Load identification capability comprehensive evaluation method based on fusion decision

Abstract

The invention relates to the technical field of power systems, in particular to a load identification capability comprehensive evaluation method based on fusion decision, which comprises the following steps: constructing an evaluation index system of load identification; constructing a fuzzy cognitive map model, connecting the subjective weight matrixes through a fuzzy relation by adopting a fuzzy cognitive map, and obtaining an optimized weight matrix W'by utilizing fuzzy C-means clustering; entropy calculation is carried out through an entropy weight method, a membership degree value of the to-be-evaluated load identification method is obtained through relation matrix multiplication of a fuzzy comprehensive evaluation analysis method, a final evaluation score is calculated, and a fuzzy comprehensive evaluation result is obtained; by multiplying a comprehensive evaluation value obtained by quantization of a comment set by a normalized fuzzy comprehensive evaluation matrix, a final evaluation score integrating all evaluation index influences can be obtained, and non-intrusive load identification capability can be sorted through the score. Therefore, the optimal load identification scheme can be selected in an auxiliary manner or the existing scheme can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power systems, and in particular to a comprehensive evaluation method for load identification capability based on fusion decision-making. Background Art

[0002] Load monitoring technology is an effective way to strengthen the power demand management capability and promote energy conservation and orderly power consumption. Load monitoring of power equipment can be divided into two main types: intrusive load monitoring and non-intrusive load monitoring. Non-intrusive monitoring technology is gradually replacing intrusive monitoring due to its advantages of non-entry and easy maintenance.

[0003] Non-intrusive load monitoring estimates the working and energy consumption status of each electrical equipment based on the electrical parameters at the power entrance, and obtains detailed power consumption status information through load identification technology.

[0004] The comprehensive evaluation method of load identification capability is an important part of power system operation and management. In modern power systems, accurate identification of load identification capability plays a key role in achieving reliable power supply, rationally planning power equipment and optimizing operation and dispatching. The comprehensive evaluation method of load identification capability can provide a decision-making basis in operation and management by analyzing and evaluating the load status of the system.

[0005] Among the existing comprehensive evaluation methods for load identification capabilities, the fuzzy cognitive map (FCM) is an effective tool for processing uncertain information. It can simulate the human thinking process and perform fuzzy classification and cluster analysis on factors in complex systems. The analytic hierarchy process (AHP) is a structured decision analysis method that allows decision makers to determine the relative weights of different factors by comparing their importance in pairs. The fuzzy comprehensive evaluation method (FCE) can select the most relevant features from a large amount of load data and improve the accuracy and efficiency of load identification.

[0006] At present, the existing technology has the following defects in the process of evaluating the composite recognition ability:

[0007] 1. The existing comprehensive evaluation method of load identification capability lacks sufficient consideration of the correlation between different indicators and cannot accurately reflect the importance of each indicator;

[0008] 2. When the amount of load data is large, the traditional FCE algorithm has too much calculation and is slow; the traditional AHP method has defects, it is difficult to fully consider the relationship between indicators, and the uncertainty information is not well processed. In addition, the complexity of the power system load requires that the evaluation method must be able to adapt to a variety of different working conditions and changes, which requires the evaluation method to be highly flexible and adaptable. Summary of the invention

[0009] In order to solve the shortcomings of the prior art, the present invention provides a comprehensive evaluation method for load identification capability based on fusion decision-making, introduces fuzzy cognitive map (FCM), combines improved analytic hierarchy process (AHP) and fuzzy comprehensive evaluation (FCE) method, constructs fuzzy cognitive map to fuzzy represent various factors of the system, and then uses improved AHP-FCE method for comprehensive evaluation, thereby achieving accurate evaluation of the system load identification capability.

[0010] The technical solution of the present invention is as follows:

[0011] A comprehensive evaluation method for load identification capability based on fusion decision-making includes the following steps:

[0012] S1: construct an evaluation index system for load identification, apply the load identification method to be evaluated to the load characteristic data set to be identified, and obtain the load identification results of each evaluation index;

[0013] S2: The judgment matrix calculated using the importance ratio scale needs to be checked for consistency to obtain the subjective weight matrix W of each evaluation index;

[0014] S3: Construct a fuzzy cognitive graph model, use the fuzzy cognitive graph to connect the subjective weight matrix W through fuzzy relations, and use fuzzy C-means clustering to obtain the optimized weight matrix W′;

[0015] S4: Calculate the entropy value by entropy weight method to obtain the objective weight matrix W′ j ;

[0016] S5: Take the weighted average method to weight W′ and W′ j Perform fusion calculation to obtain the comprehensive weight matrix W f ;

[0017] S6: The comprehensive weight matrix W f The membership value of the load identification method to be evaluated is obtained by multiplying it with the relationship matrix of the fuzzy comprehensive evaluation analysis method, and the final evaluation score M is calculated to obtain the fuzzy comprehensive evaluation result.

[0018] Further, in S1, the load identification evaluation index system includes classification performance index and application performance index, wherein the classification performance index includes accuracy P C, recall rate R E , precision R C , the harmonic mean H1 of recall and precision; application performance indicators include training time T and number of samples S;

[0019] According to the key characteristics and requirements of load identification, the evaluation index system is set up into three levels from top to bottom: target layer, criterion layer and indicator layer.

[0020] Furthermore, in S1, the load identification methods to be evaluated include a load identification method based on a convolutional neural network, a load identification method based on a recurrent neural network, and a load identification method based on a bidirectional long short-term memory network optimization algorithm. The above methods are applied to the filtered and screened PLAID data set to perform load identification on various types of electrical equipment.

[0021] Furthermore, each indicator of the indicator layer is defined as:

[0022] Accuracy: P C =(TP+TN) / (P+N)

[0023] Recall: R E =TP / (TP+FN)

[0024] Accuracy: R C =TP / (TP+FP)

[0025] Harmonic mean:

[0026] Training time metrics:

[0027] Sample number indicators:

[0028] In the formula, TP represents the number of positive samples identified as positive, that is, true positive examples; FP represents the number of negative samples identified as positive, that is, false positive examples; FN represents the number of positive samples identified as negative, that is, false negative examples; TN represents the number of negative samples identified as negative, that is, true negative examples; P represents the number of positive samples, N represents the number of negative samples, T represents the time of algorithm training, in minutes, and S represents the number of samples participating in the algorithm training.

[0029] Furthermore, in S2, the importance ratio scale calculation judgment matrix needs to be subjected to consistency check, and the checking process includes the following steps:

[0030] S2.1: Based on the evaluation index hierarchy model constructed in S1, a judgment matrix is ​​formed by comparing the importance ratio scales in pairs;

[0031] S2.2: Calculate the maximum eigenvalue λ of the judgment matrix max, consistency index CI and random consistency ratio CR. When CR is less than 0.1, the judgment matrix is ​​considered to have satisfactory consistency; otherwise, the judgment matrix needs to be adjusted until the consistency requirements are met;

[0032] S2.3: Find the maximum eigenvalue λ max The corresponding eigenvectors are normalized to obtain the subjective weight matrix W of each evaluation index.

[0033] Furthermore, in S3, the construction process of the fuzzy cognitive graph model is as follows:

[0034] S3.1: Based on the evaluation index results obtained in step 1, define the concept nodes in the fuzzy cognitive map. Each concept node represents an evaluation index. The key influencing factors are abstracted and summarized into four basic concept nodes of the FCM model, which are expressed as a four-tuple G = (C, E, A, F);

[0035] Where C={C1,C2,...,C i} represents the set of concepts that constitute the vertices of a directed graph;

[0036] Among them, the key indicators that affect the load identification ability on the fuzzy cognitive map are represented as nodes. The values ​​on the nodes represent the actual performance of each indicator under the current power grid operation status. The weights marked on the lines between the nodes reflect the interaction strength between different indicators. The weight of all nodes in the fuzzy cognitive map is W. ij ,i,j=1,2,...,C,W ij ∈[-1,1],W ij represents the strength of the connection between the jth node and the ith node; when W ij >0, it means that when the state value of the jth node increases, the state value of the i-th node will also increase; when W ij <0, it means that when the state value of the jth node increases, the state value of the i-th node decreases; when W ij =0, it means there is no relationship between the jth node and the ith node;

[0037] A represents the concept node C i To activation degree A i A(t)=(A1(t),A2(t),...,A c (t)) represents the activation degree of all concept nodes at the current time t, which is the state of G at time t, where A i (t)∈[0,1];

[0038] F represents the activation function, which calculates the concept A i The state value at time t+1:

[0039]

[0040] The commonly used activation function F is represented by the sigmoid function:

[0041]

[0042] τ represents the steepness of the activation function. The larger the τ is, the closer the shape of the sigmoid function is to the step function. When τ is set to 5, the activation function maps the state value of the node to the interval [-1,1].

[0043] Step 3.2: Convert the influencing factors into numerical values ​​and adjust the numerical values ​​to [-1, 1] for normalization;

[0044] Step 3.3: Use fuzzy C-means clustering to cluster concept nodes C i Perform clustering and continuously update the activation degree u of the concept node according to the activation function F i (t+1) Until the activation change between two adjacent iterations is less than the preset threshold, the optimized weight matrix W′ is obtained.

[0045] Furthermore, in S4, the objective weight of each evaluation index is calculated according to the entropy weight method, and the information entropy size of each index in the overall evaluation is determined by analyzing the entropy value of the index evaluation data set, and then the weight W is obtained. ′ j , the specific calculation process is as follows:

[0046] S4.1: First, construct m load identification evaluation index objects and n evaluation index judgment matrix X = (x ij ) mxn , normalized matrix:

[0047]

[0048] Among them, x ij is each element of the matrix, min{x ij} is the minimum value of each indicator, max{x ij} is the maximum value of each indicator;

[0049] S4.2: Judgment matrix X = (x ij ) mxn The weight of each index is P = (p ij ) mxn , the calculation formula is as follows:

[0050]

[0051] When using the entropy weight method to determine the indicator weight, the entropy value of the normalized judgment matrix is ​​calculated to obtain the entropy value of each indicator.j The larger it is, the less information the indicator contains and the lower its weight should be.

[0052] S4.3: Through the entropy value E j Calculate the weight W of each indicator j and the final weight W′ j :

[0053]

[0054] In the formula, 6 represents the 6 evaluation indicators in the indicator layer.

[0055] Furthermore, in S5, the weighted average method is used to calculate the weights W′ and W′ j The specific method of fusion calculation is as follows:

[0056] W f =αW′+βW′ j

[0057] Where W f Each row of represents the weight of an evaluation indicator, 0<α+β≤1.

[0058] Furthermore, in S6, the fuzzy comprehensive evaluation analysis method comprises the following specific steps:

[0059] S6.1: According to the final weight vector W determined in S5 f , construct the weight matrix of fuzzy comprehensive evaluation;

[0060] S6.2: The calculated comprehensive fuzzy value matrix U corresponding to each layer of the key factors of load identification capability;

[0061] S6.3: Normalize the comprehensive fuzzy values ​​of all factors to eliminate the influence of different dimensions or numerical values ​​on the evaluation results;

[0062] S6.4: The obtained comprehensive weight matrix W f Multiply it with the fuzzy relationship matrix R to obtain the normalized comprehensive fuzzy evaluation matrix K′, calculate the final evaluation score M, and obtain the fuzzy comprehensive evaluation result.

[0063] Furthermore, according to the evaluation results obtained in S6, a comment set V is established to express the quality of the evaluation results. The comment set is set to V = [V1, V2, V3, V4, V5], corresponding to excellent, good, average, poor and very poor, respectively.

[0064] Compared with the prior art, the present invention has the following beneficial effects:

[0065] 1. The present invention constructs a comprehensive evaluation index system. Taking into account the fact that the existing comprehensive evaluation methods for load identification capability lack the correlation between different indicators, the present invention selects key indicators that are closer to judging the load identification capability according to the actual situation, and sets the evaluation index system into three levels from top to bottom: target layer, criterion layer and indicator layer; and determines 6 types of indicators for evaluating the performance of the load identification method, which comprehensively reflects the classification performance and application performance of the load identification method.

[0066] 2. The present invention integrates fuzzy cognitive maps into the comprehensive evaluation system of non-intrusive load identification capabilities. By constructing fuzzy cognitive maps, the mutual relationship and influence between various evaluation indicators can be clearly displayed, so that the evaluation system is closer to the actual situation. In addition, the fuzzy cognitive map can also dynamically adjust the evaluation indicators according to actual needs, further improving the flexibility and adaptability of the evaluation system. An improved hierarchical analysis method is used to determine the weight of each evaluation indicator, and the complex evaluation problem is decomposed into several levels. Through paired comparison and consistency test, the relative importance weight of each indicator is obtained to ensure the objectivity and accuracy of the evaluation results. The fuzzy comprehensive evaluation method is used to quantitatively analyze each evaluation indicator, which can handle the ambiguity and uncertainty in the evaluation process. By multiplying the comprehensive evaluation value obtained by quantifying the comment set with the normalized fuzzy comprehensive evaluation matrix, the final evaluation score that integrates the influence of all evaluation indicators can be obtained. The non-intrusive load identification capability can be ranked by this score, thereby assisting in selecting the best load identification scheme or improving the existing scheme. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 It is a flow chart of the comprehensive evaluation method of load identification capability based on fusion decision of the present invention;

[0068] Figure 2 is a distribution diagram of membership results of a load identification method according to an embodiment of the present invention;

[0069] Figure 3 It is a specific flow chart of the comprehensive evaluation method of load identification capability based on fusion decision in an embodiment of the present invention. DETAILED DESCRIPTION

[0070] The following embodiments of the present invention are described in further detail in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0071] Example:

[0072] like Figure 1-3 As shown, the present invention provides a comprehensive evaluation method for load identification capability based on fusion decision-making, and the implementation process is as follows:

[0073] S1: construct an evaluation index system for load identification, apply the load identification method to be evaluated to the load characteristic data set to be identified, and obtain the load identification results of each evaluation index;

[0074] The load identification evaluation index system includes classification performance index and application performance index, among which the classification performance index includes accuracy P C , recall rate R E , precision R C , the harmonic mean H1 of recall and precision; application performance indicators include training time T and number of samples S;

[0075] According to the key characteristics and requirements of load identification, the evaluation index system is set up into three levels from top to bottom: target layer, criterion layer and indicator layer; as shown in Table 1, the indicators in the second-level criterion layer correspond to the indicators in the third-level indicator layer one by one. The first-level indicators reflect the capability of the load identification algorithm, while the third-level indicators provide more detailed evaluation dimensions.

[0076] Table 1 Evaluation index system for load identification

[0077]

[0078] The indicators in the indicator layer are defined as:

[0079] Accuracy: P C =(TP+TN) / (P+N)

[0080] Recall: R E =TP / (TP+FN)

[0081] Accuracy: R C =TP / (TP+FP)

[0082] Harmonic mean:

[0083] Training time metrics:

[0084] Sample number indicators:

[0085] In the formula, TP represents the number of positive samples identified as positive, that is, true positive examples; FP represents the number of negative samples identified as positive, that is, false positive examples; FN represents the number of positive samples identified as negative, that is, false negative examples; TN represents the number of negative samples identified as negative, that is, true negative examples; P represents the number of positive samples, N represents the number of negative samples, T represents the time of algorithm training, in minutes, and S represents the number of samples participating in algorithm training. Table 2 shows the index evaluation results of the three types of load identification algorithms in this embodiment.

[0086] Table 2 Load identification results of various methods

[0087]

[0088] S2: The judgment matrix calculated using the importance ratio scale needs to be checked for consistency to obtain the subjective weight matrix W of each evaluation index;

[0089] Among them, the importance ratio scale calculation judgment matrix needs to be checked for consistency, and the checking process includes the following steps:

[0090] S2.1: Based on the evaluation index hierarchy model constructed in S1, a judgment matrix is ​​formed by comparing the importance ratio scales in pairs;

[0091] In the judgment matrix, the comparison results are usually expressed using a scale of 1 to 9, and their meanings are shown in Table 3:

[0092] Table 3 Comparison table of the importance of quantitative indicators

[0093]

[0094] Among them, a ij Indicates the importance of indicator i relative to indicator j, a ji Indicates the importance of index j relative to index i. The two are reciprocals of each other and are expressed as a ij =1 / a ji .

[0095] S2.2: Calculate the maximum eigenvalue λ of the judgment matrix max , consistency index CI and random consistency ratio CR. When CR is less than 0.1, the judgment matrix is ​​considered to have satisfactory consistency; otherwise, the judgment matrix needs to be adjusted until the consistency requirements are met;

[0096] S2.3: Find the maximum eigenvalue λ max The corresponding eigenvectors are normalized to obtain the subjective weight matrix W of each evaluation index.

[0097] S3: Construct a fuzzy cognitive graph model, use the fuzzy cognitive graph to connect the subjective weight matrix W through fuzzy relations, and use fuzzy C-means clustering to obtain the optimized weight matrix W ′ ;

[0098] The construction process of the fuzzy cognitive graph model is as follows:

[0099] S3.1: Based on the evaluation index results obtained in step 1, define the concept nodes in the fuzzy cognitive map. Each concept node represents an evaluation index. The key influencing factors are abstracted and summarized into four basic concept nodes of the FCM model, which are expressed as a four-tuple G = (C, E, A, F);

[0100] Where C={C1,C2,...,C i} represents the set of concepts that constitute the vertices of a directed graph;

[0101] Among them, the key indicators that affect the load identification ability on the fuzzy cognitive map are represented as nodes. The values ​​on the nodes represent the actual performance of each indicator under the current power grid operation status. The weights marked on the lines between the nodes reflect the interaction strength between different indicators. The weight of all nodes in the fuzzy cognitive map is W. ij ,i,j=1,2,...,C,W ij ∈[-1,1],W ij represents the strength of the connection between the jth node and the ith node; when W ij >0, it means that when the state value of the jth node increases, the state value of the i-th node will also increase; when W ij <0, it means that when the state value of the jth node increases, the state value of the i-th node decreases; when W ij =0, it means there is no relationship between the jth node and the ith node;

[0102] A represents the concept node C i To activation degree A i A(t)=(A1(t),A2(t),...,A c (t)) represents the activation degree of all concept nodes at the current time t, which is the state of G at time t, where A i (t)∈[0,1];

[0103] F represents the activation function, which calculates the concept A i The state value at time t+1:

[0104]

[0105] The commonly used activation function F is represented by the sigmoid function:

[0106]

[0107] τ represents the steepness of the activation function. The larger the τ is, the closer the shape of the sigmoid function is to the step function. When τ is set to 5, the activation function maps the state value of the node to the interval [-1,1].

[0108] Step 3.2: Convert the influencing factors into numerical values ​​and adjust the numerical values ​​to [-1, 1] for normalization;

[0109] Step 3.3: Use fuzzy C-means clustering to cluster concept nodes C i Perform clustering and continuously update the activation degree u of the concept node according to the activation function F i(t+1) Until the activation change between two adjacent iterations is less than the preset threshold, the optimized weight matrix W′ is obtained.

[0110] The fuzzy C-means clustering described therein allows data points to have partial membership rather than just belonging to one cluster, and its key parameters include the number of clusters C and the fuzzy factor Q of the membership, and in this embodiment, Q=2.

[0111] Specifically, find the membership degree u of each sample ij and the jth concept γ determined by the fuzzy C-means algorithm j , convert the time series in the two-dimensional space into the C-dimensional feature space:

[0112]

[0113] Among them, x i is the i-th time series data point. ij Indicates the membership of sample i to the jth concept, which is used to measure the closeness between the sample and each concept. i Represents the i-th time series data point, which is the specific value of the sample data. j represents the jth concept center determined by the fuzzy C-means algorithm, that is, the center point of the jth category. n represents the number of time series sample data points. c represents the dimension of the feature space, that is, the number of clusters (number of concepts) in the fuzzy clustering. Q represents the fuzzy index, which is used to control the fuzziness of the membership, usually Q>1. ‖x i -γ j ‖ represents sample x i To Concept Center | j The Euclidean distance (or other defined distance metric) of Represents the membership degree u ij The Q-th power of is used to weight the influence of samples when updating the concept center.

[0114] The available data is processed, and the cluster center is dynamically updated according to the current data based on the fuzzy dynamic C-means clustering. The fuzzy time series set is obtained, and the fuzzy time series is used for the learning of cognitive maps. During the learning process, the cognitive map can extract rules and patterns from the data and continuously adjust the network connection weights to improve the simulation accuracy of the system behavior.

[0115] The weight matrix is ​​continuously adjusted by minimizing the error, and the termination condition is as follows:

[0116]

[0117] Among them, t is the number of iterations; ε is a very small constant, 0.001, which is used to represent the error threshold.

[0118] In the present invention, parameter optimization mainly focuses on the adjustment of the weight matrix W'. The weight matrix is ​​adjusted continuously in an iterative manner until a preset error threshold or a maximum number of iterations is reached, thereby completing the parameter optimization process.

[0119] S4: Calculate the entropy value by entropy weight method to obtain the objective weight matrix W′ j ;

[0120] The objective weight of each evaluation index is calculated according to the entropy weight method. The information entropy of each index in the overall evaluation is determined by analyzing the entropy value of the index evaluation data set, and then the weight W′ is obtained. j , the specific calculation process is as follows:

[0121] S4.1: First, construct m load identification evaluation index objects and n evaluation index judgment matrix X = (x ij ) mxn , normalized matrix:

[0122]

[0123] Among them, x ij is each element of the matrix, min{x ij} is the minimum value of each indicator, max{x ij} is the maximum value of each indicator;

[0124] S4.2: Judgment matrix X = (x ij ) mxn The weight of each index is P = (p ij ) mxn , the calculation formula is as follows:

[0125]

[0126] When using the entropy weight method to determine the indicator weight, the entropy value of the normalized judgment matrix is ​​calculated to obtain the entropy value of each indicator. j The larger it is, the less information the indicator contains and the lower its weight should be.

[0127] S4.3: Through the entropy value E j Calculate the weight W of each indicator j and the final weight W′ j :

[0128]

[0129] In the formula, E j W represents the entropy value of the jth evaluation index, which is used to measure the information content of the index. The larger the entropy value, the lower the dispersion of the index, the smaller the amount of information contained, and the lower the weight. jRepresents the weight of the jth indicator. j To calculate, the weight is inversely proportional to the entropy value, that is, the smaller the entropy value, the greater the weight. ij It represents the normalized value of the i-th sample under the j-th evaluation index, which is mainly used to standardize the original data. Represents the sum of the entropy values ​​of all six evaluation indicators. W′ j It represents the final weight after correction, which is calculated by normalization to ensure that the sum of the weights is equal to 1. 6 represents the 6 evaluation indicators in the indicator layer.

[0130] S5: Take the weighted average method to calculate the weight W′ j and W′ j Perform fusion calculation to obtain the comprehensive weight matrix W f ;

[0131] The weights W′ and W′ are calculated by weighted average method. j The specific method of fusion calculation is as follows:

[0132] W f =αW′+βW′ j

[0133] Where W f Each row of represents the weight of an evaluation indicator, 0<α+β≤1.

[0134] Table 4 shows the fusion weight results of each indicator in this embodiment:

[0135] Table 4 Fusion weights of each indicator

[0136]

[0137] From the results in Table 4, we can see that the fusion weight W of each indicator f The value is between weight W′ and weight W′ j A balance between subjective and objective weighting is achieved, avoiding the problem of excessive subjectivity and weak objectivity of weights, which is conducive to the scientific and reasonable evaluation of various load identification methods.

[0138] The weights W of each evaluation index calculated are f As input parameters for fuzzy comprehensive evaluation in subsequent steps.

[0139] S6: The comprehensive weight matrix W f The membership value of the load identification method to be evaluated is obtained by multiplying it with the relationship matrix of the fuzzy comprehensive evaluation analysis method, and the final evaluation score M is calculated to obtain the fuzzy comprehensive evaluation result.

[0140] The fuzzy comprehensive evaluation analysis method has the following specific steps:

[0141] S6.1: According to the final weight vector W determined in S5 f , construct the weight matrix of fuzzy comprehensive evaluation;

[0142] S6.2: The calculated comprehensive fuzzy value matrix U corresponding to each layer of the key factors of load identification capability;

[0143] S6.3: Normalize the comprehensive fuzzy values ​​of all factors to eliminate the influence of different dimensions or numerical values ​​on the evaluation results;

[0144] S6.4: The obtained comprehensive weight matrix W f Multiply it with the fuzzy relationship matrix R to obtain the normalized comprehensive fuzzy evaluation matrix K ′ , calculate the final evaluation score M and obtain the fuzzy comprehensive evaluation result.

[0145] According to the evaluation results obtained in S6, a comment set V is established to express the quality of the evaluation results. The comment set is set to V = [V1, V2, V3, V4, V5], which correspond to excellent, good, general, poor and very poor respectively. For the convenience of calculation, the present invention quantifies the evaluation level and assigns excellent, good, general, poor and very poor values ​​100, 80, 60, 40 and 20 respectively. The indicator evaluation criteria table is shown in Table 5.

[0146] Table 5 Index evaluation criteria

[0147]

[0148]

[0149] Since the comprehensive evaluation results are fuzzy, defuzzification technology is needed to convert the fuzzy evaluation into specific values, and the maximum membership method is selected to convert the fuzzy evaluation results into clear comprehensive evaluation scores.

[0150] Determine the fuzzy relationship matrix R, which describes the relationship between the performance of the evaluation object on each evaluation index and each evaluation level. Each row of R represents an evaluation index, and each column represents an evaluation level. The element r in the matrix ij It indicates the membership degree of the evaluation object to the jth evaluation level on the i-th evaluation index. The calculated comprehensive fuzzy value matrix G corresponding to the load identification capability evaluation index is:

[0151]

[0152] The comprehensive fuzzy values ​​of all indicators are normalized to eliminate the influence of different dimensions or numerical values ​​on the evaluation results.

[0153]

[0154] Among them, G′ ij is the normalized value, G ij is the unnormalized comprehensive fuzzy value, max(G) and min(G) are the maximum and minimum values ​​in the comprehensive fuzzy value matrix, respectively.

[0155] After normalization, the comprehensive fuzzy value of each concept node will form a probability distribution, in which each value is between 0 and 1, and their sum is 1. This ensures that the relative contributions of different factors to the evaluation results are fair and comparable, regardless of their original numerical size or dimension.

[0156] The obtained comprehensive weight matrix W f The weighted average method is used to synthesize the fuzzy relationship matrix R to obtain the fuzzy comprehensive evaluation matrix K and normalize it to obtain K ’ .

[0157]

[0158] Substitute the corresponding values ​​according to the comment set to calculate the final evaluation score M:

[0159] M=K ′ ×V

[0160] According to the defuzzified value, the comprehensive evaluation score of the load identification capability of the power system is calculated, and the grade standards are pre-set as excellent V1, good V2, general V3, poor V4 and very poor V5. The score is compared with the grade standard to determine the grade of the load identification capability of the power system. Table 6 is the membership ranking of various methods in this embodiment.

[0161] Table 6 Membership ranking of various methods

[0162]

[0163]

[0164] Referring to the results in Table 6, the load identification method based on Bi-LSTM is the best, with a comprehensive evaluation level of V1 and a comprehensive evaluation score of 94.24. Its scores for each index are higher than those of the other two methods. The load identification method based on RNN is second, with a comprehensive evaluation level of V2 and a comprehensive evaluation score of 85.53. The load identification method based on CNN has the lowest score compared to the other two algorithms, and its scores for each index are lower than those of the first two methods. The evaluation results of this embodiment are consistent with actual engineering applications, verifying the practical effectiveness of the technical method of the present invention.

[0165] The embodiments of the present invention are provided for the purpose of illustration and description. Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations of the present invention. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. A comprehensive evaluation method for load identification capability based on fusion decision-making, characterized in that: The following steps are involved: S1: construct an evaluation index system for load identification, apply the load identification method to be evaluated to the load characteristic data set to be identified, and obtain the load identification results of each evaluation index; S2: The judgment matrix calculated using the importance ratio scale needs to be checked for consistency to obtain the subjective weight matrix W of each evaluation index; S3: Construct a fuzzy cognitive graph model, use the fuzzy cognitive graph to connect the subjective weight matrix W through fuzzy relations, and use fuzzy C-means clustering to obtain the optimized weight matrix W′; S4: Calculate the entropy value by entropy weight method to obtain the objective weight matrix W′ j ; S5: Take the weighted average method to calculate the weights W′ and W′ j Perform fusion calculation to obtain the comprehensive weight matrix W f ; S6: The comprehensive weight matrix W f The membership value of the load identification method to be evaluated is obtained by multiplying it with the relationship matrix of the fuzzy comprehensive evaluation analysis method, and the final evaluation score M is calculated to obtain the fuzzy comprehensive evaluation result.

2. The method for comprehensive evaluation of load identification capability based on fusion decision-making according to claim 1, characterized in that: In S1, the load identification evaluation index system includes classification performance index and application performance index, wherein the classification performance index includes accuracy P C , recall rate R E , precision R C , the harmonic mean H1 of recall and precision; application performance indicators include training time T and number of samples S; According to the key characteristics and requirements of load identification, the evaluation index system is set up into three levels from top to bottom: target layer, criterion layer and indicator layer.

3. The method for comprehensive evaluation of load identification capability based on fusion decision-making according to claim 2, characterized in that: In S1, the load identification methods to be evaluated include a load identification method based on a convolutional neural network, a load identification method based on a recurrent neural network, and a load identification method based on a bidirectional long short-term memory network optimization algorithm. The above methods are applied to the PLAID data set after filtering and screening to perform load identification on various types of electrical equipment.

4. The method for comprehensive evaluation of load identification capability based on fusion decision-making according to claim 3 is characterized in that: The indicators of the indicator layer are defined as: Accuracy: P C =(TP+TN) / (P+N) Recall: R E =TP / (TP+FN) Accuracy: R C =TP / (TP+FP) Harmonic mean: Training time metrics: Sample number indicators: In the formula, TP represents the number of positive samples identified as positive, that is, true positive examples; FP represents the number of negative samples identified as positive, that is, false positive examples; FN represents the number of positive samples identified as negative, that is, false negative examples; TN represents the number of negative samples identified as negative, that is, true negative examples; P represents the number of positive samples, N represents the number of negative samples, T represents the time of algorithm training, in minutes, and S represents the number of samples participating in the algorithm training.

5. The method for comprehensive evaluation of load identification capability based on fusion decision-making according to claim 2, characterized in that: In S2, the importance ratio scale calculation judgment matrix needs to be checked for consistency, and the checking process includes the following steps: S2.1: Based on the evaluation index hierarchy model constructed in S1, a judgment matrix is ​​formed by comparing the importance ratio scales in pairs; S2.2: Calculate the maximum eigenvalue λ of the judgment matrix max , consistency index CI and random consistency ratio CR. When CR is less than 0.1, the judgment matrix is ​​considered to have satisfactory consistency; otherwise, the judgment matrix needs to be adjusted until the consistency requirements are met; S2.3: Find the maximum eigenvalue λ max The corresponding eigenvectors are normalized to obtain the subjective weight matrix W of each evaluation index.

6. The method for comprehensive evaluation of load identification capability based on fusion decision-making according to claim 1, characterized in that: In S3, the construction process of the fuzzy cognitive graph model is as follows: S3.1: Based on the evaluation index results obtained in step 1, define the concept nodes in the fuzzy cognitive map. Each concept node represents an evaluation index. The key influencing factors are abstracted and summarized into four basic concept nodes of the FCM model, which are expressed as a four-tuple G = (C, E, A, F); Where C={C1,C2,...,C i } represents the set of concepts that constitute the vertices of a directed graph; Among them, the key indicators that affect the load identification ability on the fuzzy cognitive map are represented as nodes. The values ​​on the nodes represent the actual performance of each indicator under the current power grid operation status. The weights marked on the lines between the nodes reflect the interaction strength between different indicators. The weight of all nodes in the fuzzy cognitive map is W. ij ,i,j=1,2,...,C,W ij ∈[-1,1],W ij represents the strength of the connection between the jth node and the ith node; A represents the concept node C i To activation degree A i A(t)=(A1(t),A2(t),...,A c (t)) represents the activation degree of all concept nodes at the current time t, which is the state of G at time t, where A i (t)∈[0,1]; F represents the activation function, which calculates the concept A i The state value at time t+1: The commonly used activation function F is represented by the sigmoid function: τ represents the steepness of the activation function. The larger the τ is, the closer the shape of the sigmoid function is to the step function. When τ is set to 5, the activation function maps the state value of the node to the interval [-1,1]. Step 3.2: Convert the influencing factors into numerical values ​​and adjust the numerical values ​​to [-1, 1] for normalization; Step 3.3: Use fuzzy C-means clustering to cluster concept nodes C i Perform clustering and continuously update the activation degree u of the concept node according to the activation function F i (t+1) Until the activation change between two adjacent iterations is less than the preset threshold, the optimized weight matrix W′ is obtained.

7. The method for comprehensive evaluation of load identification capability based on fusion decision-making according to claim 4 is characterized in that: In S4, the objective weight of each evaluation index is calculated according to the entropy weight method. The information entropy of each index in the overall evaluation is determined by analyzing the entropy value of the index evaluation data set, and then the weight W′ is obtained. j , the specific calculation process is as follows: S4.1: First, construct m load identification evaluation index objects and n evaluation index judgment matrix X = (x ij ) mxn , normalized matrix: Among them, x ij is each element of the matrix, min{x ij } is the minimum value of each indicator, max{x ij } is the maximum value of each indicator; S4.2: Judgment matrix X = (x ij ) mxn The weight of each index is P = (p ij ) mxn , the calculation formula is as follows: When using the entropy weight method to determine the indicator weight, the entropy value of the normalized judgment matrix is ​​calculated to obtain the entropy value of each indicator. j The larger it is, the less information the indicator contains and the lower its weight should be. S4.3: Through the entropy value E j Calculate the weight W of each indicator j and the final weight W′ j : In the formula, 6 represents the 6 evaluation indicators in the indicator layer.

8. The method for comprehensive evaluation of load identification capability based on fusion decision-making according to claim 1, characterized in that: In S5, the weighted average method is used to calculate the weights W′ and W′ j The specific method of fusion calculation is as follows: W f =αW′+βW′ j Where W f Each row of represents the weight of an evaluation indicator, 0<α+β≤1.

9. The method for comprehensive evaluation of load identification capability based on fusion decision-making according to claim 1, characterized in that: In S6, the fuzzy comprehensive evaluation analysis method comprises the following specific steps: S6.1: According to the final weight vector W determined in S5 f , construct the weight matrix of fuzzy comprehensive evaluation; S6.2: The calculated comprehensive fuzzy value matrix U corresponding to each layer of the key factors of load identification capability; S6.3: Normalize the comprehensive fuzzy values ​​of all factors to eliminate the influence of different dimensions or numerical values ​​on the evaluation results; S6.4: The obtained comprehensive weight matrix W f Multiply it with the fuzzy relationship matrix R to obtain the normalized comprehensive fuzzy evaluation matrix K′, calculate the final evaluation score M, and obtain the fuzzy comprehensive evaluation result.

10. The method for comprehensive evaluation of load identification capability based on fusion decision-making according to claim 9, characterized in that: According to the evaluation results obtained in S6, a comment set V is established to express the quality of the evaluation results. The comment set is set to V = [V1, V2, V3, V4, V5], corresponding to excellent, good, average, poor and very poor, respectively.

Images