Power grid cost configuration method and system based on multi-modal graph structure and fuzzy evaluation

Abstract

The invention discloses a power grid cost configuration method and system based on a multi-modal diagram structure and fuzzy evaluation, and the method comprises the steps: firstly obtaining a preset cost index, and constructing a full-life-cycle model; an evaluation system is constructed based on cost indexes, and a multi-modal graph structure network is formed; extracting a multi-modal feature vector, endowing an edge with a dynamic weight, and calculating a dynamic weight vector; embedding a fuzzy reasoning matrix and performing defuzzification to obtain a comprehensive weight; and finally, carrying out weighted calculation on the full-life cycle model based on the comprehensive weight to obtain the full-life comprehensive cost of the power grid. According to the invention, through the full-life-cycle model and the dynamic fuzzy evaluation method based on the multi-modal graph neural network, the cost of each item of the power grid is predicted and configured, the weight of each index is dynamically allocated to the evaluation object, subjective prejudice is avoided, the correlation among the indexes is considered, the error is reduced, and the evaluation accuracy is improved. Meanwhile, the whole-life-cycle model covers the cost of each stage of the power grid from purchasing to scrapping, and comprehensive evaluation of the service life of the power grid is realized.

Description

Technical Field

[0001] The present invention relates to a power grid cost configuration method, belongs to the field of power grid cost optimization, and particularly relates to a power grid cost configuration method and system based on a multimodal graph structure and fuzzy evaluation. Background Art

[0002] Grid production cost allocation methods primarily involve economic analysis of power markets and grid operations. Specifically, they focus on how to rationally allocate and calculate the various costs of grid operations, including the construction, maintenance, and operation of transmission lines, substations, and other equipment, as well as costs associated with power market transactions and dispatch. With the gradual liberalization of the power market and intensified competition, grid companies need to manage and control costs more meticulously to improve their competitiveness and profitability. At the same time, grid modernization requires substantial investments in equipment upgrades and technological transformation, and these costs also need to be rationally allocated and calculated.

[0003] In the production cost configuration of power grids, existing technologies usually adopt unified standards and methods, failing to fully consider the characteristics, differences and future changes of power grids in different regions, resulting in unreasonable cost configuration and poor accuracy; on the other hand, existing technologies often only focus on the current production cost configuration of power grids, while ignoring factors that may change in the future, making the cost configuration lack foresight and flexibility, and poor reliability. Summary of the Invention

[0004] The purpose of the present invention is to overcome the above-mentioned defects and problems in the prior art and to provide a more accurate power grid cost configuration method and system based on multimodal graph structure and fuzzy evaluation.

[0005] To achieve the above objectives, the technical solution of the present invention is: a power grid cost configuration method based on multimodal graph structure and fuzzy evaluation, comprising:

[0006] S1. Obtain the preset cost indicators of the power grid throughout its life cycle and build a full life cycle model of the power grid;

[0007] S2. Build an evaluation system based on preset cost indicators, use the evaluation indicators in the evaluation system as nodes, and build a node set; calculate the mutual information based on the connection relationship between the nodes, establish edges between the nodes, and build an edge set; build a multimodal graph structure network based on the node set and edge set;

[0008] S3. Based on the multimodal graph structure network, extract the multimodal feature vector and assign dynamic weights to the established edges; obtain the attention weight to weight the importance of the node and obtain the dynamic weight vector in the current context;

[0009] S4. Based on the dynamic weight vector, the fuzzy reasoning matrix is ​​embedded in the multimodal graph structure network to optimize the fuzzy reasoning rule process. The fuzzy weight of the fuzzy reasoning is defuzzified by the Bayesian weighted mind method to obtain the comprehensive defuzzification index weight;

[0010] S5. Based on the weights of the comprehensive defuzzification indicators, a weighted calculation is performed on the full life cycle model to obtain the comprehensive cost of the full life cycle of the power grid.

[0011] In step S1, the preset cost indicators include investment and construction costs, operating costs, maintenance costs, reliability failure costs, decommissioning costs, and electricity costs;

[0012] The expression of the full life cycle model of the power grid is as follows:

[0013] LCC=CI+CO+CM+CF+CD+CP;

[0014] Where: CI is the investment and construction cost of the grid, CO is the operation cost of the grid, CM is the maintenance cost of the grid, CF is the reliability failure cost of the grid, CD is the decommissioning cost of the grid, and CP is the electricity cost of the grid;

[0015] CI=C s +C line +C si ;

[0016] Where: C s is the investment cost of transmission tower construction, C line is the investment cost of distribution line construction, C sj Design costs for construction;

[0017] CO=C sope +C gope +C zope ;

[0018] Where: C sope is the substation operating cost, C gope is the operating cost of high-voltage distribution lines, C zope The operating cost of medium voltage distribution lines;

[0019]

[0020] Among them: PM i is the unit price of preventive maintenance for equipment i, Q i is the number of installations of device i, λ PM，i is the average annual preventive maintenance frequency of equipment i, CM j is the repair cost of faulty component j, f j is the failure probability of faulty component j, τ jis the mean repair time, r is the discount rate, and t is the t-th year in the whole life cycle;

[0021]

[0022] Of which: EENS k is the expected value of power shortage under fault scenario k, VOLL is the unit power outage loss value, RC k is the emergency repair cost of fault scenario k;

[0023] CD=C d wC r β;

[0024] Where: C d The cost of decommissioning, C r is the original value of the equipment, and β is the residual value rate;

[0025]

[0026] Where: P grid (t) is the purchased electricity in period t, ρ(t) is the time-of-use electricity price curve, C gen is the cost of self-provided power generation, P local is the self-generated power.

[0027] The construction investment costs of transmission towers and distribution lines include:

[0028]

[0029] Where: C1 is the total investment cost of transmission tower, t ope is the equipment operation time, r0 is the discount rate;

[0030]

[0031] Among them: z is the construction investment cost per unit length of the distribution line, γ2 is the tortuosity coefficient of the distribution line, J q is the set of loads supplied by the qth substation, J is the set of all load points, and l(q, j) is the straight-line distance between the qth substation and the jth load point;

[0032] Substation operating costs, high-voltage distribution line operating costs, and medium-voltage distribution line operating costs, specifically including:

[0033]

[0034]

[0035] Where: ΔP M，t is the power loss, SN is the rated capacity of the transformer, S2 is the secondary side load power of the transformer, ΔP0, ΔP k are the no-load active power loss and rated load active power loss of the transformer, β is the load factor, δ is the unit loss cost of the transformer, and ΔT is the operating time of the transformer;

[0036]

[0037] Where: P q is the load value of the qth substation, R g is the resistance per unit length of the high-voltage distribution line, U N is the rated voltage of the distribution line, γ1 is the tortuosity coefficient of the high-voltage distribution line, l(q) is the distance to the qth substation, δ is the network loss cost, and ΔT is the transformer operation time;

[0038]

[0039] Where: P j is the load value of the jth load point, R z is the resistance per unit length of the medium-voltage distribution line, γ2 is the tortuosity coefficient of the medium-voltage distribution line, and l(p,q) is the straight-line distance between the p-th substation and the q-th load point.

[0040] The step S2 specifically includes:

[0041] S21. Using the preset cost index as the evaluation index, construct an evaluation system, and use different evaluation indexes in the evaluation system as different types of nodes to construct a node set V, where V = {v1, v2, ..., v i ,...,v n}, |V| = n;

[0042] S22. Determine the size of the edge set E based on the connection relationship and connection strength between the nodes:

[0043] If the connection relationship between the nodes is a complete graph, the size of the edge set E is as follows:

[0044]

[0045] Where: n is the number of nodes;

[0046] If the connection relationship between nodes is a sparse graph, the mutual information between nodes is calculated, and its expression is as follows:

[0047] I(i,j)=lg(P(i,j) / (P(i)P(j)));

[0048] Where: I(i, j) is the mutual information between nodes i and j, P(i, j) is the joint probability distribution of nodes i and j appearing at the same time, P(i) and P(j) are the marginal probability distributions of nodes i and j appearing respectively;

[0049] Based on the mean mutual information between nodes, a dynamic adaptive threshold is set, which is expressed as follows:

[0050] θ=μ+ασ;

[0051] Where: θ is the adaptive threshold, μ is the mean of mutual information, α is the learnable parameter, and σ is the standard deviation;

[0052] Determine whether the mutual information between node i and node j is greater than the adaptive threshold θ; if so, establish an edge between node i and node j;

[0053] Assume that the feature vectors of node i and node j are as follows:

[0054] X i =[x1, x2, ..., x m ], X j =[y1, y2, ..., y m ];

[0055] Where: X i is the feature vector of node i, X j is the feature vector of node j, and m is the number of samples;

[0056] Based on the feature vectors of node i and node j, determine the connection strength S between node i and node j ij , which is expressed as follows:

[0057]

[0058] A global or local threshold θ is set based on the connection strength, and the top 20% of the threshold θ strength distribution is used as the screening condition to filter and retain the edges that meet the conditions. The expression is as follows:

[0059] E={(i, j)|S ij ≥θ};

[0060] Then the size of the edge set E is: Where: d i is the number of edges connected to node i;

[0061] S23. Based on the node set V and the edge set E, construct a multimodal graph structure network G = (V, E).

[0062] The step S3 specifically includes:

[0063] S31. The initial feature matrix of each node in the multimodal graph structure network contains text data, image data, and time series data. The multimodal feature vector is extracted from the initial feature matrix, and its expression is as follows:

[0064]

[0065] in: The feature vector generated by the BioBERT model after encoding the text data, It is the feature vector generated by the ResNet model after encoding the image data. X is the feature vector generated by the LSTM model after encoding the time series data. i is the initial feature matrix;

[0066] S32. Normalize the multimodal feature vectors and assign dynamic weights to the established edges based on a multi-layer perceptron. Capture the importance between nodes, obtain attention weights, and perform layer-by-layer propagation and fusion to obtain new node representations, which are expressed as follows:

[0067]

[0068] in: is the graph structure network where node i is at layer 0, sig is the sigmoid function, H i is the multimodal feature vector;

[0069]

[0070] in: is a graph structure network with node i at layer l+1, is the graph structure network of node j in layer l, σ is the activation function, c ij is the normalization coefficient, a ij is the attention weight of node j to node i, is the attention vector, w ij is the weight matrix, N(i) is the neighborhood set of node i, || is the feature concatenation operation, Leaky ReLU is the nonlinear transformation function, MLP is the multi-layer perceptron, H k is the feature representation of node k in the neighborhood of node i, b l is the bias term;

[0071] S33. Weight the importance of the node by the attention weight to obtain the dynamic weight vector in the current situation, which is expressed as follows:

[0072]

[0073] Where: W tis the dynamic weight vector at time t, is the output feature of node i in the Lth layer of the multi-layer graph structure network.

[0074] The step S4 specifically includes:

[0075] S41. Embed the fuzzy reasoning matrix in the multimodal graph structure network and perform multi-layer fuzzy reasoning convolution calculation. Its expression is as follows:

[0076]

[0077] in: is the l+1 layer output of the fuzzy inference matrix, A t is the adaptive adjacency matrix, is the l-layer output of the fuzzy inference matrix, r k is the vector representation of the k-th rule;

[0078]

[0079] Where: Z t is the embedding representation learned from the fuzzy inference matrix, d i d j are the degrees of node i and node j respectively;

[0080] S42. Mapping fuzzy inference rules R′ into node features of a multimodal graph structure network Perform adaptive multi-layer graph structure network calculation to obtain the fuzzy weight of fuzzy reasoning, which is expressed as follows:

[0081] R′=AdaptiveGNN(R,W t , A); R={r1, r2,..., r k ,...,r m};

[0082] Among them: AdaptiveGNN is an adaptive multi-layer graph structure network, R is the original fuzzy matrix, A is the static adjacency matrix, r k is the vector representation of the kth fuzzy inference rule;

[0083] S43. Defuzzify the fuzzy weight of fuzzy reasoning by Bayesian weighted method to obtain the comprehensive defuzzification index weight, which is expressed as follows:

[0084]

[0085] in: is the comprehensive defuzzification index weight, x i is the fuzzy weight of fuzzy reasoning, μ(x i) is the fuzzy membership function, P(μ(x i ) is the fuzzy membership function μ(x i ), b l is the bias term.

[0086] The step S5 specifically includes:

[0087] S51. Based on the weights of the comprehensive defuzzification indicators, the investment and construction costs, operating costs, maintenance costs, reliability failure costs, decommissioning costs, and electricity costs in the full life cycle model of the power grid are weighted to obtain the comprehensive cost of the full life cycle of the power grid, which is expressed as follows:

[0088] LCC t =CI t +CO t +CM t +CF t +CD t +CP t ;

[0089] Of which: LCC t is the weighted comprehensive cost of the entire life of the power grid, CI t 、CO t CM t CF t 、CD t 、CP t They are the weighted investment and construction cost, operation cost, maintenance cost, reliability failure cost, decommissioning cost, and electricity cost of the power grid;

[0090] The weighted investment and construction costs, operating costs, maintenance costs, reliability failure costs, decommissioning costs, and electricity costs of the power grid are as follows:

[0091]

[0092] in: is the dynamic weight of the investment and construction cost of the power grid after weighting based on the comprehensive defuzzification index weights, CI is the investment and construction cost of the power grid; is the dynamic weight of the grid operation cost after weighting based on the comprehensive defuzzification index weights, CO is the grid operation cost; is the dynamic weight of the maintenance cost of the power grid after weighting based on the comprehensive defuzzification index weights, CM is the maintenance cost of the power grid; is the dynamic weight of the reliability failure cost of the power grid after weighting based on the comprehensive defuzzification index weights, CF is the reliability failure cost of the power grid; is the dynamic weight of the power grid decommissioning cost after weighting based on the comprehensive defuzzification index weights, CD is the power grid decommissioning cost; is the dynamic weight of the power cost of the power grid after weighting based on the comprehensive defuzzification index weights, and CP is the power cost of the power grid.

[0093] A power grid cost configuration system based on a multimodal graph structure and fuzzy evaluation is applied to the above method, and the system includes:

[0094] The full life cycle model construction module is used to obtain the preset cost indicators of the power grid throughout its life cycle and build a full life cycle model of the power grid;

[0095] The multimodal graph structure network construction module is used to build an evaluation system based on preset cost indicators, using the evaluation indicators in the evaluation system as nodes to construct a node set; calculate the mutual information based on the connection relationship between nodes, establish edges between nodes, and construct an edge set; and construct a multimodal graph structure network based on the node set and edge set;

[0096] The dynamic weight vector acquisition module is used to extract multimodal feature vectors based on the multimodal graph structure network and assign dynamic weights to the established edges; it obtains attention weights to weight the importance of nodes and obtains the dynamic weight vector in the current context;

[0097] The fuzzy reasoning optimization and defuzzification module is used to embed the fuzzy reasoning matrix in the multimodal graph structure network based on the dynamic weight vector, optimize the fuzzy reasoning rule process, and defuzzify the fuzzy weight of the fuzzy reasoning through the Bayesian weighted mind method to obtain the comprehensive defuzzification index weight;

[0098] The power grid full life comprehensive cost calculation module is used to perform weighted calculation on the full life cycle model based on the comprehensive defuzzification index weights to obtain the full life comprehensive cost of the power grid.

[0099] A power grid cost configuration device based on a multimodal graph structure and fuzzy evaluation, the device comprising a processor and a memory; the memory is used to store computer program code and transmit the computer program code to the processor;

[0100] The processor is configured to execute the above-mentioned power grid cost configuration method based on multimodal graph structure and fuzzy evaluation according to the instructions in the computer program code.

[0101] A computer-readable storage medium stores computer-executable instructions. When the computer-executable instructions are executed on a computer, the above-mentioned power grid cost configuration method based on multimodal graph structure and fuzzy evaluation is implemented.

[0102] Compared with the prior art, the present invention has the following beneficial effects:

[0103] The present invention provides a power grid cost configuration method and system based on multimodal graph structure and fuzzy evaluation. The method first obtains preset cost indicators and constructs a full life cycle model; then constructs an evaluation system based on the cost indicators to form a multimodal graph structure network; then extracts multimodal feature vectors, assigns dynamic weights to edges, and calculates dynamic weight vectors through attention weights; then embeds a fuzzy reasoning matrix, uses the Bayesian weighted center method to defuzzify, and obtains a comprehensive weight; finally, based on the comprehensive weight, the full life cycle model is weighted and calculated to obtain the full life comprehensive cost of the power grid; in application, this design predicts and configures various costs of the power grid through the full life cycle model and a dynamic fuzzy evaluation method based on a multimodal graph neural network, and dynamically assigns the weight of each indicator to the evaluation object, avoiding subjective bias, and taking into account the correlation between various indicators to reduce errors. At the same time, the full life cycle model covers the costs of each stage of the power grid from procurement to scrapping, realizing a comprehensive evaluation of the power grid life. BRIEF DESCRIPTION OF THE DRAWINGS

[0104] Figure 1 It is a flow chart of the method of the present invention.

[0105] Figure 2 It is a system structure diagram of the present invention.

[0106] Figure 3 It is a diagram of the equipment structure of the present invention.

[0107] In the figure: full life cycle model construction module 1, multimodal graph structure network construction module 2, dynamic weight vector acquisition module 3, fuzzy reasoning optimization and defuzzification module 4, power grid full life comprehensive cost calculation module 5, processor 6, memory 7, computer program code 71. DETAILED DESCRIPTION

[0108] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0109] Example 1:

[0110] See also Figure 1 , a power grid cost configuration method based on multimodal graph structure and fuzzy evaluation, including:

[0111] S1. Obtain preset cost indicators of the power grid over its entire life cycle and construct a full life cycle model of the power grid. In step S1, the preset cost indicators include investment and construction costs, operating costs, maintenance costs, reliability failure costs, decommissioning costs, and electricity costs. The expression of the full life cycle model of the power grid is as follows:

[0112] LCC=CI+CO+CM+CF+CD+CP;

[0113] Where: CI is the investment and construction cost of the grid, CO is the operation cost of the grid, CM is the maintenance cost of the grid, CF is the reliability failure cost of the grid, CD is the decommissioning cost of the grid, and CP is the electricity cost of the grid;

[0114] CI=C s +C line +C sj ;

[0115] Where: C s is the investment cost of transmission tower construction, C line is the investment cost of distribution line construction, C sj Design costs for construction;

[0116] Furthermore, the construction investment costs of transmission towers and distribution lines are as follows:

[0117]

[0118] Where: C1 is the total investment cost of transmission tower, t ope is the equipment operating time, r0 is the discount rate;

[0119]

[0120] Among them: z is the construction investment cost per unit length of the distribution line, γ2 is the tortuosity coefficient of the distribution line, J q is the set of loads supplied by the qth substation, J is the set of all load points, and l(q, j) is the straight-line distance between the qth substation and the jth load point;

[0121] CO=C sope +C gope +C zope ;

[0122] Where: C sope is the substation operating cost, C gope is the operating cost of high-voltage distribution lines, C zope The operating cost of medium voltage distribution lines;

[0123] Furthermore, the operating costs of substations, high-voltage distribution lines, and medium-voltage distribution lines include:

[0124]

[0125] Where: ΔP M，t is the power loss, S N is the rated capacity of the transformer, S2 is the secondary side load power of the transformer, ΔP0, ΔP kare the no-load active power loss and rated load active power loss of the transformer, β is the load factor, δ is the unit loss cost of the transformer, and ΔT is the operating time of the transformer;

[0126]

[0127] Where: P q is the load value of the qth substation, R g is the resistance per unit length of the high-voltage distribution line, U N is the rated voltage of the distribution line, γ1 is the tortuosity coefficient of the high-voltage distribution line, l(q) is the distance to the qth substation, δ is the network loss cost, and ΔT is the transformer operation time;

[0128]

[0129] Where: P j is the load value of the jth load point, R z is the resistance per unit length of the medium-voltage distribution line, γ2 is the tortuosity coefficient of the medium-voltage distribution line, and l(p,q) is the straight-line distance between the p-th substation and the q-th load point.

[0130]

[0131] Among them: PM i is the unit price of preventive maintenance for equipment i, Q i is the number of installations of device i, λ PM,i is the average annual preventive maintenance frequency of equipment i, CM j is the repair cost of faulty component j, f j is the failure probability of faulty component j, τ j is the mean repair time, r is the discount rate, and t is the t-th year in the whole life cycle;

[0132]

[0133] Of which: EENS k is the expected value of power shortage under fault scenario k, VOLL is the unit power outage loss value, RC k is the emergency repair cost of fault scenario k;

[0134] CD=C d wC r β;

[0135] Where: C d The cost of decommissioning, C r is the original value of the equipment, and β is the residual value rate;

[0136]

[0137] Where: P grid (t) is the purchased electricity in period t, ρ(t) is the time-of-use electricity price curve, C gen is the cost of self-provided power generation, P local is the self-generated power.

[0138] S2. Build an evaluation system based on preset cost indicators, use the evaluation indicators in the evaluation system as nodes, and build a node set; calculate the mutual information based on the connection relationship between the nodes, establish edges between the nodes, and build an edge set; build a multimodal graph structure network based on the node set and edge set;

[0139] In the evaluation system, different evaluation indicators, such as infrastructure costs, land and rights costs, and construction costs in power grid construction, are regarded as different types of nodes, thus forming a multimodal graph structure; each node can contain different types of information, such as numerical data, text descriptions, image features, etc.

[0140] Furthermore, the step S2 specifically includes:

[0141] S21. Using the preset cost index as the evaluation index, construct an evaluation system, and use different evaluation indexes in the evaluation system as different types of nodes to construct a node set V, where V = {v1, v2, ..., v i ,...,v n}, |V| = n; each node represents an evaluation indicator (such as "infrastructure cost", "infrastructure cost"), where n is the total number of nodes.

[0142] S22. Determine the size of the edge set E based on the connection relationship and connection strength between the nodes:

[0143] If the connection relationship between the nodes is a complete graph, which means that each indicator in the evaluation system affects each other, then the size of the edge set E is: Where: n is the number of nodes.

[0144] If the connection relationship between nodes is a sparse graph, it means that not all indicators in the evaluation system have a direct impact on each other. In this solution, an adaptive threshold method based on mutual information is used to determine the node connection relationship, specifically including:

[0145] First, calculate the mutual information between nodes, which is expressed as follows:

[0146] I(i,j)=lg(P(i,j) / (P(i)P(j)));

[0147] Where: I(i, j) is the mutual information between node i and node j, P(i, j) is the joint probability distribution of node i and node j appearing at the same time, P(i) and P(j) are the marginal probability distributions of node i and node j appearing respectively;

[0148] In order to determine whether an edge needs to be established between nodes i and j based on the mutual information, a dynamic threshold θ can be used. The threshold setting is adaptive, that is, it is adjusted according to the distribution of the data, as follows:

[0149] Based on the mean mutual information between nodes, a dynamic adaptive threshold is set, which is expressed as follows:

[0150] θ=μ+ασ;

[0151] Where: θ is the adaptive threshold, μ is the mean mutual information, α is a learnable parameter (the preferred initial value is 1.5), and σ is the standard deviation;

[0152] Determine whether the mutual information between node i and node j is greater than the adaptive threshold θ; if it is, establish an edge between node i and node j; that is, when I(i,j)≥θ, establish edge e ij ∈E; where e ij represents the edge between node i and node j;

[0153] Then assume that the feature vectors of node i and node j are as follows:

[0154] X i =[x1, x2, ..., x m ], X j =[y1, y2, ..., y m ];

[0155] Where: X i is the feature vector of node i, X j is the feature vector of node j, and m is the number of samples;

[0156] Based on the feature vectors of node i and node j, determine the connection strength S between node i and node j ij , which is expressed as follows:

[0157]

[0158] A global or local threshold θ is set based on the connection strength, and the top 20% of the threshold θ strength distribution is used as the screening condition to filter and retain the edges that meet the conditions. The expression is as follows:

[0159] E={(i, j)|S ij ≥θ};

[0160] Then the size of the edge set E is: Where: d i is the number of edges connected to node i;

[0161] S23. Based on the node set V and the edge set E, construct a multimodal graph structure network G = (V, E).

[0162] S3. Based on the multimodal graph structure network, extract the multimodal feature vector and assign dynamic weights to the established edges; obtain the attention weight to weight the importance of the node and obtain the dynamic weight vector in the current context;

[0163] Furthermore, the step S3 specifically includes:

[0164] S31, the initial feature matrix X of each node i in the multimodal graph structure network i They all contain text data, image data, and time series data. The multimodal feature vector is extracted from the initial feature matrix, and its expression is as follows:

[0165]

[0166] in: The feature vectors generated by encoding text data for the BioBERT model can be used to fine-tune the BioBERT model for specific tasks, such as power grid construction cost assessment, making it more suitable for the text content in that specific field; It is the feature vector generated by the ResNet model after encoding the image data. X is the feature vector generated by the LSTM model after encoding the time series data. i is the initial feature matrix;

[0167] S32. Normalize the multimodal feature vectors and assign dynamic weights to the established edges based on a multi-layer perceptron. Capture the importance between nodes, obtain attention weights, and perform layer-by-layer propagation and fusion to obtain new node representations, which are expressed as follows:

[0168]

[0169] in: is the graph structure network where node i is at layer 0, sig is the sigmoid function, H i is the multimodal feature vector;

[0170]

[0171] in: is a graph structure network with node i at layer l+1, is the graph structure network of node j in layer l, σ is the activation function, cij is the normalization coefficient, a ij is the attention weight of node j to node i, is the attention vector, w ij is the weight matrix, N(i) is the neighborhood set of node i, || is the feature concatenation operation, Leaky ReLU is the nonlinear transformation function, MLP is the multi-layer perceptron, H k is the feature representation of node k in the neighborhood of node i, b l is the bias term.

[0172] In this plan is through The corresponding cumulative operation is obtained; in the construction of power grid, Contains the fusion information of "transmission tower material cost" (value), "geographic environment satellite map" (image), and "historical failure rate" (time series); each layer Used to capture topological relationships of different granularities: for example, when the shallow layer l = 1, local direct associations are learned (such as the linear relationship between "land cost" and "construction cost"); when the deep layer l = 3, global complex interactions are modeled (such as the nonlinear impact of "decommissioning cost" on multi-stage operation and maintenance strategies). Through layer-by-layer cumulative propagation and fusion, the complex correlation relationships between various influencing factors can be effectively captured.

[0173] S33. Weight the importance of the node by the attention weight to obtain the dynamic weight vector in the current situation, which is expressed as follows:

[0174]

[0175] Where: W t is the dynamic weight vector at time t, W is the output feature of node i in the Lth layer, i.e. the last layer, of the multi-layer graph network. t Adaptively adjust at each evaluation to reflect different evaluation scenarios and changes in the relationships between nodes.

[0176] S4. Based on the dynamic weight vector, the fuzzy reasoning matrix is ​​embedded in the multimodal graph structure network to optimize the fuzzy reasoning rule process. The fuzzy weight of the fuzzy reasoning is defuzzified by the Bayesian weighted mind method to obtain the comprehensive defuzzification index weight;

[0177] Furthermore, the step S4 specifically includes:

[0178] S41. Embed the fuzzy reasoning matrix in the multimodal graph structure network and perform multi-layer fuzzy reasoning convolution calculation. Its expression is as follows:

[0179]

[0180] in: is the l+1 layer output of the fuzzy inference matrix, A t is the adaptive adjacency matrix, is the l-layer output of the fuzzy inference matrix, r k is the vector representation of the k-th rule;

[0181] In this plan, is the updated feature of node i at layer l+1, which includes the dual optimization results of fuzzy reasoning and graph convolution. For example, in the power grid cost model, the nodes are defined as: v1: transmission tower construction cost, v2: land acquisition cost, v3: equipment maintenance cost; then the fuzzy reasoning feature is: is the fuzzy membership of the transmission tower construction cost, for example, the membership of “high cost” is 0.8; is the fuzzy membership of the land acquisition cost, for example, the membership of “medium cost” is 0.5; is the fuzzy membership of equipment maintenance cost. For example, if the membership of “low cost” is 0.3, then the output feature It is an updated feature after comprehensively considering the neighborhood node relationship and fuzzy rules. For example: The updated degree of subordination of the transmission tower construction cost is adjusted from 0.8 to 0.75 to reflect the impact of land cost.

[0182]

[0183] Where: Z t is the embedding representation learned from the fuzzy inference matrix, d i d j are the degrees of node i and node j respectively; the degrees represent the number of neighbor nodes of node i, that is, the number of nodes directly connected to node i.

[0184] S42. Map the fuzzy inference rule R′ to the node feature F of the multimodal graph structure network i l+1 , perform adaptive multi-layer graph structure network calculation to obtain the fuzzy weight of fuzzy reasoning, which is expressed as follows:

[0185] R′=AdaptiveGNN(R,W t ,A); R=(r1,r2,...,r k ,...,r m};

[0186] Among them: AdaptiveGNN is an adaptive multi-layer graph structure network, R is the original fuzzy matrix, A is the static adjacency matrix, r k is the vector representation of the kth fuzzy inference rule;

[0187] In this scheme, fuzzy inference rules are logical rules based on expert knowledge, which are used to describe the fuzzy relationship between input variables and output variables, and are usually expressed as "IF-THEN" rules.

[0188] The static adjacency matrix A refers to: nodes (such as substations, transmission lines, load points); edges: physical connection relationships (such as substations-transmission lines). Adaptive adjacency matrix A t This refers to dynamically adjusting the relationships between nodes; for example, load fluctuations leading to changes in the intensity of interactions between substations and dynamic adjustments to fault propagation paths.

[0189] If the original fuzzy matrix R's rule description is: IF "load" is "high" THEN "maintenance cost" is "high"; IF "failure frequency" is "high" THEN "retirement cost" is "high". After dynamically adjusting the rule weights through fuzzy inference rule optimization R', for example, the weight of the "load-maintenance cost" rule is increased from 0.7 to 0.8, then based on R' and A t Update node feature F i l+1 For example, the membership of the maintenance cost of the substation node is adjusted from 0.6 to 0.7; the membership of the failure risk of the load point node is adjusted from 0.4 to 0.5.

[0190] S43. Defuzzify the fuzzy weight of fuzzy reasoning by Bayesian weighted method to obtain the comprehensive defuzzification index weight, which is expressed as follows:

[0191]

[0192] in: is the comprehensive defuzzification index weight, x i is the fuzzy weight of fuzzy reasoning, μ(x i ) is the fuzzy membership function, P(μ(x i ) is the fuzzy membership function μ(x i ), b l is the bias term.

[0193] In this solution, after obtaining the weights, visualization tools such as multimodal heat maps, radar charts, and graph networks can be used to display the relationships and interactions between different dimensions and nodes, making the evaluation results more intuitive and easy to understand.

[0194] In this solution, it is preferred to dynamically update and optimize the weights through the cross entropy loss or MSE loss of the training model, so that the model can continuously improve the weight distribution and inference rules in multiple iterations. The specific loss formula is as follows:

[0195]

[0196] Where: y i is the actual evaluation index weight, The weight of the evaluation index predicted by the model.

[0197] S5. Based on the weights of the comprehensive defuzzification indicators, a weighted calculation is performed on the full life cycle model to obtain the comprehensive cost of the full life cycle of the power grid.

[0198] Furthermore, the step S5 specifically includes:

[0199] S51. Based on the weights of the comprehensive defuzzification indicators, the investment and construction costs, operating costs, maintenance costs, reliability failure costs, decommissioning costs, and electricity costs in the full life cycle model of the power grid are weighted to obtain the comprehensive cost of the full life cycle of the power grid, which is expressed as follows:

[0200] LCC t =CI t +CO t +CM t +CF t +CD t +CP t ;

[0201] Of which: LCC t is the weighted comprehensive cost of the entire life of the power grid, CI t 、CO t CM t CF t 、CD t 、CP t They are the weighted investment and construction cost, operation cost, maintenance cost, reliability failure cost, decommissioning cost, and electricity cost of the power grid;

[0202] The weighted investment and construction costs, operating costs, maintenance costs, reliability failure costs, decommissioning costs, and electricity costs of the power grid are as follows:

[0203]

[0204] in: is the dynamic weight of the investment and construction cost of the power grid after weighting based on the comprehensive defuzzification index weights, CI is the investment and construction cost of the power grid; is the dynamic weight of the grid operation cost after weighting based on the comprehensive defuzzification index weights, CO is the grid operation cost; is the dynamic weight of the maintenance cost of the power grid after weighting based on the comprehensive defuzzification index weights, CM is the maintenance cost of the power grid; is the dynamic weight of the reliability failure cost of the power grid after weighting based on the comprehensive defuzzification index weights, CF is the reliability failure cost of the power grid; is the dynamic weight of the power grid decommissioning cost after weighting based on the comprehensive defuzzification index weights, CD is the power grid decommissioning cost; is the dynamic weight of the power cost of the power grid after weighting based on the comprehensive defuzzification index weights, and CP is the power cost of the power grid.

[0205] Example 2:

[0206] See also Figure 2 A power grid cost configuration system based on a multimodal graph structure and fuzzy evaluation is provided, the system being applied to the method described in Example 1, the system comprising:

[0207] The full life cycle model construction module 1 is used to obtain the preset cost indicators of the power grid throughout its life cycle and construct the full life cycle model of the power grid;

[0208] Furthermore, the full life cycle model of the power grid constructed by the full life cycle model construction module 1 is as follows:

[0209] The preset cost indicators include investment and construction costs, operating costs, maintenance costs, reliability failure costs, decommissioning costs, and electricity costs;

[0210] The expression of the full life cycle model of the power grid is as follows:

[0211] LCC=CI+CO+CM+CF+CD+CP;

[0212] Where: CI is the investment and construction cost of the grid, CO is the operation cost of the grid, CM is the maintenance cost of the grid, CF is the reliability failure cost of the grid, CD is the decommissioning cost of the grid, and CP is the electricity cost of the grid;

[0213] CI=C s +C line +C si ;

[0214] Where: C s is the investment cost of transmission tower construction, C line is the investment cost of distribution line construction, C sj Design costs for construction;

[0215] The construction investment costs of transmission towers and distribution lines include:

[0216]

[0217] Where: C1 is the total investment cost of transmission tower, t ope is the equipment operating time, r0 is the discount rate;

[0218]

[0219] Among them: z is the construction investment cost per unit length of the distribution line, γ2 is the tortuosity coefficient of the distribution line, J q is the set of loads supplied by the qth substation, J is the set of all load points, and l(q,j) is the straight-line distance between the qth substation and the jth load point;

[0220] CO=C sope +C gope +C zope ;

[0221] Where: C sope is the substation operating cost, C gope is the operating cost of high-voltage distribution lines, C zope The operating cost of medium voltage distribution lines;

[0222] Substation operating costs, high-voltage distribution line operating costs, and medium-voltage distribution line operating costs, specifically including:

[0223]

[0224] Where: ΔP M，t is the power loss, S N is the rated capacity of the transformer, S2 is the secondary side load power of the transformer, ΔP0, ΔP k are the no-load active power loss and rated load active power loss of the transformer, β is the load factor, δ is the unit loss cost of the transformer, and ΔT is the operating time of the transformer;

[0225]

[0226] Where: P q is the load value of the qth substation, R g is the resistance per unit length of the high-voltage distribution line, U N is the rated voltage of the distribution line, γ1 is the tortuosity coefficient of the high-voltage distribution line, l(q) is the distance to the qth substation, δ is the network loss cost, and ΔT is the transformer operation time;

[0227]

[0228] Where: P j is the load value of the jth load point, R z is the resistance per unit length of the medium-voltage distribution line, γ2 is the tortuosity coefficient of the medium-voltage distribution line, and l(p,q) is the straight-line distance between the p-th substation and the q-th load point.

[0229]

[0230] Among them: PM iis the unit price of preventive maintenance for equipment i, Q i is the number of installations of device i, λ PMi is the average annual preventive maintenance frequency of equipment i, CM j is the repair cost of faulty component j, f j is the failure probability of faulty component j, τ j is the mean repair time, r is the discount rate, and t is the t-th year in the whole life cycle;

[0231]

[0232] Of which: EENS k is the expected value of power shortage under fault scenario k, VOLL is the unit power outage loss value, RC k is the emergency repair cost of fault scenario k;

[0233] CD=C d wC r β;

[0234] Where: C d The cost of decommissioning, C r is the original value of the equipment, and β is the residual value rate;

[0235]

[0236] Where: P grid (t) is the purchased electricity in period t, ρ(t) is the time-of-use electricity price curve, C gen is the cost of self-provided power generation, P local is the self-generated power.

[0237] Multimodal graph structure network construction module 2 is used to build an evaluation system based on preset cost indicators, use the evaluation indicators in the evaluation system as nodes, and build a node set; calculate the mutual information based on the connection relationship between the nodes, establish edges between the nodes, and build an edge set; and build a multimodal graph structure network based on the node set and edge set;

[0238] Furthermore, the multimodal graph structure network construction module 2 constructs a multimodal graph structure network according to the following steps:

[0239] S21. Using the preset cost index as the evaluation index, construct an evaluation system, and use different evaluation indexes in the evaluation system as different types of nodes to construct a node set V, where V = {v1, v2, ..., v i ,...,v n}, |V| = n;

[0240] S22. Determine the size of the edge set E based on the connection relationship and connection strength between the nodes:

[0241] If the connection relationship between the nodes is a complete graph, the size of the edge set E is as follows:

[0242]

[0243] Where: n is the number of nodes;

[0244] If the connection relationship between nodes is a sparse graph, the mutual information between nodes is calculated, and its expression is as follows:

[0245] I(i,j)=lg(P(i,j) / (P(i)P(j)));

[0246] Where: I(i, j) is the mutual information between nodes i and j, P(i, j) is the joint probability distribution of nodes i and j appearing at the same time, P(i) and P(j) are the marginal probability distributions of nodes i and j appearing respectively;

[0247] Based on the mean mutual information between nodes, a dynamic adaptive threshold is set, which is expressed as follows:

[0248] θ=μ+ασ;

[0249] Where: θ is the adaptive threshold, μ is the mean of mutual information, α is the learnable parameter, and σ is the standard deviation;

[0250] Determine whether the mutual information between node i and node j is greater than the adaptive threshold θ; if so, establish an edge between node i and node j;

[0251] Assume that the feature vectors of node i and node j are as follows:

[0252] X i =[x1, x2, ..., x m ], X j =[y1, y2, ..., y m ];

[0253] Where: X i is the feature vector of node i, X j is the feature vector of node j, and m is the number of samples;

[0254] Based on the feature vectors of node i and node j, determine the connection strength S between node i and node j ij , which is expressed as follows:

[0255]

[0256] A global or local threshold θ is set based on the connection strength, and the top 20% of the threshold θ strength distribution is used as the screening condition to filter and retain the edges that meet the conditions. The expression is as follows:

[0257] E={(i, j)|S ij ≥θ};

[0258] Then the size of the edge set E is as follows:

[0259]

[0260] Where: d i is the number of edges connected to node i;

[0261] S23. Based on the node set V and the edge set E, construct a multimodal graph structure network G = (V, E).

[0262] Dynamic weight vector acquisition module 3 is used to extract multimodal feature vectors based on the multimodal graph structure network and assign dynamic weights to the established edges; obtain attention weights to weight the importance of nodes and obtain the dynamic weight vector in the current context;

[0263] Furthermore, the dynamic weight vector acquisition module 3 acquires the dynamic weight vector according to the following steps:

[0264] S31. The initial feature matrix of each node in the multimodal graph structure network contains text data, image data, and time series data. The multimodal feature vector is extracted from the initial feature matrix, and its expression is as follows:

[0265]

[0266] in: The feature vector generated by the BioBERT model after encoding the text data, It is the feature vector generated by the ResNet model after encoding the image data. X is the feature vector generated by the LSTM model after encoding the time series data. i is the initial feature matrix;

[0267] S32. Normalize the multimodal feature vectors and assign dynamic weights to the established edges based on a multi-layer perceptron. Capture the importance between nodes, obtain attention weights, and perform layer-by-layer propagation and fusion to obtain new node representations, which are expressed as follows:

[0268]

[0269] in: is the graph structure network where node i is at layer 0, sig is the sigmoid function, H i is the multimodal feature vector;

[0270]

[0271] in: is a graph structure network with node i at layer l+1, is the graph structure network of node j in layer l, σ is the activation function, c ij is the normalization coefficient, a ij is the attention weight of node j to node i, is the attention vector, w ij is the weight matrix, N(i) is the neighborhood set of node i, || is the feature concatenation operation, Leaky ReLU is the nonlinear transformation function, MLP is the multi-layer perceptron, H k is the feature representation of node k in the neighborhood of node i, b l is the bias term;

[0272] S33. Weight the importance of the node by the attention weight to obtain the dynamic weight vector in the current situation, which is expressed as follows:

[0273]

[0274] Where: W t is the dynamic weight vector at time t, is the output feature of node i in the Lth layer of the multi-layer graph structure network.

[0275] The fuzzy reasoning optimization and defuzzification module 4 is used to embed the fuzzy reasoning matrix in the multimodal graph structure network based on the dynamic weight vector, optimize the fuzzy reasoning rule process, and defuzzify the fuzzy weight of the fuzzy reasoning through the Bayesian weighted mind method to obtain the comprehensive defuzzification index weight;

[0276] Furthermore, the fuzzy reasoning optimization and defuzzification module 4 obtains the comprehensive defuzzification index weight according to the following steps:

[0277] S41. Embed the fuzzy reasoning matrix in the multimodal graph structure network and perform multi-layer fuzzy reasoning convolution calculation. Its expression is as follows:

[0278]

[0279] in: is the l+1 layer output of the fuzzy inference matrix, A t is the adaptive adjacency matrix, is the l-layer output of the fuzzy inference matrix, r k is the vector representation of the k-th rule;

[0280]

[0281] Where: Z tis the embedding representation learned from the fuzzy inference matrix, d i d j are the degrees of node i and node j respectively;

[0282] S42. Mapping fuzzy inference rules R′ into node features of a multimodal graph structure network Perform adaptive multi-layer graph structure network calculation to obtain the fuzzy weight of fuzzy reasoning, which is expressed as follows:

[0283] R′=AdaptiveGNN(R,W t , A); R={r1, r2,..., r k ,...,r m};

[0284] Among them: AdaptiveGNN is an adaptive multi-layer graph structure network, R is the original fuzzy matrix, A is the static adjacency matrix, r k is the vector representation of the kth fuzzy inference rule;

[0285] S43. Defuzzify the fuzzy weight of fuzzy reasoning by Bayesian weighted method to obtain the comprehensive defuzzification index weight, which is expressed as follows:

[0286]

[0287] in: is the comprehensive defuzzification index weight, x i is the fuzzy weight of fuzzy reasoning, μ(x i ) is the fuzzy membership function, P(μ(x i ) is the fuzzy membership function μ(x i ), b l is the bias term.

[0288] The power grid life cycle comprehensive cost calculation module 5 is used to perform weighted calculation on the full life cycle model based on the comprehensive defuzzification index weights to obtain the power grid life cycle comprehensive cost.

[0289] Furthermore, the power grid lifecycle comprehensive cost calculation module 5 calculates the power grid lifecycle comprehensive cost according to the following steps:

[0290] S51. Based on the weights of the comprehensive defuzzification indicators, the investment and construction costs, operating costs, maintenance costs, reliability failure costs, decommissioning costs, and electricity costs in the full life cycle model of the power grid are weighted to obtain the comprehensive cost of the full life cycle of the power grid, which is expressed as follows:

[0291] LCC t =CI t +CO t +CMt +CF t +CD t +CP t ;

[0292] Of which: LCC t is the weighted comprehensive cost of the entire life of the power grid, CI t 、CO t CM t CF t 、CD t 、CP t They are the weighted investment and construction cost, operation cost, maintenance cost, reliability failure cost, decommissioning cost, and electricity cost of the power grid;

[0293] The weighted investment and construction costs, operating costs, maintenance costs, reliability failure costs, decommissioning costs, and electricity costs of the power grid are as follows:

[0294]

[0295] in: is the dynamic weight of the investment and construction cost of the power grid after weighting based on the comprehensive defuzzification index weights, CI is the investment and construction cost of the power grid; is the dynamic weight of the grid operation cost after weighting based on the comprehensive defuzzification index weights, CO is the grid operation cost; is the dynamic weight of the maintenance cost of the power grid after weighting based on the comprehensive defuzzification index weights, CM is the maintenance cost of the power grid; is the dynamic weight of the reliability failure cost of the power grid after weighting based on the comprehensive defuzzification index weights, CF is the reliability failure cost of the power grid; is the dynamic weight of the power grid decommissioning cost after weighting based on the comprehensive defuzzification index weights, CD is the power grid decommissioning cost; is the dynamic weight of the power cost of the power grid after weighting based on the comprehensive defuzzification index weights, and CP is the power cost of the power grid.

[0296] Example 3:

[0297] See also Figure 3 , a power grid cost configuration device based on a multimodal graph structure and fuzzy evaluation, the device comprising a processor 6 and a memory 7; the memory 7 is used to store a computer program code 71 and transmit the computer program code 71 to the processor 6;

[0298] The processor 6 is configured to execute the power grid cost configuration method based on multimodal graph structure and fuzzy evaluation described in Example 1 according to the instructions in the computer program code 71 .

[0299] This embodiment also includes a computer-readable storage medium, which stores computer-executable instructions. When the computer-executable instructions are executed on a computer, the power grid cost configuration method based on multimodal graph structure and fuzzy evaluation described in Example 1 is implemented.

[0300] Generally speaking, computer instructions for implementing the method of the present invention may be carried by any combination of one or more computer-readable storage media. Non-transitory computer-readable storage media may include any computer-readable media except for signals that are temporarily propagating.

[0301] Computer-readable storage media can be, for example, but not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or components, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: an electrical connection having one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EKROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, device or device.

[0302] Computer program code for performing the operations of the present invention can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, SMalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages, in particular, Python suitable for neural network computing and platform frameworks based on TensorFlow, PyTorch, etc. can be used. The program code can be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or to an external computer (for example, using an Internet service provider to connect through the Internet).

[0303] The above-mentioned device and non-transitory computer-readable storage medium can be referred to the detailed description of a power grid cost configuration method based on multimodal graph structure and fuzzy evaluation and its beneficial effects, which will not be repeated here.

[0304] Although the embodiments of the present invention have been shown and described above, it should be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. A person skilled in the art may change, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. A power grid cost configuration method based on multimodal graph structure and fuzzy evaluation, characterized in that: include: S1. Obtain the preset cost indicators of the power grid throughout its life cycle and build a full life cycle model of the power grid; S2. Build an evaluation system based on preset cost indicators, use the evaluation indicators in the evaluation system as nodes, and build a node set; calculate the mutual information based on the connection relationship between the nodes, establish edges between the nodes, and build an edge set; build a multimodal graph structure network based on the node set and edge set; S3. Based on the multimodal graph structure network, extract the multimodal feature vector and assign dynamic weights to the established edges; Obtain attention weights to weight the importance of nodes and obtain the dynamic weight vector in the current context; S4. Based on the dynamic weight vector, the fuzzy reasoning matrix is ​​embedded in the multimodal graph structure network to optimize the fuzzy reasoning rule process. The fuzzy weight of the fuzzy reasoning is defuzzified by the Bayesian weighted mind method to obtain the comprehensive defuzzification index weight; S5. Based on the weights of the comprehensive defuzzification indicators, a weighted calculation is performed on the full life cycle model to obtain the comprehensive cost of the full life cycle of the power grid.

2. The power grid cost configuration method based on multimodal graph structure and fuzzy evaluation according to claim 1 is characterized by: In step S1, the preset cost indicators include investment and construction costs, operating costs, maintenance costs, reliability failure costs, decommissioning costs, and electricity costs; The expression of the full life cycle model of the power grid is as follows: LCC=CI+CO+CM+CF+CD+CP; Where: CI is the investment and construction cost of the grid, CO is the operation cost of the grid, CM is the maintenance cost of the grid, CF is the reliability failure cost of the grid, CD is the decommissioning cost of the grid, and CP is the electricity cost of the grid; CI=C s +C line +C sj ; Where: C s is the investment cost of transmission tower construction, C line is the investment cost of distribution line construction, C sj Design costs for construction; CO=C sope +C gope +C zope ; Where: C sope is the substation operating cost, C gope is the operating cost of high-voltage distribution lines, C zope The operating cost of medium voltage distribution lines; Among them: PM i is the unit price of preventive maintenance for equipment i, Q i is the number of installations of device i, λ PM，i is the average annual preventive maintenance frequency of equipment i, CM j is the repair cost of faulty component k, f j is the failure probability of faulty component j, τ j is the mean repair time, r is the discount rate, and t is the t-th year in the whole life cycle; Of which: EENS k is the expected value of power shortage under fault scenario k, VOLL is the unit power outage loss value, RC k is the emergency repair cost of fault scenario k; CD=C d Toilets r β; Where: C d The cost of decommissioning, C r is the original value of the equipment, and β is the residual value rate; Where: P grid (t) is the purchased electricity in period t, ρ(t) is the time-of-use electricity price curve, C gen is the cost of self-provided power generation, P local is the self-generated power.

3. The power grid cost configuration method based on multimodal graph structure and fuzzy evaluation according to claim 2 is characterized by: The construction investment costs of transmission towers and distribution lines include: Where: C1 is the total investment cost of the transmission tower, t ope is the equipment operation time, r0 is the discount rate; Among them: z is the construction investment cost per unit length of the distribution line, γ2 is the tortuosity coefficient of the distribution line, J q is the set of loads supplied by the qth substation, J is the set of all load points, and l(q,j) is the straight-line distance between the qth substation and the jth load point; Substation operating costs, high-voltage distribution line operating costs, and medium-voltage distribution line operating costs, specifically including: Where: ΔP M，t is the power loss, S N is the rated capacity of the transformer, S2 is the secondary side load power of the transformer, ΔP0, ΔP k are the no-load active power loss and rated load active power loss of the transformer, β is the load factor, δ is the unit loss cost of the transformer, and ΔT is the operating time of the transformer; Where: P q is the load value of the qth substation, R g is the resistance per unit length of the high-voltage distribution line, U N is the rated voltage of the distribution line, γ1 is the tortuosity coefficient of the high-voltage distribution line, l(q) is the distance to the qth substation, δ is the network loss cost, and ΔT is the transformer operation time; Where: P j is the load value of the jth load point, R z is the resistance per unit length of the medium-voltage distribution line, γ2 is the tortuosity coefficient of the medium-voltage distribution line, and l(p,q) is the straight-line distance between the p-th substation and the q-th load point.

4. The power grid cost configuration method based on multimodal graph structure and fuzzy evaluation according to claim 1 is characterized by: The step S2 specifically includes: S21. Using the preset cost index as the evaluation index, construct an evaluation system, and use different evaluation indexes in the evaluation system as different types of nodes to construct a node set V, where V = {v1, v2, ..., v i ,...,v n }, |V| = n; S22. Determine the size of the edge set E based on the connection relationship and connection strength between the nodes: If the connection relationship between the nodes is a complete graph, the size of the edge set E is as follows: Where: n is the number of nodes; If the connection relationship between nodes is a sparse graph, the mutual information between nodes is calculated, and its expression is as follows: I(i,j)=lg(P(i,j) / (P(i)P(j))); Where: I(i, j) is the mutual information between nodes i and j, P(i, j) is the joint probability distribution of nodes i and j appearing at the same time, P(i) and P(j) are the marginal probability distributions of nodes i and j appearing respectively; Based on the mean mutual information between nodes, a dynamic adaptive threshold is set, which is expressed as follows: θ=μ+ασ; Where: θ is the adaptive threshold, μ is the mean of mutual information, α is the learnable parameter, and σ is the standard deviation; Determine whether the mutual information between node i and node j is greater than the adaptive threshold θ; if so, establish an edge between node i and node j; Assume that the feature vectors of node i and node j are as follows: X i =[x1,x2,...,x m ],X j =[y1,y2,...,y m ]; Where: X i is the feature vector of node i, X j is the feature vector of node j, and m is the number of samples; Based on the feature vectors of node i and node j, determine the connection strength S between node i and node j ij , which is expressed as follows: A global or local threshold θ is set based on the connection strength, and the top 20% of the threshold θ strength distribution is used as the screening condition to filter and retain the edges that meet the conditions. The expression is as follows: E={(ij)|S ij ≥θ}; Then the size of the edge set E is as follows: Where: d i is the number of edges connected to node i; S23. Based on the node set V and the edge set E, construct a multimodal graph structure network G = (V, E).

5. The power grid cost configuration method based on multimodal graph structure and fuzzy evaluation according to claim 1 is characterized by: The step S3 specifically includes: S31. The initial feature matrix of each node in the multimodal graph structure network contains text data, image data, and time series data. The multimodal feature vector is extracted from the initial feature matrix, and its expression is as follows: in: The feature vector generated by the BioBERT model after encoding the text data, It is the feature vector generated by the ResNet model after encoding the image data. X is the feature vector generated by the LSTM model after encoding the time series data. i is the initial feature matrix; S32. Normalize the multimodal feature vectors and assign dynamic weights to the established edges based on a multi-layer perceptron. Capture the importance between nodes, obtain attention weights, and perform layer-by-layer propagation and fusion to obtain new node representations, which are expressed as follows: in: is the graph structure network where node i is at layer 0, sig is the sigmoid function, H i is the multimodal feature vector; in: is a graph structure network with node i at layer l+1, is the graph structure network of node j in layer l, σ is the activation function, c ij is the normalization coefficient, a ij is the attention weight of node j to node i, is the attention vector, w ij is the weight matrix, N(i) is the neighborhood set of node i, || is the feature concatenation operation, Leaky ReLU is the nonlinear transformation function, MLP is the multi-layer perceptron, H k is the feature representation of node k in the neighborhood of node i, b l is the bias term; S33. Weight the importance of the node by the attention weight to obtain the dynamic weight vector in the current situation, which is expressed as follows: Where: W t is the dynamic weight vector at time t, is the output feature of node i in the Lth layer of the multi-layer graph structure network.

6. The power grid cost configuration method based on multimodal graph structure and fuzzy evaluation according to claim 5 is characterized by: The step S4 specifically includes: S41. Embed the fuzzy reasoning matrix in the multimodal graph structure network and perform multi-layer fuzzy reasoning convolution calculation. Its expression is as follows: in: is the l+1 layer output of the fuzzy inference matrix, A t is the adaptive adjacency matrix, is the l-layer output of the fuzzy inference matrix, r k is the vector representation of the k-th rule; Where: Z t is the embedding representation learned from the fuzzy inference matrix, d i d j are the degrees of node i and node j respectively; S42. Mapping fuzzy inference rules R′ into node features of a multimodal graph structure network Perform adaptive multi-layer graph structure network calculation to obtain the fuzzy weight of fuzzy reasoning, which is expressed as follows: R'=AdaptiveGNN(R,W t ,A); R＝{r1，r2，...，r k ，...，r m}； Among them: AdaptiveGNN is an adaptive multi-layer graph structure network, R is the original fuzzy matrix, A is the static adjacency matrix, r k is the vector representation of the kth fuzzy inference rule; S43. Defuzzify the fuzzy weight of fuzzy reasoning by Bayesian weighted method to obtain the comprehensive defuzzification index weight, which is expressed as follows: in: is the comprehensive defuzzification index weight, x i is the fuzzy weight of fuzzy reasoning, μ(x i ) is the fuzzy membership function, P(μ(x i ) is the fuzzy membership function μ(x i ), b l is the bias term.

7. The power grid cost configuration method based on multimodal graph structure and fuzzy evaluation according to claim 6 is characterized by: The step S5 specifically includes: S51. Based on the weights of the comprehensive defuzzification indicators, the investment and construction costs, operating costs, maintenance costs, reliability failure costs, decommissioning costs, and electricity costs in the full life cycle model of the power grid are weighted to obtain the comprehensive cost of the full life cycle of the power grid, which is expressed as follows: LCC t =CI t +CO t +CM t +CF t +CD t +CP t ; Of which: LCC t is the weighted comprehensive cost of the entire life of the power grid, CI t 、CO t CM t CF t 、CD t 、CP t They are the weighted investment and construction cost, operation cost, maintenance cost, reliability failure cost, decommissioning cost, and electricity cost of the power grid; in: is the dynamic weight of the investment and construction cost of the power grid after weighting based on the comprehensive defuzzification index weights, CI is the investment and construction cost of the power grid; in: is the dynamic weight of the grid operation cost after weighting based on the comprehensive defuzzification index weights, CO is the grid operation cost; in: is the dynamic weight of the maintenance cost of the power grid after weighting based on the comprehensive defuzzification index weights, CM is the maintenance cost of the power grid; in: is the dynamic weight of the reliability failure cost of the power grid after weighting based on the comprehensive defuzzification index weights, CF is the reliability failure cost of the power grid; in: is the dynamic weight of the power grid decommissioning cost after weighting based on the comprehensive defuzzification index weights, CD is the power grid decommissioning cost; in: is the dynamic weight of the power cost of the power grid after weighting based on the comprehensive defuzzification index weights, and CP is the power cost of the power grid.

8. A power grid cost configuration system based on multimodal graph structure and fuzzy evaluation, characterized in that: The system is applied to the method according to any one of claims 1 to 7, and the system comprises: A full life cycle model construction module (1) is used to obtain preset cost indicators of the power grid during its full life cycle and to construct a full life cycle model of the power grid; The multimodal graph structure network construction module (2) is used to construct an evaluation system based on a preset cost indicator, use the evaluation indicators in the evaluation system as nodes, and construct a node set; calculate the mutual information based on the connection relationship between the nodes, establish the edges between the nodes, and construct an edge set; and construct a multimodal graph structure network based on the node set and the edge set; The dynamic weight vector acquisition module (3) is used to extract multimodal feature vectors based on the multimodal graph structure network and assign dynamic weights to the established edges; obtain the attention weight to weight the importance of the node and obtain the dynamic weight vector in the current situation; The fuzzy reasoning optimization and defuzzification module (4) is used to embed the fuzzy reasoning matrix in the multimodal graph structure network based on the dynamic weight vector, optimize the fuzzy reasoning rule process, and defuzzify the fuzzy weight of the fuzzy reasoning through the Bayesian weighted mind method to obtain the comprehensive defuzzification index weight; The power grid life cycle comprehensive cost calculation module (5) is used to perform weighted calculation on the full life cycle model based on the comprehensive defuzzification index weights to obtain the power grid life cycle comprehensive cost.

9. A power grid cost configuration device based on multimodal graph structure and fuzzy evaluation, characterized by: The device comprises a processor (6) and a memory (7); the memory (7) is used to store computer program code (71) and transmit the computer program code (71) to the processor (6); The processor (6) is configured to execute the power grid cost configuration method based on multimodal graph structure and fuzzy evaluation according to any one of claims 1 to 7 according to instructions in the computer program code (71).

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions. When the computer-executable instructions are executed on a computer, the power grid cost configuration method based on multimodal graph structure and fuzzy evaluation according to any one of claims 1 to 7 is implemented.

Images