learning factor are obtained according to the current iteration number corresponding to the current population.

[0085] It is judged whether the individual optimal particle and / or the global optimal particle are improved continuously for N stag generations.

[0086] If the individual optimal particle and / or the global optimal particle are improved, each particle in the current population is updated according to the individual optimal particle, the global optimal particle, the adaptive inertia weight and the adaptive learning factor.

[0087] If the individual optimal particle and / or the global optimal particle are improved continuously for N stag generations, each particle in the current population is updated according to the individual optimal particle, the global optimal particle, the adaptive inertia weight, the adaptive learning factor and the chaotic mapping sequence.

[0088] The population obtained after updating is taken as a new current population, and the step of calculating the fitness value of each particle in the current population according to the responsibility region division result corresponding to each particle in the current population and the target function is re-executed, and the subsequent steps are re-executed.

[0089] Until the current iteration number reaches the preset iteration number, the optimal particle that minimizes the target function is obtained according to the global optimal particle corresponding to the current iteration number.

[0090] Optionally, obtaining the adaptive inertia weight according to the current iteration number corresponding to the current population can include:

[0091] According to the fitness value of each particle in the current population, the average fitness value, the maximum fitness value and the minimum fitness value of the current population, the inertia weight adjustment amount corresponding to each particle in the current population is calculated.

[0092] According to the inertia weight adjustment amount corresponding to each particle in the current population and the current iteration number corresponding to the current population, the adaptive inertia weight corresponding to each particle in the current population is obtained.

[0093] Exemplarily, the embodiments of the present application can employ a particle swarm optimization algorithm, such as an adaptive chaotic particle swarm optimization (ACPSO) algorithm, to solve the multi-objective optimization problem defined in the foregoing steps to determine the optimal number of deployed edge computing units k opt and their deployment locations S opt . The ACPSO searches for the optimal solution by simulating the social behavior of a particle swarm and introduces adaptive parameter adjustment and a chaotic mechanism to enhance search efficiency and avoid falling into local optima, as shown in the following detailed process. Figure 3

[0094] Exemplarily, when initializing the population, a real number coding method can be employed, each particle can contain two parts of information of the number of deployed edge computing units k and the deployment locations S, and an initial population is generated through chaotic mapping to improve population diversity:

[0095] z i+1 = 4z i (1-z i ), z i ∈(0, 1);

[0096]

[0097] where z i is a chaotic mapping sequence, is the initial position of the i-th particle, corresponding to a vector composed of the number of deployed edge computing units k and the deployment locations S.

[0098] Exemplarily, when updating the particles, the velocity and position update equations of the particles are as follows:

[0099]

[0100] where vi(t) is the velocity of the i-th particle at the t-th iteration, corresponding to the change of the vector composed of the number of deployed edge computing units k and the deployment locations S. xi(t) is the position of the i-th particle at the t-th iteration, corresponding to the vector composed of the number of deployed edge computing units k and the deployment locations S. pi(t) is the individual optimal particle, that is, the optimal value of the i-th particle from the 1st iteration to the t-th iteration. g(t) is the global optimal particle, that is, the optimal particle in the t-th iteration process. w is an adaptive inertia weight, c1 and c2 are adaptive learning factors, and r1 and r2 are random numbers in the interval (0, 1).

[0101] ​​In this design, an adaptive inertia weight w is used to balance the capabilities of global search and fine-grained local search.

[0102]

[0103] Among them, w max and w min These are the maximum and minimum values ​​of the adaptive inertia weight, respectively. T max The maximum number of iterations, also known as the preset number of iterations, is given by t, where t is the current iteration number. Δw is the inertia weight adjustment, and w1 and w2 are adjustment coefficients. f is the fitness value of the current particle. avg f is the average fitness value of the current population. min f is the minimum fitness value of the current population. max This represents the maximum fitness value of the current population.

[0104] To further improve the optimization efficiency of the self-recovery strategy, an adaptive learning factor is designed:

[0105]

[0106]

[0107] Among them, c 1i Let c1 be the initial value, and c 1f The final value of c1, c 2i Let c2 be the initial value, and c 2f This is the final value of c2.

[0108] Based on this, to avoid the self-recovery strategy optimization getting trapped in local optima, when the algorithm continuously N stag If no improvements are made, a chaotic local search mechanism is triggered:

[0109]

[0110] Where ε is the search step size coefficient, x ub x is the upper bound of the search space. lb This is the lower bound of the search space.

[0111] Ultimately, the optimal ECU configuration scheme obtained by the ACPSO algorithm converges, i.e., the optimal number of ECUs deployed, k. opt Optimal ECU deployment location set S opt And the optimal responsibility area division associated with this configuration. Together, they form the foundation for deploying physical systems. For example... Figure 4 As shown, based on this optimization result, it can be used to deploy in each regional control center. The ECU is configured with specific self-recovery strategies and operating logic.

[0112] For example, these strategies can specify in detail the operation procedures of each ECU in its responsible area and the coordination with other ECUs, such as including the confirmation of faults in the area, the isolation operation sequence, the load recovery method using local resources (while meeting the grid operation constraints), and the information interaction protocol and decision rules for boundary fault coordination and cross-area load support with adjacent area ECUs. These specific strategies can be designed in advance and fixed in the ECU device in the form of rule base, algorithm program, etc.

[0113] Furthermore, after the system is put into operation, the actual performance indicators such as actual recovery time, recovery range, operation accuracy, success rate, etc. of each fault self-recovery event can be monitored and recorded, and these actual performances are compared with the design target or simulation expectation. If there is a deviation or deficiency in performance, the strategy parameters deployed in the ECU can be adjusted. These adjustable parameters can involve fault detection threshold, load recovery priority setting, communication timeout threshold, safety margin, coordination response mechanism parameters, etc. The parameter adjustment based on actual operation feedback can be completed by artificial periodic maintenance, or can be designed to have certain self-learning ability for online adaptive adjustment, thereby forming a continuous improvement cycle of "configuration deployment-operation monitoring-performance evaluation-strategy optimization", constituting a closed-loop optimization mechanism, and continuously improving the performance of the power distribution network fault self-recovery system and the adaptability to changes in the operating environment.

[0114] The embodiment of the application aims at the problem that the discrete distributed edge units under the traditional architecture often have difficulty in forming effective coordination when facing complex cross-area faults, and the statically divided responsibility areas cannot adapt to the dynamic topology characteristics of the power distribution network, thereby directly leading to the difficulty in breaking through the existing bottleneck of self-recovery efficiency and power supply reliability. A three-level collaborative optimization system is constructed: first, a dynamic game model between edge units is established by potential game theory, the decision-making behavior of each unit is mapped to a unified potential function space, and the distributed decision-making process converges to a globally optimal Nash equilibrium state, thereby fundamentally solving the cross-area coordination problem. On this basis, the wavelet transform technology with multi-scale analysis capability is introduced, the power distribution network topology is analyzed in time and frequency domains, the dynamic characteristic spectrum of the key nodes in the network hierarchy is captured, and the adaptive division of the responsibility area boundary is realized. At the same time, aiming at the strategy optimization demand in the complex fault scene, an adaptive particle swarm algorithm with chaotic disturbance mechanism is developed, the dynamic inertia weight mechanism of which can effectively balance the global exploration and local optimization ability, and cooperates with the real-time topology perception module to form a closed-loop optimization system.

[0115] The embodiment of the application realizes systematic innovation of power distribution network fault processing efficiency by fusing the technical architecture of potential game collaborative decision-making, multi-scale dynamic division and adaptive strategy optimization. Compared with the traditional centralized architecture, the edge computing unit collaborative mechanism constructed by the method significantly shortens the fault recovery response cycle, and greatly expands the load coverage range of recoverable power supply, especially in handling cross-regional complex faults, which shows stronger collaborative recovery capability. The multi-scale topology analysis technology based on wavelet transform breaks through the limitations of traditional static regional division, so that the responsibility area boundary can dynamically adapt to the network topology change and fault propagation characteristics, significantly improving the load balancing of the edge unit. By introducing the adaptive optimization algorithm of the chaos disturbance mechanism, the system can automatically adjust the self-recovery strategy according to the real-time network state, reduce resource redundancy, and ensure the reliability of the recovery process. This technical system can still maintain efficient fault isolation and power supply recovery capability when dealing with complex scenarios such as distributed energy access and dynamic topology reconstruction, providing key technical support for building an intelligent power distribution network with autonomous decision-making and dynamic evolution capability.

[0116] It should be understood that the size of the serial number of each step in the above embodiment does not mean the order of execution, and the execution order of each process should be determined according to its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiment of the application.

[0117] The following is the device embodiment of the application. For details not described in detail, please refer to the corresponding method embodiments described above.

[0118] Figure 5 The structure of the edge computing unit configuration device for power distribution network fault self-recovery provided by the embodiment of the application is shown. For ease of illustration, only the parts related to the embodiment of the application are shown, and the details are as follows:

[0119] As shown in Figure 5 The edge computing unit configuration device for power distribution network fault self-recovery includes a first processing module 51, a second processing module 52 and a third processing module 53.

[0120] The first processing module 51 is configured to take each candidate deployment node of the edge computing unit in the power distribution network as a game participant, construct a potential function based on the self-recovery benefit, operation cost and collaborative benefit coefficient between adjacent game participants of each game participant, and construct a target function according to the potential function.

[0121] The second processing module 52 is configured to take the deployment quantity and deployment position of the edge computing unit in the power distribution network as a particle, divide the responsibility area of the power distribution network for fault self-recovery according to the deployment quantity corresponding to each particle, and obtain the responsibility area division result corresponding to each particle.

[0122] The third processing module 53 is configured to perform optimization on the particles according to the responsibility area division result and the target function, obtain an optimal particle that minimizes the target function, and determine the deployment quantity, the deployment position, and the responsibility area division result corresponding to the optimal particle as an edge computing unit configuration scheme under fault self-recovery of the power distribution network.

[0123] In a possible implementation, the first processing module 51 can be configured to construct a benefit function of each game participant based on self-recovery benefits, operation costs, and synergy benefit coefficients between the game participant and adjacent game participants; and construct a potential function according to the benefit function of each game participant based on potential game theory.

[0124] In a possible implementation, the first processing module 51 can be configured to calculate self-recovery time, power supply recovery range, and self-recovery reliability of the power distribution network under different deployment schemes of the edge computing unit in the power distribution network; normalize the self-recovery time, the power supply recovery range, and the self-recovery reliability to obtain normalized self-recovery time, normalized power supply recovery range, and normalized self-recovery reliability; and construct a target function according to the normalized self-recovery time, the normalized power supply recovery range, the normalized self-recovery reliability, and the potential function.

[0125] In a possible implementation, the first processing module 51 can be configured to obtain fault detection time, fault positioning time, isolation decision time, control instruction communication and issuing time, and power supply recovery execution time of the power distribution network under different deployment schemes of the edge computing unit in the power distribution network; and calculate a sum of the fault detection time, the fault positioning time, the isolation decision time, the control instruction communication and issuing time, and the power supply recovery execution time to obtain self-recovery time of the power distribution network under the corresponding deployment scheme.

[0126] In a possible implementation, the second processing module 52 can be configured to model the topology of the power distribution network as a graph structure, calculate an adjacency matrix and a degree matrix corresponding to the graph structure, and construct a graph Laplacian matrix according to the adjacency matrix and the degree matrix; perform spectral decomposition on the graph Laplacian matrix to obtain eigenvalues and eigenvectors of the graph Laplacian matrix; obtain a graph wavelet operator under each preset analysis scale based on a preset wavelet kernel function and each preset analysis scale, in combination with the eigenvalues and the eigenvectors; calculate a response intensity of each graph wavelet operator acting on a unit signal centered on each node in the graph structure, denoted as a target eigenvalue of each node under each preset analysis scale; construct a feature matrix according to each target eigenvalue; and cluster all nodes in the power distribution network according to the feature matrix and a clustering number corresponding to each particle, and use a clustering result as a responsibility area division result corresponding to each particle.

[0127] In a possible implementation, the second processing module 52 can also be configured to initialize a population based on a chaotic mapping to obtain an initial population, the initial population including a plurality of particles.

[0128] In a possible implementation, the third processing module 53 can be configured to take the initial population as a current population, calculate an adaptive value of each particle in the current population according to the responsibility area division result corresponding to each particle in the current population and the target function, determine an individual optimal particle and a global optimal particle of the current population according to the adaptive value of each particle in the current population, and obtain an adaptive inertia weight and an adaptive learning factor according to a current iteration number corresponding to the current population. stag If the individual optimal particle and / or the global optimal particle have improved for N consecutive iterations, update each particle in the current population according to the individual optimal particle, the global optimal particle, the adaptive inertia weight, and the adaptive learning factor. stag If the individual optimal particle and / or the global optimal particle have not improved for N consecutive iterations, update each particle in the current population according to the individual optimal particle, the global optimal particle, the adaptive inertia weight, the adaptive learning factor, and a chaotic mapping sequence; take the population obtained after the update as a new current population, and re-perform the step of calculating the adaptive value of each particle in the current population according to the responsibility area division result corresponding to each particle in the current population and the target function, and subsequent steps; and until the current iteration number reaches a preset iteration number, obtain an optimal particle that minimizes the target function according to the global optimal particle corresponding to the current iteration number.

[0129] In a possible implementation, the third processing module 53 can be configured to calculate an inertia weight adjustment amount corresponding to each particle in the current population according to the fitness value of each particle in the current population, the average fitness value of the current population, the maximum fitness value and the minimum fitness value; and obtain an adaptive inertia weight corresponding to each particle in the current population according to the inertia weight adjustment amount corresponding to each particle in the current population and the current iteration number corresponding to the current population.

[0130] Figure 6 is a schematic diagram of an electronic device provided by an embodiment of the present application. As shown in Figure 6 the electronic device 6 of this embodiment includes a processor 60 and a memory 61. The memory 61 stores a computer program 62. The processor 60 implements the steps in each of the above method embodiments when executing the computer program 62. Alternatively, the processor 60 implements the functions of each module / unit in each of the above apparatus embodiments when executing the computer program 62.

[0131] For example, the computer program 62 can be divided into one or more modules / units, which are stored in the memory 61 and executed by the processor 60 to complete the present application. The one or more modules / units can be a series of computer program instruction segments capable of completing a specific function, which are used to describe the execution process of the computer program 62 in the electronic device 6.

[0132] The electronic device 6 can include, but is not limited to, the processor 60 and the memory 61. Those skilled in the art can understand that Figure 6 The electronic device 6 is only an example and does not constitute a limitation on the electronic device 6, and can include more or fewer components than those shown, or combine certain components, or different components, for example, the electronic device 6 can also include an input / output device, a network access device, a bus, etc.

[0133] For the convenience and brevity of description, only the above-mentioned division of functional modules / units is taken as an example for illustration. In actual application, the above-mentioned functions can be completed by different functional modules / units according to needs. The above-mentioned modules / units can be realized in the form of hardware, software or a combination of hardware and software.

[0134] In the above embodiments, the description of each embodiment has its own emphasis, and the parts not described or recorded in detail in a certain embodiment can be referred to the related description of other embodiments. If there is no special description and no logical conflict, the terms and / or descriptions of different embodiments are consistent and can be mutually referenced. The technical features in different embodiments can be combined to form a new embodiment according to their inherent logical relationship.

[0135] The above-described embodiments 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 embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A method for configuring an edge computing unit for power distribution network fault self-recovery, characterized in that, The method comprises: each candidate deployment node of the edge computing unit in the power distribution network is taken as a game player, a potential function is constructed based on the self-recovery benefit, operation cost and synergy benefit coefficient between adjacent game players of each game player, and a target function is constructed according to the potential function; the deployment quantity and deployment position of the edge computing unit in the power distribution network are taken as a particle, the fault self-recovery responsibility area of the power distribution network is divided according to the deployment quantity corresponding to each particle, and a responsibility area division result corresponding to each particle is obtained; the particle is optimized according to the responsibility area division result and the target function, an optimal particle that minimizes the target function is obtained, and the deployment quantity, deployment position and responsibility area division result corresponding to the optimal particle are determined as an edge computing unit configuration scheme under the fault self-recovery of the power distribution network.

2. The method of claim 1, wherein, The potential function is constructed based on the self-recovery benefit, operation cost and synergy benefit coefficient between adjacent game players of each game player, and comprises: a benefit function of each game player is constructed based on the self-recovery benefit, operation cost and synergy benefit coefficient between adjacent game players of each game player; a potential function is constructed according to the benefit function of each game player based on the potential game theory.

3. The method of claim 1, wherein, The target function is constructed according to the potential function, and comprises: the self-recovery time, power supply recovery range and self-recovery reliability of the power distribution network under different deployment schemes of the edge computing unit in the power distribution network are calculated; the self-recovery time, power supply recovery range and self-recovery reliability are all normalized to obtain normalized self-recovery time, normalized power supply recovery range and normalized self-recovery reliability; the target function is constructed according to the normalized self-recovery time, normalized power supply recovery range, normalized self-recovery reliability and potential function.

4. The method of claim 3, wherein, The self-recovery time of the power distribution network under different deployment schemes of the edge computing unit in the power distribution network is calculated, and comprises: the fault detection time, fault positioning time, isolation decision time, control instruction communication and issuing time and power supply recovery execution time of the power distribution network under different deployment schemes of the edge computing unit in the power distribution network are obtained; the sum of the fault detection time, fault positioning time, isolation decision time, control instruction communication and issuing time and power supply recovery execution time is calculated to obtain the self-recovery time of the power distribution network under the corresponding deployment scheme.

5. The method of claim 1, wherein, The fault self-recovery responsibility area of the power distribution network is divided according to the deployment quantity corresponding to each particle, and a responsibility area division result corresponding to each particle is obtained, and comprises: the topological structure of the power distribution network is modeled as a graph structure, an adjacency matrix and a degree matrix corresponding to the graph structure are calculated, and a graph Laplacian matrix is constructed according to the adjacency matrix and the degree matrix; the graph Laplacian matrix is spectrally decomposed to obtain the eigenvalue and eigenvector of the graph Laplacian matrix; obtaining a graph wavelet operator under each preset analysis scale based on a preset wavelet kernel function and each preset analysis scale, in combination with the characteristic value and the characteristic vector; calculating a response intensity of each graph wavelet operator acting on a unit signal centered on each node in the graph structure, denoted as a target characteristic value of each node under each preset analysis scale; constructing a feature matrix according to each target characteristic value; clustering all nodes in the power distribution network according to the feature matrix and the cluster number, and taking the clustering result as the responsibility area division result corresponding to each particle.

6. The method of claim 1, wherein, Before performing fault self-recovery responsibility area division on the power distribution network according to the deployment number corresponding to each particle to obtain the responsibility area division result corresponding to each particle, the method further includes: initializing a population based on chaotic mapping to obtain an initial population, the initial population including a plurality of particles.

7. The method of claim 6, wherein, optimizing the particles according to the responsibility area division result and the target function to obtain an optimal particle that minimizes the target function, including: taking the initial population as a current population, calculating the fitness value of each particle in the current population according to the responsibility area division result corresponding to each particle in the current population and the target function; determining an individual optimal particle and a global optimal particle of the current population according to the fitness value of each particle in the current population, and obtaining an adaptive inertia weight and an adaptive learning factor according to a current iteration number corresponding to the current population; determining whether the individual best particle and / or the global best particle is continuous N stag not improved; if the individual optimal particle and / or the global optimal particle are improved, updating each particle in the current population according to the individual optimal particle, the global optimal particle, the adaptive inertia weight, and the adaptive learning factor; if the individual optimal particle and / or the global optimal particle is not improved for N consecutive generations, updating each particle in the current population according to the individual optimal particle, the global optimal particle, the adaptive inertia weight, the adaptive learning factor, and a chaotic mapping sequence stag if the individual optimal particle and / or the global optimal particle is not improved for N consecutive generations, updating each particle in the current population according to the individual optimal particle, the global optimal particle, the adaptive inertia weight, the adaptive learning factor, and a chaotic mapping sequence taking the population obtained after the update as a new current population, and re-executing the step of calculating the fitness value of each particle in the current population according to the responsibility area division result corresponding to each particle in the current population and the target function, and subsequent steps; until the current iteration number reaches a preset iteration number, obtaining the optimal particle that minimizes the target function according to the global optimal particle corresponding to the current iteration number.

8. The method of claim 7, wherein, The adaptive inertia weight includes: calculating an inertia weight adjustment amount corresponding to each particle in the current population according to the fitness value of each particle in the current population, an average fitness value, a maximum fitness value, and a minimum fitness value of the current population; obtaining an adaptive inertia weight corresponding to each particle in the current population according to the inertia weight adjustment amount corresponding to each particle in the current population and the current iteration number corresponding to the current population.

9. An edge computing unit configuration apparatus for power distribution network fault self-recovery, characterized in that, including: the first processing module is configured to take each candidate deployment node of an edge computing unit deployed in a power distribution network as a game participant, construct a potential function based on the self-recovery benefit, operation cost, and synergy benefit coefficient between adjacent game participants of each game participant, and construct a target function according to the potential function; The second processing module is configured to divide the power distribution network into a responsibility area according to a deployment quantity of each particle corresponding to the deployment quantity, and obtain a responsibility area division result corresponding to each particle; The third processing module is configured to optimize the particle according to the responsibility area division result and the target function, obtain an optimal particle that minimizes the target function, and determine the deployment quantity, the deployment position, and the responsibility area division result corresponding to the optimal particle as an edge computing unit configuration scheme under the power distribution network fault self-recovery.

10. An electronic device, comprising: A computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the method according to any one of claims 1 to 8 when executing the computer program.

Images