Substation equipment state evaluation system and method based on AHP-entropy weight fusion and improved fuzzy comprehensive evaluation

Abstract

The invention relates to a substation equipment state evaluation system based on AHP-entropy weight fusion and improved fuzzy comprehensive evaluation. The substation equipment state evaluation system comprises a feature acquisition module, an adaptive scoring module, a membership calculation module, a weight calculation module, a fuzzy synthesis module and a classification optimization module. The invention further relates to a substation equipment state evaluation method based on AHP-entropy weight fusion and improved fuzzy comprehensive evaluation. The method comprises the steps that S1, a multi-strategy nonlinear scoring rule is established; s2, constructing a dynamic membership function; s3, carrying out an AHP-entropy weight fusion algorithm; s4, improving a fuzzy synthesis operator; and S5, integrating a classification optimization strategy. According to the method, the degree of fitting of the equipment degradation law is improved to 92% through nonlinear scoring, and the weight error is reduced by 35% through the AHP-entropy weight fusion algorithm; the evaluation system provided by the invention is tested on 5000 groups of simulation data, the recall rate of a'serious anomaly 'class reaches 93.5%, and meanwhile, the time consumed by single evaluation is as long as lt; 2ms.

Description

Technical Field

[0001] This invention belongs to the field of substation equipment condition assessment technology, specifically relating to a substation equipment condition assessment system and method based on AHP-entropy weight fusion and improved fuzzy comprehensive evaluation. Background Technology

[0002] As a core tool for power equipment operation and maintenance management, substation condition assessment systems have evolved from traditional periodic maintenance models to modern systems featuring "real-time monitoring and intelligent diagnosis." Current advanced condition assessment systems exhibit several significant technical characteristics, which collectively drive profound changes in substation operation and maintenance management models.

[0003] A modern substation condition assessment system is a complex, comprehensive technical platform. Its architecture typically employs a layered design, dividing the system into multiple functionally defined and collaborative components. These components work together to complete the entire process from data acquisition to condition assessment and decision support. Its main components include real-time monitoring and data acquisition, data storage and processing, and intelligent analysis and condition assessment.

[0004] As the foundational component of the system, the data acquisition layer bears the crucial responsibility of acquiring equipment status information. This layer consists of various sensors and monitoring devices deployed on different types of equipment in the substation, forming a sensing network covering the entire station. The data acquisition scope of modern assessment systems has expanded from traditional electrical parameters (such as voltage and current) to include diverse status parameters such as temperature, humidity, partial discharge, and oil chromatography.

[0005] The data storage and processing layer is responsible for organizing, storing, and analyzing monitoring data. It introduces big data technology and establishes a unified data model centered on the equipment. For data storage, distributed storage and parallel computing technologies are employed to construct a distributed parallel computing platform for power transmission and transformation equipment condition assessment, enabling rapid processing of massive amounts of data based on big data technology.

[0006] The intelligent analysis layer achieves in-depth assessment of equipment status through various algorithm models. This layer typically includes multiple functional modules: the basic status monitoring module handles fundamental functions such as data visualization and over-limit alarms; the advanced analysis module utilizes machine learning, pattern recognition, and other technologies to achieve fault diagnosis and prediction. By employing multiple algorithms to establish a more comprehensive analysis system for real-time monitoring data, including advanced functions such as dynamic assessment of equipment load capacity, differentiated status evaluation, fault prediction, and risk assessment, it provides a more comprehensive reference for operation and maintenance decisions.

[0007] With the deepening of smart grid construction, substation condition assessment systems have made significant progress in terms of technology and application scope. However, in actual operation, the following problems still exist due to limitations in the application of intelligent algorithms and the scientific nature of assessment methods: 1. The weight allocation is highly subjective: the traditional fuzzy comprehensive evaluation method relies on expert experience to set weights and does not consider the dynamic correlation between features.

[0008] 2. Linearity of scoring rules: Existing methods use a linear deduction system (such as deducting a fixed number of points for each additional refusal to operate), which cannot reflect the non-linear characteristics of equipment degradation.

[0009] 3. Uniqueness of membership functions: Traditional triangular or trapezoidal membership functions do not take into account the actual data distribution, resulting in insufficient evaluation resolution.

[0010] 4. Class imbalance problem: The "severe anomaly" class accounts for less than 5% of the equipment status samples, resulting in poor generalization ability of the classification model.

[0011] 5. Limitations of the synthesis operator: The traditional M(∧,∨) operator only retains local extremum information and loses the contribution of features in other dimensions. Summary of the Invention

[0012] The purpose of this invention is to overcome the shortcomings of the prior art and provide a substation equipment condition assessment system and method based on AHP-entropy weight fusion and improved fuzzy comprehensive evaluation.

[0013] The technical problem solved by this invention is achieved through the following technical solution: A substation equipment condition assessment system based on AHP-entropy weight fusion and improved fuzzy comprehensive evaluation includes a feature acquisition module, an adaptive scoring module, a membership degree calculation module, a weight calculation module, a fuzzy synthesis module, and a classification optimization module. The feature acquisition module acquires eight features: Class I alarms, Class II alarms, number of no-operation times, number of false-operation times, family defects, implementation degree of countermeasures, device defects, and service life. The adaptive scoring module generates an 8×1 scoring vector using nonlinear scoring rules. The membership calculation module calculates five levels of membership based on the Gaussian kernel function. The membership weight calculation module generates an 8-dimensional weight vector using the AHP-entropy weight fusion algorithm. The fuzzy synthesis module performs weighted Euclidean synthesis to obtain a comprehensive state evaluation value. The classification optimization module uses the SMOTE oversampling and SVM hyperparameter search algorithms to complete the state classification.

[0014] A substation equipment condition assessment method based on AHP-entropy weight fusion and improved fuzzy comprehensive evaluation, employing the aforementioned assessment system, comprises the following steps: S1. Multi-strategy nonlinear scoring rules: Design differentiated scoring functions such as exponential, step, and logarithmic scores for different feature types; S2. Construction of dynamic membership function: Combining Gaussian kernel function and data clustering results, adaptively adjust membership center value and shape parameter; S3, AHP-Entropy Weight Fusion Algorithm: The baseline weight is determined by AHP verified by the consistency ratio (CR), and dynamically adjusted by combining information entropy correction; S4. Improved fuzzy synthesis operator: Weighted Euclidean synthesis is used to replace the traditional small-large operator, preserving full-dimensional feature information; S5. Integrated classification optimization strategy: Combining SMOTE oversampling and SVM hyperparameter search algorithm to improve the recognition rate of rare categories.

[0015] Moreover, S1 specifically refers to: Input: Original feature vector X = (x1,x2,...,x8); Output: Rating vector S = (s1,s2,...,s8), where S i ∈ [0,100] ; - Class I alarms (critical failure events): exponential decay S1 = 100 - 15 × (x1^1.5); - Class II alarms (minor fault events): linear decay; S² = 100 - 8 × x²; - Refusal to move / malfunction (action reliability): Step-wise decay; S3 = 100 - 20 × min(x3,3) (20 points will be deducted for each of the first 3 attempts) S4 = 100 - 15 × min(x4,4) (15 points will be deducted for each of the first 4 attempts) - Family defect (systematic risk): log decay; S5 = 100 - 25 × ln(x5+1); - Implementation of countermeasures (preventive measures): Linear positive indicators S6 = 100 - 5 × x6; - Equipment defect (current service issue): linear decay; S7 = 100 - 7 × x7; - Service life (equipment aging): segmented degradation; If x8 ≤ 3: S8 = 100 - 3 × x8, deduct 3 points for each of the first 3 years; Otherwise: S8 = 85 - 10 × (x8 - 3), and 10 points will be deducted each year after 3 years; - Exponential decay: S i = 100 - a × x^b, Class I alarm: a=15, b=1.5; - Stepped decay: S i = 100 - c × min(x,d), Rejected: c=20, d=3; - Logarithmic decay: S i = 100 - e × ln(x+1), family defect: e=25; - Piecewise function: S i = { 100-3x (x ≤ 3); 85-10(x-3) (x > 3)}.

[0016] Moreover, S2 specifically refers to: Input: rating vector S; Output: Membership matrix R ∈ ^(8×5); 1) The five-level state center values ​​were determined by K-means clustering, and the optimal center values ​​were [85, 75, 65, 55, 45]). 2) Construct the Gaussian membership function: rᵢⱼ = e^(-k × (sᵢ - cⱼ)^2); Where: k = 0.02, shape parameter; cⱼ = [85,75,65,55,45], fifth-level state center value; 3) Parameter optimization: The shape parameter k is determined by grid search, with a step size of 0.01~0.05 and a center value of 0.005; the center value is dynamically updated by clustering historical data.

[0017] Moreover, S3 specifically refers to: Input: Feature scoring matrix S ∈ ^(n×8); Output: Fusion weight vector W ∈ ^8; 1) Construct the judgment matrix: Construct a reciprocal matrix A by comparing each of the 35 pairs of terms (satisfying aᵢⱼ = 1 / aⱼᵢ). 2) AHP weight calculation: Calculate the maximum eigenvalue λ_max and the corresponding eigenvector W_AHP; Consistency check: CR = (λ_max - n) / [(n-1)×RI] < 0.1, otherwise the judgment matrix is ​​automatically corrected; 3) Entropy weight correction: Calculate information entropy: eⱼ = -Σ(pᵢⱼ × ln pᵢⱼ); Calculate weight: wⱼ = (1 - eⱼ) / Σ(1 - eⱼ); 4) Fusion weight: W_final = αW_AHP + (1-α)W_entropy, where α=0.6 is the optimal value verified by experiments.

[0018] Moreover, S4 specifically refers to: Input: Membership matrix R ∈ ^(8×5), weight vector W ∈ ^8; Output: Comprehensive evaluation value vector B ∈ ^5; Improved operator: Bⱼ = Σ(wᵢ × rᵢⱼ), i=1~8.

[0019] Moreover, S5 specifically refers to: 1) SMOTE oversampling: The "Severe Anomaly" class (label 4) is oversampled by 5 times, and the K-nearest neighbor parameter k=5; 2) SVM hyperparameter search: Parameter range: Regularization parameter C: [0.1, 0.5, 1, 5, 10]; Kernel width γ: [0.01, 0.05, 0.1, 0.5, 1]; Kernel function: {rbf, poly, sigmoid}; 3) Multi-core learning strategy: Dynamically select the optimal kernel function and compare the performance of different kernel functions through cross-validation.

[0020] The advantages and beneficial effects of the invention are as follows: 1. Scoring rule optimization: Non-linear scoring improves the fitting accuracy of equipment degradation patterns to 92% (from 78%).

[0021] 2. Weight allocation accuracy: The AHP-entropy weight fusion algorithm reduces weight error by 35%.

[0022] 3. Classification performance: Tested on 5000 sets of simulated data, the recall rate of the "serious anomaly" class reached 93.5% (originally 68.2%).

[0023] 4. Computational efficiency: Single evaluation time <2ms (Intel i7-11800H). Detailed Implementation

[0024] The present invention will be further described in detail below through specific embodiments. The following embodiments are merely descriptive and not limiting, and should not be used to limit the scope of protection of the present invention.

[0025] A substation equipment condition assessment system and method based on AHP-entropy weight fusion and improved fuzzy comprehensive evaluation, the innovation of which lies in: An innovative substation equipment condition assessment system based on AHP-entropy weight fusion and improved fuzzy comprehensive evaluation includes a feature acquisition module, an adaptive scoring module, a membership degree calculation module, a weight calculation module, a fuzzy synthesis module, and a classification optimization module. The feature acquisition module acquires eight features: Class I alarms, Class II alarms, number of no-operation times, number of false-operation times, family defects, implementation degree of countermeasures, device defects, and service life. The adaptive scoring module generates an 8×1 scoring vector using nonlinear scoring rules. The membership calculation module calculates five levels of membership based on the Gaussian kernel function. The membership weight calculation module generates an 8-dimensional weight vector using the AHP-entropy weight fusion algorithm. The fuzzy synthesis module performs weighted Euclidean synthesis to obtain a comprehensive state evaluation value. The classification optimization module uses the SMOTE oversampling and SVM hyperparameter search algorithms to complete the state classification.

[0026] A substation equipment condition assessment method based on AHP-entropy weight fusion and improved fuzzy comprehensive evaluation is innovative in that it employs the aforementioned assessment system, and the method comprises the following steps: S1. Multi-strategy nonlinear scoring rules: Design differentiated scoring functions such as exponential, step, and logarithmic scores for different feature types; S2. Construction of dynamic membership function: Combining Gaussian kernel function and data clustering results, adaptively adjust membership center value and shape parameter; S3, AHP-Entropy Weight Fusion Algorithm: The baseline weight is determined by AHP verified by the consistency ratio (CR), and dynamically adjusted by combining information entropy correction; S4. Improved fuzzy synthesis operator: Weighted Euclidean synthesis is used to replace the traditional small-large operator, preserving full-dimensional feature information; S5. Integrated classification optimization strategy: Combining SMOTE oversampling and SVM hyperparameter search algorithm to improve the recognition rate of rare categories.

[0027] Specifically, S1 is: Input: Original feature vector X = (x1,x2,...,x8); Output: Rating vector S = (s1,s2,...,s8), where S i ∈ [0,100] ; - Class I alarms (critical failure events): exponential decay S1 = 100 - 15 × (x1^1.5); - Class II alarms (minor fault events): linear decay; S² = 100 - 8 × x²; - Refusal to move / malfunction (action reliability): Step-wise decay; S3 = 100 - 20 × min(x3,3) (20 points will be deducted for each of the first 3 attempts) S4 = 100 - 15 × min(x4,4) (15 points will be deducted for each of the first 4 attempts) - Family defect (systematic risk): log decay; S5 = 100 - 25 × ln(x5+1); - Implementation of countermeasures (preventive measures): Linear positive indicators S6 = 100 - 5 × x6; - Equipment defect (current service issue): linear decay; S7 = 100 - 7 × x7; - Service life (equipment aging): segmented degradation; If x8 ≤ 3: S8 = 100 - 3 × x8, deduct 3 points for each of the first 3 years; Otherwise: S8 = 85 - 10 × (x8 - 3), and 10 points will be deducted each year after 3 years; - Exponential decay: S i = 100 - a × x^b, Class I alarm: a=15, b=1.5; - Stepped decay: S i = 100 - c × min(x,d), Rejected: c=20, d=3; - Logarithmic decay: S i = 100 - e × ln(x+1), family defect: e=25; - Piecewise function: S i = { 100-3x (x ≤ 3); 85-10(x-3) (x > 3)}.

[0028] Specifically, S2 is: Input: rating vector S; Output: Membership matrix R ∈ ^(8×5); 1) The five-level state center values ​​were determined by K-means clustering, and the optimal center values ​​were [85, 75, 65, 55, 45]). 2) Construct the Gaussian membership function: rᵢⱼ = e^(-k × (sᵢ - cⱼ)^2); Where: k = 0.02, shape parameter; cⱼ = [85,75,65,55,45], fifth-level state center value; 3) Parameter optimization: The shape parameter k is determined by grid search, with a step size of 0.01~0.05 and a center value of 0.005; the center value is dynamically updated by clustering historical data.

[0029] Specifically, S3 is: Input: Feature scoring matrix S ∈ ^(n×8); Output: Fusion weight vector W ∈ ^8; 1) Construct the judgment matrix: Construct a reciprocal matrix A by comparing each of the 35 pairs of terms (satisfying aᵢⱼ = 1 / aⱼᵢ). 2) AHP weight calculation: Calculate the maximum eigenvalue λ_max and the corresponding eigenvector W_AHP; Consistency check: CR = (λ_max - n) / [(n-1)×RI] < 0.1, otherwise the judgment matrix is ​​automatically corrected; 3) Entropy weight correction: Calculate information entropy: eⱼ = -Σ(pᵢⱼ × ln pᵢⱼ); Calculate weight: wⱼ = (1 - eⱼ) / Σ(1 - eⱼ); 4) Fusion weight: W_final = αW_AHP + (1-α)W_entropy, where α=0.6 is the optimal value verified by experiments.

[0030] Specifically, S4 is: Input: Membership matrix R ∈ ^(8×5), weight vector W ∈ ^8; Output: Comprehensive evaluation value vector B ∈ ^5; Improved operator: Bⱼ = Σ(wᵢ × rᵢⱼ), i=1~8.

[0031] Specifically, S5 is: 1) SMOTE oversampling: The "Severe Anomaly" class (label 4) is oversampled by 5 times, and the K-nearest neighbor parameter k=5; 2) SVM hyperparameter search: Parameter range: Regularization parameter C: [0.1, 0.5, 1, 5, 10]; Kernel width γ: [0.01, 0.05, 0.1, 0.5, 1]; Kernel function: {rbf, poly, sigmoid}; 3) Multi-kernel learning strategy: Dynamically select the optimal kernel function and compare the performance of different kernel functions through cross-validation. Although embodiments of the invention have been disclosed for illustrative purposes, those skilled in the art will understand that various substitutions, variations, and modifications are possible without departing from the spirit and scope of the invention and the appended claims. Therefore, the scope of the invention is not limited to the contents disclosed in the embodiments.

Claims

1. A substation equipment condition assessment system based on AHP-entropy weight fusion and improved fuzzy comprehensive evaluation, characterized in that: It includes a feature acquisition module, an adaptive scoring module, a membership calculation module, a weight calculation module, a fuzzy synthesis module, and a classification optimization module; The feature acquisition module acquires eight features: Class I alarms, Class II alarms, number of times the device refused to operate, number of times the device malfunctioned, family defects, implementation degree of countermeasures, device defects, and service life. The adaptive scoring module generates an 8×1 scoring vector using nonlinear scoring rules; the membership calculation module calculates five levels of membership based on the Gaussian kernel function; the membership weight calculation module generates an 8-dimensional weight vector using the AHP-entropy weight fusion algorithm; the fuzzy synthesis module performs weighted Euclidean synthesis to obtain a comprehensive state evaluation value; and the classification optimization module uses the SMOTE oversampling and SVM hyperparameter search algorithms to complete state classification.

2. A method for assessing the condition of substation equipment based on AHP-entropy weight fusion and improved fuzzy comprehensive evaluation, characterized in that: Using the evaluation system as described in claim 1, the steps of the method are as follows: S1. Multi-strategy nonlinear scoring rules: Design differentiated scoring functions such as exponential, step, and logarithmic scores for different feature types; S2. Construction of dynamic membership function: Combining Gaussian kernel function and data clustering results, adaptively adjust membership center value and shape parameter; S3, AHP-Entropy Weight Fusion Algorithm: The baseline weight is determined by AHP verified by the consistency ratio (CR), and dynamically adjusted by combining information entropy correction; S4. Improved fuzzy synthesis operator: Weighted Euclidean synthesis is used to replace the traditional small-large operator, preserving full-dimensional feature information; S5. Integrated classification optimization strategy: Combining SMOTE oversampling and SVM hyperparameter search algorithm to improve the recognition rate of rare categories.

3. The substation equipment condition assessment method based on AHP-entropy weight fusion and improved fuzzy comprehensive evaluation according to claim 2, characterized in that: Specifically, S1 is: Input: Original feature vector X = (x1,x2,...,x8); Output: Rating vector S = (s1,s2,...,s8), where S i ∈ [0,100] ; - Class I alarms (critical failure events): exponential decay S1 = 100 - 15 × (x1^1.5); - Class II alarms (minor fault events): linear decay; S² = 100 - 8 × x²; - Refusal to move / malfunction (action reliability): Step-wise decay; S3 = 100 - 20 × min(x3,3) (20 points will be deducted for each of the first 3 attempts) S4 = 100 - 15 × min(x4,4) (15 points will be deducted for each of the first 4 attempts) - Family defect (systematic risk): log decay; S5 = 100 - 25 × ln(x5+1); - Implementation of countermeasures (preventive measures): Linear positive indicators S6 = 100 - 5 × x6; - Equipment defect (current service issue): linear decay; S7 = 100 - 7 × x7; - Service life (equipment aging): segmented degradation; If x8 ≤ 3: S8 = 100 - 3 × x8, deduct 3 points for each of the first 3 years; Otherwise: S8 = 85 - 10 × (x8 - 3), and 10 points will be deducted each year after 3 years; - Exponential decay: S i = 100 - a × x^b, Class I alarm: a=15, b=1.5; - Stepped decay: S i = 100 - c × min(x,d), Rejected: c=20, d=3; - Logarithmic decay: S i = 100 - e × ln(x+1), family defect: e=25; - Piecewise function: S i = { 100-3x (x ≤ 3); 85-10(x-3) (x > 3)} 。 4. The substation equipment condition assessment method based on AHP-entropy weight fusion and improved fuzzy comprehensive evaluation according to claim 2, characterized in that: Specifically, S2 is: Input: rating vector S; Output: Membership matrix R ∈ ^(8×5); 1) The five-level state center values ​​were determined by K-means clustering, and the optimal center values ​​were [85, 75, 65, 55, 45]). 2) Construct the Gaussian membership function: rᵢⱼ = e^(-k × (sᵢ - cⱼ)^2); Where: k = 0.02, shape parameter; cⱼ = [85,75,65,55,45], fifth-level state center value; 3) Parameter optimization: The shape parameter k is determined by grid search, with a step size of 0.01~0.05 and a center value of 0.005; the center value is dynamically updated by clustering historical data.

5. The substation equipment condition assessment method based on AHP-entropy weight fusion and improved fuzzy comprehensive evaluation according to claim 2, characterized in that: Specifically, S3 is: Input: Feature scoring matrix S ∈ ^(n×8); Output: Fusion weight vector W ∈ ^8; 1) Construct the judgment matrix: Construct a reciprocal matrix A by comparing each of the 35 pairs of terms (satisfying aᵢⱼ = 1 / aⱼᵢ). 2) AHP weight calculation: Calculate the maximum eigenvalue λ_max and the corresponding eigenvector W_AHP; Consistency check: CR = (λ_max - n) / [(n-1)×RI] < 0.1, otherwise the judgment matrix is ​​automatically corrected; 3) Entropy weight correction: Calculate information entropy: eⱼ = -Σ(pᵢⱼ × ln pᵢⱼ); Calculate the weights: wⱼ = (1 - eⱼ) / Σ(1 - eⱼ); 4) Fusion weight: W_final = αW_AHP + (1-α)W_entropy, where α=0.6 is the optimal value verified by experiments.

6. The substation equipment condition assessment method based on AHP-entropy weight fusion and improved fuzzy comprehensive evaluation according to claim 2, characterized in that: Specifically, S4 is: Input: Membership matrix R ∈ ^(8×5), weight vector W ∈ ^8; Output: Comprehensive evaluation value vector B ∈ ^5; Improved operator: Bⱼ = Σ(wᵢ × rᵢⱼ), i=1~8.

7. The substation equipment condition assessment method based on AHP-entropy weight fusion and improved fuzzy comprehensive evaluation according to claim 2, characterized in that: Specifically, S5 is: 1) SMOTE oversampling: The "Severe Anomaly" class (label 4) is oversampled by 5 times, and the K-nearest neighbor parameter k=5; 2) SVM hyperparameter search: Parameter range: Regularization parameter C: [0.1, 0.5, 1, 5, 10]; Nuclear width γ: [0.01, 0.05, 0.1, 0.5, 1]; Kernel functions: {rbf, poly, sigmoid}; 3) Multi-core learning strategy: Dynamically select the optimal kernel function and compare the performance of different kernel functions through cross-validation.

Images