Power distribution network risk assessment method for massive distributed resource access

Abstract

The invention discloses a power distribution network risk assessment method for massive distributed resource access, and the method comprises the following steps: dividing a risk domain according to the related information of a power distribution network, carrying out the stratified sampling of the risk domain, carrying out the Monte Carlo simulation, obtaining a sample matrix of risk indexes, inputting a random forest model, and calculating the importance score of the risk indexes, screening key risk indexes based on the importance threshold; taking the importance score as an objective weight, calculating a subjective weight by using an analytic hierarchy process, calculating a difference degree according to the subjective and objective weights, and fusing the subjective and objective weights to obtain a comprehensive weight; and obtaining a key risk index characteristic value in a to-be-evaluated scene, calculating the membership degree of the key risk index after adjusting the boundary parameter of the membership degree function according to the related information of the to-be-evaluated scene, calculating the comprehensive risk index according to the comprehensive weight and the membership degree of the key risk index, and judging the risk level. The method can fully consider the topological characteristics of the power distribution network, is suitable for the dynamic access of distributed resources, and achieves the precise risk assessment.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of new energy grid connection, and particularly relates to a power distribution network risk assessment method for massive distributed resource access. BACKGROUND

[0002] With the deepening of global energy transformation, the proportion of distributed energy (DER) such as solar energy, wind energy and small hydropower in the power system is rapidly increasing, and the traditional centralized power grid is gradually changing to a more flexible power distribution network structure. Distributed resources are considered as a key force to promote low-carbon economy and sustainable development, but their large-scale access also brings many new challenges to the operation of the power distribution network. The inherent volatility and uncertainty characteristics of distributed resources, such as solar and wind energy, are affected by multiple factors such as weather, season, time, etc., resulting in unstable power generation. In addition, their wide distribution and low power characteristics make it impossible to be centrally controlled like traditional centralized power generation, which significantly increases the difficulty of accurately assessing the operation risks that may be brought about by their access. SUMMARY

[0003] The technical problem to be solved by the present application is to provide a power distribution network risk assessment method for massive distributed resource access, which solves the above problems of the prior art by deeply integrating electrical topology information into the traditional risk assessment process to form an adaptive assessment system combining topology awareness and data-driven, which can fully consider the topology characteristics of the power distribution network, adapt to the dynamic access of distributed resources, and realize a new method of accurate risk assessment, providing scientific technical support for the safe and stable operation of the power distribution network with massive distributed resource access.

[0004] To solve the above technical problems, the technical scheme adopted by the present application is as follows: A power distribution network risk assessment method for massive distributed resource access, comprising the following steps: Obtain the power distribution network topology structure, DER access information and load distribution data, calculate the electrical distance matrix according to the power distribution network topology structure, DER access information and load distribution data, divide the risk domain according to the electrical distance matrix, select the probability distribution model corresponding to the preset risk index, stratified sampling of the risk domain and Monte Carlo simulation using the probability distribution model to obtain the sample matrix of the risk index; Input the sample matrix of the risk index into the random forest model, the random forest model calculates the comprehensive weight of the sample according to the risk propagation density index and the node criticality index, takes the comprehensive weight of all samples contained by the node as the weight of the node, calculates the importance score of each risk index according to the purity improvement value and the node weight when the node is split, and takes the risk index with an importance score greater than an importance threshold as a key risk index; The importance score of the key risk indicator is taken as an objective weight of each key risk indicator, a subjective weight of each key risk indicator is calculated by using the analytic hierarchy process, a difference degree is calculated according to the subjective weight and the objective weight of each key risk indicator, the subjective weight and the objective weight of each key risk indicator are fused according to the difference degree, and a comprehensive weight of each key risk indicator is obtained; The feature value of each key risk indicator in the to-be-evaluated scene is obtained, and the DER penetration rate and the average value of node criticality of the to-be-evaluated scene are obtained, the boundary parameter of the membership function is adjusted according to the DER penetration rate and the average value of node criticality, the membership of each key risk indicator to different risk levels is calculated according to the membership function after the boundary parameter is adjusted, and the comprehensive risk index is calculated according to the comprehensive weight and the membership of each key risk indicator, and the risk level is determined.

[0005] Further, the electrical distance matrix is an n*n symmetric matrix, n is the total number of nodes of the power distribution network, and the mathematical expression of the elements in the electrical distance matrix is as follows:

[0006] Among them, d ij is the electrical distance between node i and node j , R ij 、X ij are the equivalent resistance and reactance of node i to node j , P DER,i 、P DER,j are the DER installed capacity at node i and node j , P load,i 、P load,j are the load capacity at node i and node j ; and α is a DER penetration rate correction coefficient.

[0007] Further, when the risk propagation domain is divided according to the electrical distance matrix, the optimal clustering division scheme that minimizes the value of the clustering objective function is found by K-means algorithm iteration, and the clustering objective function is composed of the weighted sum of the square sum of the electrical distance from all nodes to the cluster center to which they belong and the DER output variance in each risk domain, and the mathematical expression is as follows:

[0008] Among them, K is the number of clusters;c k For the first k Risk domain Cluster centers, k∈[1, K ]; Represents the k-th risk domain From node i in the cluster to its cluster center c k The sum of squares of electrical distances; Var( P DER,k ) is the first k Variance of DER output within each domain; β As a balancing factor, cluster centers c k The calculation formula is:

[0009] Among them, | Z k |for the first k Risk domain The number of nodes within; Indicates the risk domain Z k Sum of risk indicator vectors for all nodes within the node; v i For nodes i The risk indicator vector, specifically the first element in the electrical distance matrix. i The vector of all electrical distances. Furthermore, when selecting a probability distribution model corresponding to a preset risk indicator, stratifying the risk domain, and using the probability distribution model for Monte Carlo simulation, the following steps are included: Based on the probability distribution model corresponding to each risk indicator, establish the load probability model and DER output probability model corresponding to each risk indicator; Traversing all risk domains, for the current risk domain, firstly, Latin hypercube sampling is used to generate a corresponding baseline random vector. Then, the DER output probability model is used to perform an inverse probability transformation to convert the baseline random vector into output samples. Finally, for nodes in the current risk domain, the correlation matrix C is used as the basis for the transformation. k The Cholesky decomposition is used to generate relevant random vectors. The load probability model is used to perform inverse probability transformation to convert the random vectors into load samples. Finally, power flow calculation is performed based on the output samples and load samples to obtain the risk index calculation results corresponding to the current risk domain. Record all risk indicator calculation results. If the number of simulations is less than the specified number, iterate through all risk domains again to obtain and record all risk indicator calculation results until the number of simulations reaches the specified number. A sample matrix is ​​constructed based on the calculation results of all risk indicators. The sample matrix includes... NGroups of simulated samples, each group containing m The values ​​of each risk indicator are set, where N is the specified number of times and m is the preset total number of risk indicators.

[0010] Furthermore, the mathematical expressions for the elements in the correlation matrix are as follows:

[0011] in, For risk domain Z k internal nodes i With nodes j The correlation, d ij For nodes i and nodes j The electrical distance between them, where σ is the correlation attenuation coefficient.

[0012] Furthermore, the mathematical expression for the overall weight of the samples is as follows:

[0013] in, For the first i The overall weight of each sample, where γ and λ are adjustment parameters. For the first i The risk transmission density index of each sample For the first i The node criticality index for each sample is expressed mathematically as follows:

[0014]

[0015] in, For the first i The neighboring nodes of the node containing each sample (neighbors) i ) nodes j DER installed capacity at the location d ij For the first i The node where each sample is located and the node j Electrical distance between them, BC i For the first i Betweenness centrality of the nodes containing each sample P DER,i For the first i DER installed capacity at the node where each sample is located P load,i For the first i The load capacity at the node where each sample is located It is the maximum value of betweenness centrality.

[0016] Furthermore, when calculating the importance score of each risk indicator based on the purity improvement value at the time of node splitting and the node weight, the following are included: Calculate the original importance vector for each risk indicator. I raw,m The mathematical expression is as follows:

[0017] in, w n For nodes in a random forest model n The weight of node n is the risk indicator used. m Split nodes; Δ impurity ( n ) is a node n The improvement in purity before and after splitting; The original importance vector of each risk indicator is normalized to obtain the importance score of each risk indicator, as shown in the following mathematical expression:

[0018] in, , I raw =[ I raw,1 , I raw,2 ,···, I raw,m ] , T The number of decision trees in the random forest model. The weighted importance of the m-th risk indicator is expressed mathematically as follows:

[0019] Among them, nodes t (m) In the decision tree of the random forest model, the first... t Using risk indicators in trees m All nodes that split.

[0020] Furthermore, the mathematical expression for the overall weight of each key risk indicator is as follows:

[0021] in, Indicates objective weighting. Indicates subjective weighting. The adaptive factor is expressed mathematically as follows:

[0022] wherein, is a specified coefficient, Δ0 is a difference threshold, and Δ is a difference, and the mathematical expression is as follows:

[0023] wherein, and respectively represent the objective weight and subjective weight of the jth key risk indicator, and m' is the total number of key risk indicators.

[0024] Further, when the boundary parameters of the membership function are adjusted according to the DER penetration rate and the average value of the node criticality, the mathematical expression is as follows:

[0025]

[0026] wherein, a 1, b 1 is the transition interval boundary between low risk and medium risk of the membership function, a 3, b 3 is the transition interval boundary between medium risk and high risk of the membership function, , is the adjusted transition interval boundary between low risk and medium risk, , is the adjusted transition interval boundary between medium risk and high risk, τ is the DER penetration rate influence coefficient, and ξ is the node criticality influence coefficient, is the average value of the node criticality, is the DER penetration rate.

[0027] Further, when the comprehensive risk index is calculated and the risk level is determined according to the comprehensive weight of each key risk indicator and the membership, the following steps are included: The membership calculation results of all key risk indicators are arranged in rows to form an m×3-dimensional fuzzy relationship matrix R, and each row of the matrix R represents the distribution of a key risk indicator at low, medium and high risk levels; The weighted average type fuzzy synthesis operator is used to perform fuzzy synthesis operation on the comprehensive weight of each key risk indicator and the fuzzy relationship matrix R to obtain a comprehensive evaluation vector, and the mathematical expression is as follows: B= [ b 1, b 2, b 3] wherein, , is the comprehensive weight of the jth key risk indicator, The membership of the jth key risk indicator for the kth risk level, k = 1 represents low risk, k = 2 represents medium risk, and k = 3 represents high risk; The comprehensive risk index is obtained by multiplying the quantitative value of the comment set and the comprehensive evaluation vector, and the mathematical expression is as follows:

[0028] Wherein, V represents the quantitative value of the comment set; The dynamic threshold is calculated according to the DER penetration rate and the average value of the node criticality, and the comprehensive risk index is compared with the dynamic threshold to determine the risk level, and the mathematical expression of the dynamic threshold is as follows:

[0029] Wherein, And The low-risk dynamic threshold and the high-risk dynamic threshold are respectively represented, if the comprehensive risk index is less than the low-risk dynamic threshold, the risk level is low risk, if the comprehensive risk index is greater than the high-risk dynamic threshold, the risk level is high risk, otherwise the risk level is medium risk.

[0030] Compared with the prior art, the advantages of the present application are as follows: First, the physical accuracy of risk simulation is improved. By constructing the electrical distance matrix and performing risk domain hierarchical division, the Monte Carlo simulation sample fully reflects the constraint effect of the distribution network topology on risk propagation and the spatial correlation of distributed resources, overcoming the problem of lack of spatial correlation in the traditional independent sampling method, and the simulation result is more consistent with the actual operation scene.

[0031] Second, the pertinence of risk indicator extraction is enhanced. By designing the sample weight mechanism of the risk propagation density index and the node criticality index, the importance of the risk indicators output by the random forest can truly reflect the physical propagation path of the risk in the power grid, accurately identify the key risk sources, and significantly improve the effectiveness of risk indicator screening compared with the standard random forest method.

[0032] Third, the scientificity of weight determination is improved. By defining the subjective and objective weight difference index to realize adaptive fusion of subjective and objective weights, both the data-driven objective statistical law and the expert's domain knowledge are utilized, the advantages of the two are complementary, and the limitations of pure subjective judgment are avoided.

[0033] Fourth, the dynamic adaptability of risk evaluation is realized. By establishing the quantitative mapping relationship between the boundary parameters of the membership function and the DER penetration rate and the node criticality, the risk level division standard can be dynamically adjusted with the system state, overcoming the hysteresis of the traditional fixed threshold method, and improving the timeliness and accuracy of the evaluation result. BRIEF DESCRIPTION OF DRAWINGS

[0034] Figure 1 Technical roadmap for the method of embodiments of the application.

[0035] Figure 2 Flowchart for the method of embodiments of the application.

[0036] Figure 3 Flowchart for the Monte Carlo simulation.

[0037] Figure 4 Flowchart for the hierarchical topology-aware Monte Carlo simulation based on the electrical distance matrix in embodiments of the application.

[0038] Figure 5 Flowchart for the calculation of the risk domain in embodiments of the application.

[0039] Figure 6 Detailed flowchart for the Monte Carlo simulation sampling in embodiments of the application.

[0040] Figure 7 Illustration of the low-risk membership function.

[0041] Figure 8 Illustration of the medium-risk membership function.

[0042] Figure 9 Illustration of the high-risk membership function. DETAILED DESCRIPTION

[0043] The application is further described below in conjunction with the accompanying drawings and specific preferred embodiments, but the scope of protection of the application is not limited thereby.

[0044] Currently, the design and control system of most power distribution networks is still based on the traditional centralized power system architecture, and the following technical bottlenecks are faced when dealing with massive distributed resource access: First, the existing Monte Carlo simulation method has a defect in modeling spatial correlation. The conventional Monte Carlo method adopts an independent random sampling strategy when dealing with the time-space coupling characteristics of distributed resources, and considers the power distribution network as a homogeneous system, without fully considering the influence of the power distribution network topology on the propagation of distributed resource fluctuations. This leads to a lack of spatial correlation in the simulation samples, making it difficult to accurately depict the amplification effect of distributed resource cluster effect on power grid risk, and resulting in a deviation between the risk assessment results and the actual operating conditions.

[0045] Second, the standard random forest algorithm lacks topological awareness. The traditional random forest treats all risk indicators and samples equally when assessing the importance of risk indicators, without considering the spatial relationship between samples, and fails to reflect the influence of electrical topological distance on risk propagation in distributed resource access scenarios. This leads to a mismatch between the importance ranking of risk indicators and the actual risk propagation path, making it difficult to accurately identify key risk sources and provide scientific basis for targeted risk control measures.

[0046] Third, the conventional AHP method is highly subjective and lacks dynamic adaptability. The traditional AHP method relies entirely on expert subjective judgment to construct the judgment matrix, and in the face of the emerging scenario of massive distributed resource access, it lacks objective data support, leading to a possible deviation of weight allocation from the actual risk distribution. In addition, as the penetration rate of distributed resources continues to change, the importance of risk indicators also evolves dynamically, and existing methods are difficult to automatically adapt to such changes, requiring frequent manual adjustments.

[0047] Fourth, the existing fuzzy comprehensive evaluation method uses fixed evaluation criteria. The traditional fuzzy comprehensive evaluation method uses fixed membership functions and evaluation matrices, which fail to reflect the influence of distributed resource access location, capacity, and timing characteristics on risk grade division boundaries. As the penetration rate of DERs in the power grid increases and node functions are dynamically adjusted, the fixed threshold leads to evaluation criteria that lag behind the actual state of the system, which may result in "misjudging low risk" or "over-conservatism", making the evaluation results lack of pertinence and timeliness.

[0048] To solve the above problems, the embodiment proposes a power distribution network risk assessment method for massive distributed resource access, as shown in Figure 1 The electrical topological information is systematically integrated into the whole process of risk assessment, forming an adaptive evaluation system combining topological awareness and data-driven, which can fully consider the electrical topological characteristics of the power distribution network and the spatio-temporal coupling characteristics of distributed resource access, and realize accurate risk assessment of the power distribution network under the scenario of massive distributed resource access.

[0049] As shown in Figure 2 The method includes the following steps: S1) Obtain the power distribution network topology structure, DER access information and load distribution data, calculate the electrical distance matrix according to the power distribution network topology structure, DER access information and load distribution data, divide the risk domain according to the electrical distance matrix, select the probability distribution model corresponding to the preset risk indicator, and perform Monte Carlo simulation on the risk domain using the probability distribution model to obtain the sample matrix of the risk indicator; S2) input the sample matrix of risk indicators into a random forest model, calculate the comprehensive weight of the sample according to the risk propagation density index and the node criticality index, take the comprehensive weight of all samples contained by the node as the weight of the node, calculate the importance score of each risk indicator according to the purity improvement value when the node is split and the node weight, and take the risk indicator with an importance score greater than an importance threshold as a key risk indicator; S3) take the importance score of the key risk indicator as the objective weight of each key risk indicator, calculate the subjective weight of each key risk indicator using the analytic hierarchy process, calculate the difference degree according to the subjective weight and the objective weight of each key risk indicator, fuse the subjective weight and the objective weight of each key risk indicator according to the difference degree, and obtain the comprehensive weight of each key risk indicator; S4) obtain the characteristic value of each key risk indicator under the scene to be evaluated, and the DER penetration rate and the average node criticality of the scene to be evaluated, adjust the boundary parameter of the membership function according to the DER penetration rate and the average node criticality, calculate the membership of each key risk indicator to different risk levels according to the membership function after adjusting the boundary parameter, and calculate the comprehensive risk index and determine the risk level according to the comprehensive weight and the membership of each key risk indicator.

[0050] The steps are described below.

[0051] In step S1 of the embodiment, the preset risk indicators are obtained by systematic literature research, policy interpretation and historical data analysis, and all kinds of risk factors faced by the distributed resource access distribution network are comprehensively sorted out. From the equipment level (transformer overload, line fault, protection misoperation, etc.), system level (voltage limit, frequency deviation, power imbalance, etc.), and environmental level (meteorological factors, load fluctuation, new energy output uncertainty, etc.), multidimensional identification is performed, and an m-risk indicator system covering internal and external factors is established, laying a foundation for subsequent risk indicator extraction and evaluation.

[0052] Step S1 of the embodiment is based on the distributed resource access distribution network scene, and a large number of sample data are constructed by using Monte Carlo method for multiple random sampling simulation to generate a large number of scenes. As shown in Figure 3 The traditional Monte Carlo simulation regards the distribution network as a homogeneous system, ignoring the influence of DER access location and electrical topology on risk propagation.

[0053] In order to solve the defects of the traditional Monte Carlo simulation method, step S1 of the embodiment proposes a hierarchical topology-aware mechanism based on an electrical distance matrix, which divides the distribution network into different risk propagation domains, so that the simulation samples are more consistent with the real distributed resource access distribution network risk coupling characteristics, and the process is as follows Figure 4As shown, based on the hierarchical topology-aware Monte Carlo simulation technology of electrical distance matrix, the electrical distance matrix considering the resistance, reactance, DER installed capacity and load capacity between nodes is constructed, the improved K-means clustering algorithm is used to divide the distribution network into several risk propagation domains, Latin hypercube sampling is used between domains, and correlation matrix is used to generate correlation random vectors within the domain, so that the simulation sample conforms to the real risk coupling characteristics, which includes the following steps: S11) Obtain input data, including: Distribution network topology: node connection relationship, line parameters (R, X); DER access information: access node location, rated capacity, output characteristics; Load distribution: load node location, load curve; Historical operation data: fault record, voltage monitoring data.

[0054] S12) Construct the electrical distance matrix: According to the distribution network topology, DER access information and load distribution data, the electrical distance between node i and node j is defined d ij is: (1) In the formula: d ij is the electrical distance between node i and node j , R ij 、X ij and i are the equivalent resistance and reactance from node j to node P , DER,i 、P DER,j and i are the DER installed capacity at node j ; P load,i 、P load,j and i are the load capacity at node j ; α is the DER penetration rate correction coefficient, and the value range is [0.2, 0.5].

[0055] Therefore, the electrical distance matrix construction process is as follows: 1) Based on the network topology, the shortest electrical path between any two nodes is calculated by using Dijkstra algorithm; 2) Calculate the electrical distance matrix D between nodes in the whole network: dij ] n×n The electrical distance matrix is ​​an n×n symmetric matrix, where n is the total number of nodes in the distribution network. The elements in the electrical distance matrix represent the electrical distances between any two nodes in the distribution network topology. d ij ; 3) Normalize the electrical distance matrix.

[0056] S13) Risk propagation domain hierarchical division: In this embodiment, when dividing the risk propagation domain according to the electrical distance matrix, the optimal clustering partitioning scheme that minimizes the value of the clustering objective function is specifically found through iterative searching using an improved K-means algorithm. The calculation process is as follows: Figure 5 As shown, the clustering objective function consists of the weighted sum of the squared electrical distances from all nodes to their respective cluster centers and the variance of DER output within each risk domain. The mathematical expression is as follows: (2) in, K The number of clusters (determined based on network size and DER distribution density, recommended) is... K =3~5); c k For the first k Risk domain Cluster centers, k∈[1, K ]; Represents the k-th risk domain From node i in the cluster to its cluster center c k The sum of squares of electrical distances; Var( P DER,k ) is the first k Variance of DER output within each domain; β This is a balancing factor, with a reference value of 0.3. Among them, the cluster centers... c k The calculation formula is: (3) Among them, | Z k |for the first k Risk domain The number of nodes within; Indicates the risk domain Z k Sum of risk indicator vectors for all nodes within the node; v i For nodes i The risk indicator vector, specifically the first element in the electrical distance matrix. i Row-represented nodes i A vector of all electrical distances to all other nodes. The optimal clustering scheme that makes the value of the clustering objective function minimum is finally output by iterative searching through the improved K-means algorithm K risk domains Z 1 ,Z 2 ,...,Z K} since the objective function J By considering the electrical distance and DER output fluctuation, the optimal clustering scheme that makes the value of the clustering objective function minimum is found through iterative searching by the K-means algorithm, which can ensure that the nodes in the same risk domain are strongly electrically coupled and have similar DER characteristics. J

[0057] S14) Hierarchical topology-aware Monte Carlo simulation: The Monte Carlo method is to calculate the probability or expected value of an event according to the probability distribution of random sample variables and sample space, and to use the probability or expected value as the probability or expected value of the event sought. Independent and identically distributed assumption is adopted for random sampling of all nodes, and spatial correlation in the risk domain is ignored. In order to solve the problems existing in the traditional Monte Carlo method, the application introduces a hierarchical topology-aware mechanism of electrical distance in the Monte Carlo simulation, realizes accurate division of the risk propagation domain, better simulates the operating state of distributed resources under different environmental conditions, and generates a large amount of simulation sample data, so that the simulation sample is more in line with the real risk coupling characteristics. The following improvements are mainly made: (1) In-domain correlation modeling: For the nodes in the same risk domain Z k , a spatial correlation matrix C k is introduced to model the correlation of nodes in the domain during Monte Carlo simulation. The mathematical expression is: (4) In the formula: σ is the correlation decay coefficient; i, j is the node number.

[0058] (2) Hierarchical sampling strategy: Inter-domain: independent sampling of each risk domain, Latin hypercube sampling (LHS) is used to ensure uniform coverage of the sample space.

[0059] In-domain: based on the correlation matrix C k , a correlated random vector is generated by Cholesky decomposition.

[0060] (3) Probability distribution: In order to simulate the operating scenario of the real power grid, different probability distributions are used to describe each risk index variable in this embodiment. Common probability distributions include: ​1) Beta distribution is a probability distribution that describes a family of curves on the interval (0, 1). It has two shape parameters a and b, and both parameters must be greater than 0. In MATLAB, you can generate random numbers from a Beta distribution using the betarnd function. The basic usage is: R = betarnd(A, B) (5) This function takes shape parameters a and b and returns a random number generated from a Beta distribution.

[0061] 2) The gamrnd function is used to generate random numbers from a gamma distribution with specified shape and scale parameters. The gamma distribution is a continuous probability distribution commonly used to model waiting time or lifetime data. The basic syntax is: r = gamrnd(a, b) (6) This syntax generates a random number from a gamma distribution with shape parameter a and scale parameter b.

[0062] 3) Exponential distribution is used to describe the time interval between events. MATLAB provides various functions to handle exponential distribution, including generating random numbers, calculating probability density function, etc.

[0063] By selecting appropriate probability distribution models for each risk indicator, we can obtain sample data that reflects the real operation of the power grid under different scenarios.

[0064] In summary, selecting the probability distribution model corresponding to the preset risk indicator, stratified sampling the risk domain, and using the probability distribution model for Monte Carlo simulation, the specific sampling process is as shown in Figure 6 , including the following steps: Establish the load probability model and DER output probability model corresponding to each risk indicator according to the probability distribution model corresponding to each risk indicator; Traverse all risk domains, first generate the corresponding benchmark random vector using Latin hypercube sampling for the current risk domain, then use the DER output probability model to perform probability inverse transformation to convert the benchmark random vector into output samples, and for the nodes in the current risk domain, based on the correlation matrix C k , generate the relevant random vector using Cholesky decomposition, perform probability inverse transformation on the random vector using the load probability model to convert it into load samples, and finally perform power flow calculation based on the output samples and load samples to obtain the risk indicator calculation results corresponding to the current risk domain; Record all risk indicator calculation results, if the simulation times are less than the specified times, traverse all risk domains again to obtain and record all risk indicator calculation results until the simulation times reach the specified times; According to all risk indicator calculation results, establish a sample matrix X ∈ RN×m , the sample matrix comprises N groups of simulation samples, each group of samples containing m values of the risk indicators, N is a specified number of times, and m is a preset total number of risk indicators.

[0065] Step S2 of this embodiment uses a random forest model to screen key risk indicators. Random forest is a machine learning method based on decision tree integration, which is widely used in classification and regression problems. The core idea of random forest is to construct multiple decision trees and use the combined prediction results of these decision trees to improve model accuracy and avoid overfitting. However, the random forest algorithm treats input data as independent risk indicator vectors and does not consider the spatial relationship between samples at all. However, in a power distribution network, nodes are not independent - they are coupled through electrical lines, and risks can propagate along electrical paths.

[0066] To solve the problem of the traditional random forest model, step S2 of this embodiment proposes an electrical topology weighted random forest risk indicator extraction method, which embeds electrical distance and risk domain information into the training process of the random forest, designs a multiplication weight model of risk propagation density index (RDI) and node criticality index (NCI) to incorporate sample weights, and multiplies the purity improvement value and node weight when splitting the decision tree node. Not only does it learn the statistical law of data, but more importantly, it learns the physical propagation law of risk in the electrical network, so that the importance ranking of risk indicators can reflect the actual propagation path of risk in the power grid, thereby screening key risk indicators. Specifically, the following steps are included: S21) Obtain input data, including: a sample matrix X ∈ R N×m generated by Monte Carlo simulation; corresponding risk labels (labeled by expert scoring or historical failure data): y ∈ R N×1 ; risk domain division results: { Z 1 ,Z 2 ,...,Z K}; electrical distance matrix: D.

[0067] S22) Topology information is incorporated into sample weights: give each training sample a topological-based first i sample comprehensive weight w i , which is multiplied by two independent factors to describe the risk importance of the node from different dimensions. In this embodiment, the mathematical expression of the comprehensive weight of the sample is as follows: (7) in, For the first i The overall weight of each sample, with γ and λ being adjustment parameters, and it is recommended to take values ​​of γ=2 and λ=0.5; For the first i Risk transmission density index of each sample For the first i The node criticality index of each sample. Samples with high RDI correspond to areas with dense risk propagation, while samples with high NCI correspond to topologically critical nodes. These samples are given higher weights during training.

[0068] RDI (Risk Propagation Density): a spatial dimension describing the "risk level of the surrounding environment"; NCI (Node Criticality): a structural dimension describing the "system position of the node itself". The overall weighting of the samples uses the multiplication of these two orthogonal dimensions rather than their addition, meaning neither is dispensable—even if a node is surrounded by dense DERs (high RDI), if it is located at the network edge (low NCI), its overall risk is still limited; and vice versa. This multiplicative structure better reflects the non-linear superposition effect of risk than an additive structure.

[0069] In this embodiment, the first i Risk transmission density index (RDI) of each sample i The Sigmoid function is used to handle this, and its mathematical expression is as follows: (8) in, For the first i The neighboring nodes of the node containing each sample (neighbors) i ) nodes j DER installed capacity at the location d ij For the first i The node where each sample is located and the node j The electrical distance between them, the formula reflects the sample i The degree of clustering and electrical coupling strength of DERs around the node, sample i neighbor nodes ( i Only includes the first i Each sample node is directly connected to the node via a line.

[0070] In this embodiment, the node criticality index (NCI) of the i-th sample i Using a linear function, the mathematical expression is as follows: (9) Among them, BC i For the first iThe betweenness centrality of the nodes containing each sample characterizes the degree of keyness of the node in the topology. P DER,i For the first i DER installed capacity at the node where each sample is located P load,i For the first i The load capacity at the node where each sample is located It is the maximum value of betweenness centrality.

[0071] S23) Calculate the importance of the weighted risk index: The traditional random forest risk indicator importance calculation process is as follows: Traverse all trees, find all nodes split by the risk indicator in each tree, sum the purity decreases of these nodes, divide by the number of trees, and take the average. For the m-th risk indicator, its weighted importance calculation formula is: (10) In the formula: T The number of nodes in the decision tree. t(m) For the first t Indicators used in trees m All nodes that split; Δ impurity ( n ) is a node n The improvement in purity before and after splitting. Related concepts are explained below: (1) Node splitting: in each internal node n The algorithm needs to select a risk indicator to "segment" the samples. For example, the risk indicator "voltage deviation" can be selected. A threshold is set: voltage deviation < 0.03, left child node: samples with voltage deviation < 0.03; right child node: samples with voltage deviation ≥ 0.03.

[0072] (2) Purity measures the "disorder" of samples within a node. The mathematical definition of purity is: (11) In the formula: P r It is a node n The Middle r The proportion of class samples.

[0073] The core idea behind the basic calculation logic of the importance of risk indicators is that the more times a risk indicator is used for splitting in a decision tree, and the greater the decrease in purity caused by each split, the more important this risk indicator becomes.

[0074] The traditional method does not distinguish the importance of the nodes. The same contribution is made in splitting the key nodes (high weight samples) and in splitting the edge nodes (low weight samples). To solve the problems existing in the traditional method, the embodiment calculates the weighted risk indicator importance by multiplying the purity decrease value of the node by the weight of the node, and the mathematical expression is as follows: (12) wherein the weight of the node n is equal to the sum of the weights of all samples contained by the node, and the expression is as follows: (13) wherein, w n is the weight of the node n w i is the weight of the i-th sample; samples( i ) is the sample set contained by the node n If a risk indicator is frequently used for splitting in a high weight node (key area), the importance of the risk indicator will be significantly improved. This reflects the key role of the risk indicator in the risk propagation path. n

[0075] S24) Normalization of risk indicator importance: In the embodiment, when the importance score of each risk indicator is calculated according to the purity improvement value and the weight of the node when the node is split, the original importance vector I raw,m obtained after the traversal and accumulation of each tree is normalized in two steps, including: (1) The original importance vector I raw,m of each risk indicator is calculated, and the mathematical expression is as follows: (14) wherein, w n is the weight of the node n in the random forest model, the node n is the node split using the risk indicator m ; Δ impurity ( n ) is the purity improvement value before and after the node n is split; and thus the original importance vector I raw of all risk indicators is obtained as follows: I raw,1 , I raw,2 ,···, I raw,m ]. ​​

[0076] (2) Normalize the original importance vector of each risk indicator to obtain the importance score of each risk indicator. The process is as follows: First, take the average: (15) Normalize back to [0,1]: (16) S25) Output result: The importance of normalized risk indicators is ranked as follows: I =[ I 1, I 2,..., I m ],in I m Indicates the first m The importance score of each risk indicator is calculated, and then risk indicators with an importance score greater than the importance threshold are selected as key risk indicators.

[0077] In step S3 of this embodiment, the key risk indicators identified in step S2 are transformed into a quantifiable and operable risk assessment indicator system. The weight of each indicator is determined, and the weight of the normalized risk indicator importance is obtained as 1, thus forming a scientific and complete risk assessment system.

[0078] Traditional Analytic Hierarchy Process (AHP) relies entirely on expert subjective judgment and is susceptible to cognitive biases; while purely objective weighting methods (such as entropy weighting) ignore professional knowledge. To overcome the shortcomings of traditional methods, step S3 of this embodiment proposes an adaptive weight fusion mechanism based on the degree of difference. This mechanism uses the importance of the risk indicators output by the random forest as objective weights, combines them with subjective weights calculated by the AHP, defines a difference index between subjective and objective weights, and constructs an adaptive fusion factor using the Sigmoid function. The fusion ratio is dynamically adjusted according to the degree of difference between the subjective and objective weights. Specifically, the following steps are included: S31) Obtain input data, including: Importance vector of risk metrics output by random forest: I’ =[ I 1, I 2,..., I m’ (m' represents the number of key risk indicators after screening). Expert judgment matrix: .

[0079] S32) Objective weight calculation: Directly use the importance of the risk indicators output by the random forest as objective weights: w objI’ .

[0080] S33) Subjective weight calculation: Subjective weight is calculated by using AHP w sub , including the following four steps: (1) Establish a hierarchical structure model: Analyze the relationship between the index factors in the system and establish the hierarchical structure of the system. In order to accurately evaluate the operation risk level caused by the access of massive distributed resources to the distribution network, a risk index system containing m' key indicators is constructed.

[0081] (2) Construct a comparison and discrimination matrix: After the hierarchical structure is established, the evaluator determines the importance of each layer of different factors relative to the factors of the previous layer based on his own knowledge, experience and analysis, starting from the first criterion layer and gradually moving down. The weight of the latter relative to the former can be calculated when a two-by-two comparison matrix is constructed. When the factor of the previous level is used as the comparison criterion, a comparison scale a ij can be used to express the relative importance of the i-th index factor and the j-th index factor in the next level. In this embodiment, the AHP 1-9 scale method is used, as shown in Table 1: Table 1 AHP 1-9 scale table Scale Meaning 1 Indicator i and indicator j are of equal importance in comparison to each other 3 Indicator i is slightly more important than indicator j in comparison to each other 5 Indicator i is significantly more important than indicator j in comparison to each other 7 Indicator i is strongly more important than indicator j in comparison to each other 9 Indicator i is extremely more important than indicator j in comparison to each other 2、4、6、8 Importance is somewhere between the two scales above Reciprocal Importance of two indicators is reversed in comparison to each other After the membership and dominance relationship between the upper and lower levels of indicators is established, the relative importance of each layer of factors is determined and expressed in data to quantitatively describe the judgment matrix of the hierarchical structure. (3) Weight calculation and normalization of each layer: The relative weight of the compared elements with respect to the criterion is calculated from the judgment matrix, and consistency check is performed.

[0082] There are three different methods for weight calculation, which are arithmetic mean method, geometric mean method and risk index value method. In this embodiment, the risk index value and risk index vector of each judgment matrix under a single criterion are calculated by using the risk index value method, mainly using the formula to calculate the value, where represents the judgment matrix, represents the maximum risk index value of , and represents the corresponding risk index vector, each element in is represented by, that is, the weight value of the single level ordering. The specific calculation steps are as follows: a) Normalize the judgment matrix A: (17) Obtain the normalized matrix After normalization, the sum of the elements in each column of B is 1.

[0083] b) Add the elements of matrix B row by row to obtain the vector. ,in: (18) c) Normalize the vector V to obtain the weight vector. ,in: (19) d) Calculate the maximum risk index value using the following formula: (20) In the above formula express The i-th element, where n is the order.

[0084] (4) Consistency check Because of the complexity of the target problem and the ambiguity and diversity of people's understanding of the problem, the judgment matrices given by different people may not be completely identical. Therefore, consistency testing is the basis for the objective accuracy of the judgment results. Consistency testing is one of the necessary steps in the analytic hierarchy process (AHP). Passing the consistency test is a prerequisite for the calculated weights to be meaningful; otherwise, the results will not fully reflect the true weights of the indicators. If the consistency test fails, the data must be corrected, a new set of weights calculated, and the test repeated until the consistency test is passed; otherwise, the above process is repeated.

[0085] The consistency index is as follows: (twenty one) In the formula, n is the order of the judgment matrix. When the consistency ratio is satisfied... The consistency test of the judgment matrix is ​​performed; otherwise, it indicates that the evaluation of the indicators in the matrix is ​​unreasonable, and the judgment matrix data needs to be corrected until the consistency test is passed. The mean value of the consistency indicator RI is determined according to Table 2 below: Table 2. Mean values ​​of consistency index n 1 2 3 4 5 6 7 8 9 RI 0 0 0.52 0.89 1.12 1.26 1.36 1.41 1.46 S34) Differential Adaptive Fusion: Define the difference Δ between subjective and objective weights: (twenty two) in, and respectively represent the objective weight and subjective weight of the jth key risk indicator, and m' is the total number of key risk indicators.

[0086] Thus, the mathematical expression of the adaptive fusion weight is as follows: (23) wherein, represents the objective weight, represents the subjective weight, is an adaptive factor, and the adaptive factor (Δ) adopts a Sigmoid function, and the mathematical expression is as follows: (24) wherein, is a specified coefficient, Δ0 is a difference threshold, and the parameters are set as follows: = 10: control the steepness of the curve; Δ0 = 0.3: difference threshold.

[0087] S35) Output the result: the comprehensive weight vector , which satisfies that the sum is 1.

[0088] Step S4 of the embodiment realizes risk assessment by calculating the comprehensive risk index of the scene to be evaluated based on the comprehensive weight of the key risk indicator.

[0089] The traditional fuzzy comprehensive evaluation adopts a fixed risk level division threshold. In the process of power grid evolution, the proportion of DER access continues to grow, and the node function is dynamically adjusted. The fixed threshold leads to that the evaluation standard lags behind the actual state of the system, and the problems of "misjudgment of low risk" or "excessive conservatism" may occur. There is an urgent need for a risk evaluation method that can adaptively adjust the evaluation standard, so that the evaluation result not only reflects the real-time state of the system, but also considers the influence of the topology characteristics.

[0090] To solve the problems existing in the traditional method, step S4 of the embodiment proposes a dynamic threshold fuzzy comprehensive evaluation method based on DER penetration rate and node criticality. By dynamically adjusting the boundary parameters of the membership function to construct an adaptive membership function, a quantitative mapping relationship between the boundary parameters of the membership function and the system characteristics is established, the adaptive adjustment of the risk evaluation standard is realized, the weighted average type fuzzy synthesis operator is used to calculate the comprehensive risk index and determine the risk level. Specifically, the following steps are included: S41) Obtain the input data, including: the comprehensive weight vector: ; the characteristic value of the risk indicator of the scene to be evaluated: x [ x 1, x 2,...,x m ]; DER penetration of the scenario: ; Average node criticality of the scenario: .

[0091] S42) Dynamic threshold adjustment: Traditional fuzzy evaluation uses fixed criteria: V1 low risk (V<30); V2 medium risk (30≤V≤70); V3 high risk (V>70). This embodiment proposes a coupling adjustment mechanism for DER penetration and node criticality, achieving the synergistic effect of the two factors through multiplication, avoiding the imbalance of adjustment amplitude caused by simple linear superposition. The complete process is as follows: Step 1: According to the DER penetration and node criticality of the scenario, calculate the dynamic boundary parameter, realize the dynamic adjustment of the membership function boundary parameter: In this embodiment, when adjusting the boundary parameter of the membership function according to the DER penetration and the average node criticality, the mathematical expression is as follows: The tightening adjustment formula of the low-risk and medium-risk boundary: (25) The relaxation adjustment formula of the high-risk boundary: (26) Wherein, a 1, b 1 is the transition interval boundary of the low-risk and medium-risk of the membership function (in this embodiment, the values are 3%, 5%), a 3, b 3 is the transition interval boundary of the medium-risk and high-risk of the membership function (in this embodiment, the values are 7%, 10%), so the medium-risk interval of this embodiment is [ a 1, b 3] = [3%, 10%], wherein [5%, 7%] is completely medium risk; , is the adjusted transition interval boundary of the low-risk and medium-risk, , is the adjusted transition interval boundary of the medium-risk and high-risk, τ is the DER penetration influence coefficient, controlling the strength of penetration rate on threshold adjustment, the reference value is 0.3; ξ is the node criticality influence coefficient, controlling the strength of node position on threshold adjustment, the reference value is 0.2; is the average node criticality, quantitatively evaluating the average importance of the nodes involved in the scenario, the value range is [0, 1]; DER penetration is the ratio of distributed energy to total installed capacity, with a range of [0, 1].

[0092] Step 2: Calculate the membership degree of each key risk indicator: The membership degree of each key risk indicator is calculated using the membership function after adjusting the boundary parameters, and the mathematical expression is as follows: (27) Where x j is the eigenvalue of the jth key risk indicator.

[0093] 1) Low risk membership function f low (Z-shaped function: from the left, to maintain 1, then linearly decrease to become 0) The specific graph is shown in Figure 7 : x j ≤ : completely belongs to low risk (membership = 1), μv 1( x j )=1; < x j < : linear transition from low risk; x j ≥ : completely not belong to low risk (membership = 0), μv 1( x j )=0.

[0094] 2) Medium risk membership function f medium (trapezoidal function: linearly rising from 0 on the left, with a platform of 1 in the middle, and linearly descending to 0 on the right) The specific graph is shown in Figure 8 : x j ≤ : completely not belong to medium risk (membership = 0), μv 2( x j )=0; < x j < : Linear transition to medium risk; ≤ x j ≤ : Fully belongs to medium risk (membership = 1), μv 2( x j )=1; < x j < : Linear transition to medium risk; x j ≥ : Fully does not belong to low risk (membership = 0), μv 2( x j )=0 3) High risk membership function f high (S-shaped function: starts from 0 on the left, rises linearly to a 3 ’ starts, reaches 1 and remains) The specific graph is shown in b 3 ’ : Figure 9 x j ≤ : Fully does not belong to high risk (membership = 0), μv 3( x j )=0; < x j < : Linear transition to high risk; x j ≥ : Fully belongs to high risk (membership = 1), μv 3( x j )=1.

[0095] Next, according to the comprehensive weight and membership of each key risk indicator, the comprehensive risk index is calculated and the risk level is determined, including the following steps: Step 3: Build fuzzy relationship matrix: ​Arrange the membership degree calculation results of all key risk indicators obtained in step 2 in rows to form an m×3 fuzzy relation matrix R. Each row of matrix R represents the distribution of a key risk indicator at the low, medium and high risk levels, reflecting the fuzzy characteristics of the current state of the indicator.

[0096] Step 4: Fuzzy comprehensive evaluation calculation: A weighted average fuzzy synthesis operator is used to integrate the weights of each key risk indicator. A fuzzy synthesis operation is performed with the fuzzy relation matrix R to obtain the comprehensive evaluation vector B. The three components of the comprehensive evaluation vector B represent the overall degree to which the evaluated object belongs to low, medium, and high risk, respectively. The mathematical expressions are as follows: B=w 综合 ·R=[ b 1, b 2, b 3] (28) in, , The comprehensive weight of the j-th key risk indicator. Let represent the membership degree of the j-th key risk indicator to the k-th risk level, where k=1 indicates low risk, k=2 indicates medium risk, and k=3 indicates high risk.

[0097] Step 5: Calculate the overall risk index: The comprehensive risk index is obtained by multiplying the quantitative value of the comment set with the comprehensive evaluation vector. In this embodiment, the quantitative value of the comment set V=[15,50,85] (representing representative values ​​of low, medium, and high risk, respectively) is introduced, and the mathematical expression is as follows: (29) The value of RI ranges from [0, 100], with a higher value indicating a higher level of risk.

[0098] Step 6: Determine the risk level based on dynamic thresholds: A dynamic threshold is calculated based on the DER penetration rate and the average node criticality. The comprehensive risk index is compared with the dynamic threshold to determine the risk level. The mathematical expression for the dynamic threshold is as follows: (30) in, and These represent the low-risk dynamic threshold and the high-risk dynamic threshold, respectively. If the comprehensive risk index is less than the low-risk dynamic threshold, the risk level is low; if the comprehensive risk index is greater than the high-risk dynamic threshold, the risk level is high; otherwise, the risk level is medium.

[0099] Through dynamic threshold evaluation, differentiated risk control measures can be formulated for different DER penetration rates and node importance, improving resource allocation efficiency.

[0100] To sum up, the application proposes a power distribution network risk assessment method for massive distributed resource access, establishes a risk assessment model combining Monte Carlo simulation method and random forest algorithm, which can effectively assess the risk of distributed power supply access to the distribution network. Monte Carlo simulation provides a variety of possible operating scenarios, and through simulating the power grid risk under different situations; while the random forest algorithm helps to identify the key factors affecting the power grid risk through risk index importance analysis. Finally, based on the analytic hierarchy process-fuzzy comprehensive evaluation method, the evaluation index weight is determined, which can provide effective decision support for power grid managers and ensure the stability and reliability of the power grid. With the continuous increase of new energy proportion, the application can automatically adapt to system changes without manual parameter adjustment, and has good sustainability. Through contribution decomposition, the main risk factors are clearly identified, providing clear decision basis for operation personnel.

[0101] Those skilled in the art will appreciate that embodiments of the application can be provided as methods, systems, or computer program products. Accordingly, the application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the application can take the form of a computer program product implemented on one or more computer-readable storage media (including, but not limited to, disk memory, CD-ROMs, optical storage devices, etc.) containing computer usable program code. The application is described with reference to flowcharts and / or block diagrams illustrating the architecture, and / or computer program products according to embodiments of the application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, as well as combinations of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing apparatus to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing apparatus produce an apparatus that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in a flow or multiple flows and / or blocks Figure 1 The functions specified in a flow or multiple flows and / or blocks These computer program instructions can also be stored in a computer-readable memory that can direct the computer or other programmable data processing apparatus to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including instruction means, which implements the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in a flow or multiple flows and / or blocks Figure 1the functions specified in the flow or flows and / or blocks of the flow or flows. Such computer program instructions can also be loaded into a computer or other programmable data processing devices, so that a series of operational steps are generated to realize the computer-implemented processing, and the instructions executed on the computer or other programmable devices provide a process for realizing the functions specified in the flow or flows and / or blocks of the flow or flows. Figure 1 the functions specified in the flow or flows and / or blocks of the flow or flows. Such computer program instructions can also be loaded into a computer or other programmable data processing devices, so that a series of operational steps are generated to realize the computer-implemented processing, and the instructions executed on the computer or other programmable devices provide a process for realizing the functions specified in the flow or flows and / or blocks of the flow or flows. Figure 1 the functions specified in the flow or flows and / or blocks of the flow or flows. Such computer program instructions can also be loaded into a computer or other programmable data processing devices, so that a series of operational steps are generated to realize the computer-implemented processing, and the instructions executed on the computer or other programmable devices provide a process for realizing the functions specified in the flow or flows and / or blocks of the flow or flows.

[0102] The above description is only the preferred embodiments of the present application, and the protection scope of the present application is not limited to the above-mentioned embodiments. Any improvement and modification made by those skilled in the art without departing from the principles of the present application shall also be considered as falling within the protection scope of the present application.

Claims

1. A power distribution network risk assessment method for mass distributed resource access, characterized in that, Includes the following steps: The distribution network topology, DER access information, and load distribution data are obtained. The electrical distance matrix is ​​calculated based on the distribution network topology, DER access information, and load distribution data. The risk domain is divided according to the electrical distance matrix. The probability distribution model corresponding to the preset risk index is selected. The risk domain is sampled in layers and Monte Carlo simulation is performed using the probability distribution model to obtain the sample matrix of the risk index. The sample matrix of risk indicators is input into the random forest model. The random forest model calculates the comprehensive weight of the samples based on the risk propagation density index and the node criticality index. The comprehensive weight of all samples contained in a node is used as the weight of the node. The importance score of each risk indicator is calculated based on the purity improvement value when the node splits and the node weight. Risk indicators with an importance score greater than the importance threshold are designated as key risk indicators. The importance scores of key risk indicators are used as the objective weights of each key risk indicator, and the analytic hierarchy process is used to calculate the subjective weights of each key risk indicator. The degree of difference between the subjective and objective weights of each key risk indicator is calculated, and the subjective and objective weights of each key risk indicator are integrated based on the degree of difference to obtain the comprehensive weight of each key risk indicator. Obtain the characteristic values ​​of each key risk indicator in the scenario to be evaluated, as well as the DER penetration rate and the average node criticality of the scenario to be evaluated. Adjust the boundary parameters of the membership function based on the DER penetration rate and the average node criticality. Calculate the membership degree of each key risk indicator to different risk levels based on the membership function after adjusting the boundary parameters. Calculate the comprehensive risk index and determine the risk level based on the comprehensive weight and membership degree of each key risk indicator.

2. The power distribution network risk assessment method for mass distributed resource access according to claim 1, characterized in that, The electrical distance matrix is ​​an n×n symmetric matrix, where n is the total number of nodes in the distribution network. The mathematical expressions for the elements in the electrical distance matrix are as follows: wherein, d ij is the electrical distance between the nodes i and the node j , R ij 、X ij are the equivalent resistance and reactance of the nodes i to the node j , respectively; P DER,i 、P DER,j are the DER installed capacity at the nodes i and j , respectively; P load,i 、P load,j are the load capacity at the nodes i and j , respectively; and a is the DER penetration correction factor.

3. The power distribution network risk assessment method for mass distributed resource access according to claim 2, characterized in that, When dividing the risk propagation domain based on the electrical distance matrix, the optimal clustering scheme that minimizes the clustering objective function is found iteratively using the K-means algorithm. The clustering objective function is composed of the weighted sum of the squared electrical distances from all nodes to their respective cluster centers and the variance of DER output within each risk domain. The mathematical expression is as follows: in, K The number of clusters; c k For the first k Risk domain Cluster centers, k∈[1, K ]; Represents the k-th risk domain From node i in the cluster to its cluster center c k The sum of squares of electrical distances; Var( P DER,k ) is the first k Variance of DER output within each domain; β As a balancing factor, cluster centers c k The calculation formula is: wherein, Z k | is the number of nodes within the k risk domain ; represents the summation of risk indicator vectors of all nodes within the Z risk domain k v i is the risk indicator vector of node i , specifically the vector of all electrical distances in the electrical distance matrix in the i row.

4. The power distribution network risk assessment method for mass distributed resource access according to claim 1, characterized in that, When selecting a probability distribution model corresponding to a preset risk indicator, performing stratified sampling of the risk domain, and using the probability distribution model for Monte Carlo simulation, the following steps are included: Based on the probability distribution model corresponding to each risk indicator, establish the load probability model and DER output probability model corresponding to each risk indicator; For each risk domain, a benchmark random vector is generated by Latin hypercube sampling, and then the benchmark random vector is converted into an output sample by using the DER output probability model for probability inverse transformation. For the nodes in the current risk domain, a relevant random vector is generated by Cholesky decomposition based on the correlation matrix C k , and then the random vector is converted into a load sample by using the load probability model for probability inverse transformation. Finally, the risk index calculation results corresponding to the current risk domain are obtained by power flow calculation based on the output sample and the load sample. Record all risk indicator calculation results. If the number of simulations is less than the specified number, iterate through all risk domains again to obtain and record all risk indicator calculation results until the number of simulations reaches the specified number. A sample matrix is established according to the calculation results of all risk indicators, and the sample matrix comprises N Group simulation samples, each group of samples containing m Values of the m risk indicators, N is a specified number of times, and m is a preset total number of risk indicators.

5. The power distribution network risk assessment method for mass distributed resource access according to claim 4, characterized in that, The mathematical expressions for the elements in the correlation matrix are as follows: wherein, is a risk domain Z k an inner node i a correlation with a node j d ij is a node i and an electrical distance between the node j and the node, and σ is a correlation decay coefficient.​ 6. The power distribution network risk assessment method for mass distributed resource access according to claim 1, characterized in that, The mathematical expression for the overall weight of the samples is as follows: wherein, is the integrated weight of the i th sample, and γ, λ are the adjustment parameters, is the risk propagation density index of the i th sample, is the node criticality index of the i th sample, and the corresponding mathematical expression is as follows: wherein, is the DER installed capacity at the node where the i-th sample is located, i i is the DER installed capacity at the node where the i-th sample is located, j d ij is the electrical distance between the node where the i-th sample is located and the node i i j i is the betweenness centrality of the node where the i-th sample is located, P DER,i i is the DER installed capacity at the node where the i-th sample is located, P load,i i is the load capacity at the node where the i-th sample is located, is the betweenness centrality maximum value.​​​​​​​ 7. The power distribution network risk assessment method for mass distributed resource access according to claim 1, characterized in that, When calculating the importance score of each risk indicator based on the purity improvement value at the time of node splitting and the node weight, the following are included: calculating a raw importance vector for each risk indicator I raw,m The mathematical expression is as follows: in, w n For nodes in a random forest model n The weight of node n is the risk indicator used. m Split nodes; Δ impurity ( n ) is a node n The improvement in purity before and after splitting; The original importance vector of each risk indicator is normalized to obtain the importance score of each risk indicator, as shown in the following mathematical expression: wherein, , I raw [ I raw,1 , I raw,2 ,···, I raw,m ], T is the number of decision trees in the random forest model, is the weighted importance of the mth risk indicator, which is mathematically expressed as follows: wherein nodes t (m) for each tree in the random forest model t using the risk metric m for all nodes split in the decision tree.

8. The power distribution network risk assessment method for mass distributed resource access according to claim 1, characterized in that, The mathematical expression of the comprehensive weight of each key risk indicator is as follows: wherein, represents the objective weight, represents the subjective weight, is an adaptive factor, mathematically expressed as follows: wherein is a specified coefficient, Δ0is a threshold of difference degree, and Δ is a difference degree, and the mathematical expression is as follows: wherein, and respectively represent the objective weight and subjective weight of the jth key risk indicator, and m' is the total number of key risk indicators.

9. The power distribution network risk assessment method for mass distributed resource access according to claim 1, characterized in that, When the boundary parameters of the membership function are adjusted according to the DER penetration and the average value of the node criticality, the mathematical expression is as follows: wherein, a 1, b 1 is the transition interval boundary between low risk and medium risk of the membership function, a 3, b 3 is the transition interval boundary between medium risk and high risk of the membership function, , is the adjusted transition interval boundary between low risk and medium risk, , is the adjusted transition interval boundary between medium risk and high risk, τ is the DER penetration rate influence coefficient, and ξ is the node criticality influence coefficient, is the node criticality mean, is the DER penetration rate.

10. The power distribution network risk assessment method for mass distributed resource access according to claim 1, characterized in that, When the comprehensive risk index is calculated according to the comprehensive weight of each key risk indicator and the membership, and the risk level is determined, the following steps are included: The membership calculation results of all key risk indicators are arranged in rows to form a fuzzy relationship matrix R of mx3 dimensions. Each row of the matrix R represents the distribution of a key risk indicator at the low, medium and high risk levels. The fuzzy synthetic operation of the comprehensive weight of each key risk indicator and the fuzzy relationship matrix R is performed by using the weighted average type fuzzy synthetic operator to obtain a comprehensive evaluation vector. The mathematical expression is as follows: B= [ b 1, b 2, b 3] wherein, , is the comprehensive weight of the jth key risk indicator, is the membership degree of the jth key risk indicator for the kth risk level, k = 1 represents low risk, k = 2 represents medium risk, and k = 3 represents high risk. The comprehensive risk index is obtained by multiplying the quantitative value of the comment set and the comprehensive evaluation vector. The mathematical expression is as follows: Wherein, V represents the quantitative value of the comment set; The dynamic threshold is calculated according to the DER penetration and the average value of the node criticality. The comprehensive risk index is compared with the dynamic threshold to determine the risk level. The mathematical expression of the dynamic threshold is as follows: wherein, and respectively represent a low risk dynamic threshold and a high risk dynamic threshold, the risk level is low risk if the composite risk index is less than the low risk dynamic threshold, the risk level is high risk if the composite risk index is greater than the high risk dynamic threshold, and otherwise the risk level is medium risk.

Images