Man-machine hybrid driven new energy site selection index system construction method and system

Abstract

The invention relates to a man-machine hybrid drive new energy site selection index system construction method and system, and the method comprises the steps: generating a defuzzification collective decision matrix through a preliminary index system constructed by a new energy site selection expert and the evaluation information of a decision maker, and carrying out the intelligent optimization of the index system through a machine based on the matrix, and obtaining a final index system of new energy development site selection. The human intelligence and machine intelligence of new energy site selection experts and decision makers are combined, so that the inherent strong subjectivity problem in the new energy development site selection index system construction process is solved, and the scientific preciseness and the automation intelligent level of new energy development site selection index system construction are improved. In the index system optimization process, the index redundancy analysis model can be used for analyzing the preliminary index system, redundant indexes in the preliminary index system are removed, and key indexes having a stronger distinguishing effect on decision-making schemes are reserved, so that the index system is optimized.

Description

Technical Field

[0001] The present application relates to the field of intelligent site selection technology for the development of new energy sources such as wind power and solar photovoltaics, and in particular to a method and system for constructing a human-machine hybrid-driven new energy site selection index system. Background Art

[0002] New energy development site selection is the process of selecting the best location for efficient use of renewable energy by comprehensively considering factors such as resources and the environment. A reasonable indicator system is not only a prerequisite for new energy site selection, but also the key to ensuring the accuracy and reliability of site selection, which directly affects the effectiveness of new energy site selection decisions. In order to avoid waste of resources and ecological damage, it is particularly important to build a scientific new energy site selection indicator system.

[0003] At present, the existing research on site selection for renewable energy development mainly focuses on the construction of decision-making models. For example, in wind energy: construct a multi-index decision-making model for offshore wind power site selection based on the Bayesian best-worst method; establish a multi-index decision-making model for offshore wind power based on game theory; in solar energy: use probabilistic language information to establish a multi-stage decision support framework for the site selection of solar power plants; use geographic information systems and hierarchical analysis to determine the best site for solar photovoltaic power plants, etc.

[0004] However, the above-mentioned new energy development site selection method does not fully consider the reasonable construction of the index system, and regards the new energy development site selection index system as a secondary component of the decision-making model. It usually relies solely on the subjective judgment of decision makers to determine the site selection index system.

[0005] That is to say, in the existing new energy development site selection process, the indicator system construction method has technical problems such as strong subjectivity and low degree of automation and intelligence. Summary of the invention

[0006] Based on this, it is necessary to provide a method for constructing a new energy site selection index system driven by a human-machine hybrid and a system for constructing a new energy site selection index system driven by a human-machine hybrid to address the above technical problems.

[0007] In order to achieve the above objectives, the embodiments of the present invention adopt the following technical solutions:

[0008] On the one hand, a method for constructing a new energy site selection index system driven by human-machine hybrid is provided, including:

[0009] Obtain the preliminary indicator system and evaluation information for new energy development site selection; the preliminary indicator system is constructed and improved by new energy site selection experts, and the evaluation information is the information on the evaluation of each indicator of the preliminary indicator system by new energy site selection decision makers under various alternative plans;

[0010] According to the evaluation information, the decision matrix is ​​obtained by using trapezoidal fuzzy numbers, and the decision matrix is ​​integrated and defuzzified to obtain a defuzzified collective decision matrix.

[0011] According to the defuzzified collective decision-making matrix, machine intelligence is used to optimize the analysis of the preliminary indicator system, and the final indicator system is obtained based on the analysis results.

[0012] In one embodiment, the step of optimizing and analyzing the preliminary indicator system using machine intelligence according to the defuzzified collective decision matrix includes:

[0013] According to the defuzzified collective decision-making matrix, the preliminary indicator system is analyzed using the indicator redundancy analysis model to remove redundant indicators in the preliminary indicator system.

[0014] In one embodiment, the step of optimizing and analyzing the preliminary indicator system using machine intelligence according to the defuzzified collective decision matrix, after removing redundant indicators, further includes:

[0015] Based on the defuzzified collective decision matrix and redundant indicators, the machine learning algorithm is used to quantitatively measure the ability of each key indicator combination to distinguish between alternatives and obtain the evaluation results; each key indicator is the remaining indicator except the redundant indicator;

[0016] Based on the evaluation results, the final indicator system is obtained.

[0017] In one embodiment, the step of analyzing the preliminary indicator system using the indicator redundancy analysis model according to the defuzzified collective decision matrix and removing redundant indicators in the preliminary indicator system includes:

[0018] Calculate the correlation coefficient between any two secondary indicators under the same primary indicator;

[0019] According to each correlation coefficient, a correlation coefficient matrix is ​​constructed, and the inverse matrix of the correlation coefficient matrix is ​​solved;

[0020] According to the inverse matrix, calculate the partial correlation coefficient between any two secondary indicators under the same primary indicator;

[0021] According to each partial correlation coefficient, the ability of each secondary indicator to distinguish between alternatives is quantified to obtain quantitative results;

[0022] Based on the quantitative results, the redundant indicators in the preliminary indicator system are deleted.

[0023] In one embodiment, the step of obtaining the evaluation result by using a machine learning algorithm to quantitatively measure the ability of differentiating between alternatives for each key indicator combination according to the defuzzified collective decision matrix and redundant indicators includes:

[0024] According to the defuzzified collective decision matrix and each key indicator, a classification feature matrix is ​​obtained;

[0025] Construct a random forest model based on the classification feature matrix;

[0026] According to the random forest model, calculate the weight and ranking of each key indicator;

[0027] According to the weights and rankings, all key indicator combinations are obtained, and the ability to distinguish between alternatives for each indicator combination is quantitatively measured to obtain the evaluation results.

[0028] In one embodiment, the step of obtaining a classification feature matrix according to the defuzzified collective decision matrix and each key indicator includes:

[0029] According to the defuzzified collective decision matrix and key indicators, a detailed decision data matrix is ​​obtained;

[0030] The performance coefficients of the alternative plans are calculated according to the refined decision data matrix, and the alternative plans are classified according to the performance coefficients to obtain the classification types of the alternative plans;

[0031] According to the refined decision data matrix and each classification type, a classification feature matrix is ​​obtained.

[0032] On the other hand, a human-machine hybrid driven new energy site selection index system construction system is also provided, including:

[0033] The information acquisition module is used to obtain the preliminary indicator system and evaluation information for the site selection of new energy development; the preliminary indicator system is constructed and improved by the new energy site selection experts, and the evaluation information is the information on the evaluation of each indicator of the preliminary indicator system by the decision makers of the new energy site selection under each alternative plan;

[0034] An information processing module is used to obtain a decision matrix using trapezoidal fuzzy numbers according to the evaluation information, integrate and defuzzify the decision matrix, and obtain a defuzzified collective decision matrix;

[0035] The indicator system optimization module is used to optimize and analyze the preliminary indicator system based on the defuzzified collective decision matrix using machine intelligence, and obtain the final indicator system based on the analysis results.

[0036] In one embodiment, in the above-mentioned human-machine hybrid driven new energy site selection index system construction system, the index system optimization module includes:

[0037] The redundancy analysis submodule is used to analyze the preliminary indicator system based on the defuzzified collective decision-making matrix and use the indicator redundancy analysis model to remove redundant indicators in the preliminary indicator system.

[0038] In one embodiment, in the above-mentioned human-machine hybrid driven new energy site selection index system construction system, the index system optimization module also includes:

[0039] The evaluation submodule is used to obtain the evaluation results by using a machine learning algorithm to quantitatively measure the ability of differentiating between alternatives for each key indicator combination based on the defuzzified collective decision matrix and redundant indicators; each key indicator is the remaining indicator except the redundant indicator;

[0040] The optimization submodule is used to obtain the final indicator system based on the evaluation results.

[0041] One or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages:

[0042] The above-mentioned human-machine hybrid driven new energy site selection index system construction method and system generates a defuzzified collective decision matrix through the preliminary index system constructed by the new energy site selection experts and the evaluation information of the decision makers. Based on the matrix, the index system is optimized by machine intelligence to obtain the final index system for new energy development site selection. Combining the human intelligence of the new energy site selection expert decision makers with machine intelligence not only solves the inherent strong subjectivity problem in the process of constructing the new energy development site selection index system, but also improves the scientific rigor and automation and intelligence level of the construction of the new energy development site selection index system. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the conventional technology, the drawings required for use in the embodiments or the conventional technology descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0044] Figure 1 A schematic diagram of a process for constructing a human-machine hybrid driven new energy site selection index system in one embodiment;

[0045] Figure 2 The figure is a flow chart of the RF-RFE (Random Forest-Recursive Feature Elimination) algorithm in one embodiment. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0047] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used in the specification of this application are only for the purpose of describing specific embodiments and are not intended to limit this application.

[0048] It should be noted that the reference to "embodiment" in this document means that the specific features, structures or characteristics described in conjunction with the embodiment may be included in at least one embodiment of the present invention. The presentation of the phrase at various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments.

[0049] The following will describe the implementation of the present invention in detail with reference to the accompanying drawings in the embodiment diagram of the present invention.

[0050] In one embodiment, Figure 1 As shown, the embodiment of the present application provides a method for constructing a new energy site selection index system driven by human-machine hybrid, including the following processing steps S12-16:

[0051] S12, obtaining a preliminary indicator system and evaluation information for new energy development site selection; the preliminary indicator system is constructed and improved by new energy site selection experts, and the evaluation information is the information for new energy site selection decision makers to evaluate each indicator of the preliminary indicator system under each alternative plan.

[0052] It is understandable that, first of all, the new energy site selection experts refer to the existing new energy development site selection research literature, relevant reports and sustainable development requirements, and combine their own experience and systematic judgment to form a draft of the new energy development site selection index system from a sustainable perspective. It is also possible to design a questionnaire or interview framework based on the draft of the new energy development site selection index system, and further modify and improve the draft of the new energy development site selection index system through surveys and interviews with more industry professionals, so as to determine the preliminary new energy development site selection index system.

[0053] The decision makers of new energy site selection are also experts in the field of new energy site selection. They can be the same group of experts who build the preliminary indicator system, or they can be decision makers with more technical authority. k (k=1,2,…,p) can use language terms (very bad, very bad, bad, medium, good, very good, excellent) to describe each alternative (i.e., alternative development location, which can be recorded as A i , i=1,2,…,m) under each index C j (j=1,2…,n) gives evaluation information.

[0054] S14, according to the evaluation information, a decision matrix is ​​obtained by using trapezoidal fuzzy numbers, and the decision matrix is ​​integrated and defuzzified to obtain a defuzzified collective decision matrix.

[0055] It can be understood that the TrFN (trapezoidal fuzzy number) individual decision matrix can be obtained based on the trapezoidal fuzzy numbers corresponding to the language terms. Due to the diversity of new energy development site selection indicators and the incompleteness of decision information, the preference evaluation information provided by the decision model usually has a certain degree of fuzziness and uncertainty. Trapezoidal fuzzy numbers can effectively represent and quantify the fuzzy uncertainty in the decision maker's preference evaluation information and improve the information quality. Therefore, decision makers use TrFNs to express their preference evaluation information to ensure the scientificity and reliability of the construction of the new energy development site selection indicator system.

[0056] Among them, TrFNs can be defined as: is a TrFN, where a1 and a4 are trapezoidal fuzzy numbers The upper and lower bounds of satisfy the condition 0≤a1≤a2≤a3≤a4≤1. If a1=a2=a3=a4, then the trapezoidal fuzzy number Degenerates to a real number. The membership function of TrFN can be expressed as:

[0057]

[0058] The language terms and corresponding TrFNs are shown in Table 1:

[0059] Table 1:

[0060]

[0061] The TrFN individual decision matrix can be written as: but:

[0062]

[0063] in is a TrFN, which represents the evaluation value of the i-th alternative given by the k-th decision maker relative to the j-th indicator. k Indicates E k The individual decision matrix.

[0064] For all individual decision matrices X k Integrate to form a collective decision-making matrix that includes the collective wisdom of domain experts Specifically, the integration process can be:

[0065]

[0066] Among them, TrFNWA is the n-dimensional arithmetic weighted average operator of TrFNs, is a collection of TrFNs, ω i is a trapezoidal fuzzy number The weight of

[0067] Then defuzzification is performed, and the defuzzification process can be:

[0068]

[0069] in, For a TrFN.

[0070] After defuzzification, the defuzzified collective decision matrix X is obtained.

[0071] S16, based on the defuzzified collective decision matrix, use machine intelligence to optimize and analyze the preliminary indicator system, and obtain the final indicator system based on the analysis results.

[0072] It can be understood that the preliminary indicator system can be analyzed and optimized through machine intelligence to obtain a better indicator system.

[0073] Specifically, after obtaining the defuzzified collective decision matrix X, the principal component analysis (PCA) method can be used. PCA can reduce the dimension and convert the indicator data to a few principal components through linear transformation to reduce noise and redundant information. By extracting the principal components in the data, the indicators that contribute most to the final decision can be identified, thereby removing redundant indicators and optimizing the preliminary indicator system. It can also be optimized using an autoencoder based on deep learning. The autoencoder is an unsupervised learning model that discovers important hidden features by compressing and reconstructing high-dimensional data. Using the autoencoder to train the preliminary indicator system can identify important indicators and reduce redundancy, thereby improving the effectiveness of the indicator system.

[0074] The preliminary indicator system can also be analyzed using an indicator redundancy analysis model based on the defuzzified collective decision matrix X to remove redundant indicators in the preliminary indicator system. Specifically, a PC-ANOVA (partial correlation variance analysis) indicator redundancy analysis model can be constructed.

[0075] Based on the above methods, the random forest model can be further used to perform quantitative optimization analysis on the indicator combination. Specifically, for example, after removing redundant indicators, the machine learning algorithm can be used to quantitatively measure the ability to distinguish between alternative options for each key indicator combination (each key indicator is the remaining indicator except the redundant indicator) according to the defuzzified collective decision matrix and redundant indicators to obtain evaluation results; then, the final indicator system can be obtained based on the evaluation results.

[0076] In the above-mentioned human-machine hybrid driven new energy site selection index system construction method, the preliminary index system constructed by the new energy site selection experts and the evaluation information of the decision makers are used to generate a defuzzified collective decision matrix. Based on this matrix, the index system is optimized by machine intelligence to obtain the final index system for new energy development site selection. Combining the human intelligence of new energy site selection experts and decision makers with machine intelligence not only solves the inherent strong subjectivity problem in the process of constructing the new energy development site selection index system, but also improves the scientific rigor and automation and intelligence level of the construction of the new energy development site selection index system.

[0077] In one embodiment, the preliminary indicator system is analyzed using an indicator redundancy analysis model based on the defuzzified collective decision matrix to remove redundant indicators in the preliminary indicator system.

[0078] It can be understood that the defuzzified collective decision matrix can be input into the indicator redundancy analysis model, and the similarity and dependency between indicators can be calculated using correlation analysis and other methods. By setting the correlation threshold, highly redundant indicator pairs can be identified, and redundant indicators with low contribution can be removed, retaining the most representative key indicators, and obtaining an optimized indicator system.

[0079] Specifically, the correlation coefficient between any two secondary indicators under the same primary indicator can be calculated first.

[0080] Understandable, we can set r jl is the correlation coefficient between the jth and lth indicators, then r jl The calculation formula can be defined as:

[0081]

[0082] where x ij is a certain value in the matrix X, which represents the evaluation value of the i-th alternative relative to the j-th indicator. and are the average values ​​of the jth and lth indicators under m alternatives, respectively. m represents the total number of alternatives.

[0083] According to each correlation coefficient, a correlation coefficient matrix is ​​constructed, and the inverse matrix of the correlation coefficient matrix is ​​solved.

[0084] It can be understood that the correlation coefficient r can be used jl (j=1,2,…,n,l=1,2,…,n) construct the correlation coefficient matrix R, and then solve the inverse matrix R of R - ,but:

[0085]

[0086] According to the inverse matrix, the partial correlation coefficient between any two secondary indicators under the same primary indicator is calculated.

[0087] Understandable, PCC can be set jl is the partial correlation coefficient between the jth and lth indicators, then:

[0088]

[0089] PCC jl ∈[-1,1],|PCC jl The size of | has a positive impact on the correlation between the jth and lth indicators.

[0090] According to each partial correlation coefficient, the ability of each secondary indicator to distinguish between alternatives is quantified to obtain quantitative results.

[0091] It can be understood that, specifically, it can be for |PCC jl |≥0.6, quantify the ADA (ability to distinguish between alternatives) of each indicator. Inspired by the F-value method in variance analysis and combined with the actual situation of new energy development site selection, an improved F-value method is proposed to quantify the ADA of each indicator in the redundant indicator pair. Assuming that indicators j and l constitute a redundant indicator pair, then:

[0092]

[0093] Among them, F(j) and F(l) represent the ability of indicator j and indicator l to distinguish between alternatives respectively. The higher the F(j) value, the stronger the ability of indicator j to distinguish between alternatives. ij It is a certain value in the matrix X, which represents the evaluation value of the i-th alternative relative to the j-th indicator. It is the average value of all indicator values ​​under the jth and lth indicators. and are the average values ​​of the jth and lth indicators under m alternatives, respectively. m represents the total number of alternatives.

[0094] Based on the quantitative results, the redundant indicators in the preliminary indicator system are deleted.

[0095] Specifically, the indicator with a lower ADA value can be deleted from the redundant indicator pair. The F(j)j=1,2,...,n value can be used as the basis for deletion, and the indicator with a smaller F value can be deleted from the redundant indicator pair.

[0096] In the above-mentioned human-machine hybrid driven new energy site selection index system construction method, by quantitatively analyzing the alternative scheme differentiation ability of redundant indicators in each secondary index, it is possible to accurately identify and delete redundant indicators, retain key indicators that have a stronger differentiation effect on decision-making schemes, and thus optimize the index system. It can improve the accuracy of evaluation results, reduce redundancy between indicators, and improve model calculation efficiency and scientific decision-making of new energy site selection.

[0097] In one embodiment, after removing redundant indicators, a RF-RFE (Random Forest-Recursive Feature Elimination) machine learning algorithm is constructed (the RF-RFE algorithm flow chart is as follows Figure 2 The method is shown in Figure 2 to quantify the ability of the indicator system to distinguish between alternatives. Specifically, it includes: generating a classification feature matrix based on the defuzzified collective decision matrix, and using the random forest model to calculate the weight and ranking of each key indicator. Subsequently, all key indicator combinations are quantitatively evaluated, and finally the evaluation results of all indicator combinations are obtained. Specifically:

[0098] According to the defuzzified collective decision matrix and each key indicator, the classification feature matrix is ​​obtained.

[0099] The method for obtaining the classification feature matrix, in one embodiment, is specifically:

[0100] The refined decision data matrix can be obtained based on the defuzzified collective decision matrix and each key indicator. That is, according to the defuzzified matrix X and the deletion result of redundant indicators, the specified redundant indicator data columns are removed from the matrix X to obtain the refined decision data matrix X. d .

[0101] Then, the performance coefficient of each alternative plan is calculated according to the refined decision data matrix, and each alternative plan is classified according to each performance coefficient to obtain the classification type of each alternative plan.

[0102] That is: using the refined decision data matrix X d , calculate the performance coefficient of the alternative solution:

[0103]

[0104] AOP i ) represents the performance coefficient of the jth alternative. ij is the matrix X dThe elements in represent the evaluation value of the i-th alternative relative to the j-th indicator. Subsequently, the classification type of the alternative is determined based on the AOP coefficient. Specifically, the decision maker defines the number of classification types of the alternatives (for example, two or three types) according to their unique needs, and then determines the category of each alternative by setting an appropriate threshold. Considering that the subjectively set threshold is difficult to accurately reflect the natural distribution of the data, important information may be ignored or unnecessary noise may be introduced. At the same time, it may also lead to an unbalanced number of alternatives for each classification type, which in turn has an adverse effect on the performance of the subsequent indicator combination optimization algorithm. Therefore, an equal-frequency binning method considering the distribution of AOP data is also constructed to objectively and automatically determine the classification threshold of the alternative, avoiding the shortcomings of the subjective threshold and effectively solving the problem of unbalanced number of alternatives for each classification type.

[0105] If the decision maker defines the number of classification types of alternative options as Y categories, then Y-1 thresholds need to be determined. The threshold calculation method is as follows:

[0106]

[0107] Among them, θ e represents the e-th classification threshold, e=1,2,…,Y-1 and θ1<θ2<…<θ Y-1 .

[0108] AOPs={AOP(A1),AOP(A2),…,AOP(A m )}

[0109] Sort(AOPs) = {AOP(A (1) )AOP(A (2) )…,AOP(A (m) )}

[0110] Sort(AOPs) represents an ordered set of AOPs sorted from small to large. There exists AOP(A(1)) <AOP(A (2) )<… <AOP(A (m) ). represents rounding down. f(&,#) means extracting the value at position # from the ordered set &. Y is the total number of classification types of alternatives. The specific meaning of each classification type can be determined according to actual conditions. Without loss of generality, AOP (A i ) is larger, the alternative A i The better the classification, the better. For example, the alternatives can be divided into two types: ideal development site and suitable development site. That is, Y = 2, AOP (A i )<θ1 indicates that the ith alternative is a suitable development site, θ1≤AOP(A i) indicates that the ith alternative is the ideal development site.

[0111] According to the refined decision data matrix and the classification types of alternative plans, a classification feature matrix is ​​obtained.

[0112] That is: classification feature matrix X Id as follows:

[0113]

[0114] Among them, X Id is the classification feature matrix, which is also the input data matrix of the indicator combination optimization model. represents the classification type of the i-th alternative, e = 1, 2, ..., Y. In this embodiment, Y = 2, then and They correspond to the ideal development site and the suitable development site respectively.

[0115] Construct a random forest model based on the classification feature matrix.

[0116] Understandably, using X Id Construct a random forest. First, the data matrix X Id Each row in is considered as a separate sample, that is, there are a total of m samples in the original sample set. Subsequently, m samples are randomly extracted from the original sample set as the training data set for a single decision tree. Next, indicators are randomly selected from the indicator set to form an indicator subset (usually including Using the training data set and the indicator subset, the CART algorithm is used to divide the data set, create decision tree subnodes and build a decision tree. Finally, the above operations are repeated to obtain different decision trees, thus forming a random forest.

[0117] According to the random forest model, the weight and importance ranking of each key indicator are calculated.

[0118] It can be understood that the weights of each indicator are calculated based on the random forest model. The main calculation idea is to quantify the importance of each indicator by measuring its contribution to the random forest classification. Gini impurity is a common indicator for measuring the classification contribution in random forests. Therefore, the classification contribution of an indicator can be quantified by the change in the reduction of the Gini impurity of the indicator. The details are as follows:

[0119] ① Calculate the Gini impurity of the indicator at each node in the decision tree. Specifically, for any node q in the decision tree, the calculation formula for Gini impurity can be:

[0120]

[0121] Where Gini(q,Cj ) is based on the indicator C j The Gini impurity value of node q. Y is the total number of sample categories. qy Represents the proportion of samples of the yth category in node q.

[0122] For node q, if according to the indicator C j To split, split the node q by the index C j The reduction of Gini impurities caused is calculated as follows:

[0123]

[0124] Where ΔGini(q,C j ) indicates that the split node q is composed of the index C j The reduction in Gini impurity caused by |Q| represents the total number of samples in node q. L |and|Q R | is the left child node q formed by the splitting of node q L and right child node q R The total number of samples. Gini(q L ,C j ) and Gini(q R ,C j ) is the node q L and q R Gini impurity.

[0125] ② Calculate the importance of the indicator on a single decision tree. Indicator C j The importance of any decision tree t can be calculated by the indicator C j It is calculated by summing the reduction in Gini impurity of all nodes in decision tree t:

[0126]

[0127] in is the indicator C on the decision tree t j The total Gini impurity reduction, that is, the index C j Importance in decision tree t. is the decision tree t with index C j The set of all associated split nodes.

[0128] ③Calculate the importance of the indicator. For a random forest consisting of T decision trees, any indicator C can be measured as follows: j Importance in the entire random forest:

[0129]

[0130] The importance of indicators can also be normalized:

[0131]

[0132] in, Indicator C j The normalized weight of; if there are n indicators, then:

[0133]

[0134] Sort by the normalized weight of each key indicator.

[0135] According to the weights and rankings, all key indicator combinations are obtained, and the ability to distinguish between alternatives for each indicator combination is quantitatively measured to obtain the evaluation results.

[0136] It can be understood that the RF (Random Forest) algorithm has significant advantages in dealing with multi-indicator combinations and complex data relationships. It can make full use of the information of each indicator in the indicator combination and capture the complex interactions between them. Accuracy is an important manifestation of the performance of the RF model. By comparing the accuracy of the RF model under different indicator combinations, we can intuitively see which indicator combinations can better distinguish alternatives, thereby revealing the ADA of the current indicator combination.

[0137] The cross-validation method is introduced to train the RF model and evaluate its accuracy in small sample scenarios. This method divides the data set into U subsets of equal size, uses each subset as a test set in turn, and uses the remaining subsets as training sets to train the model. This ensures full use of the data and reduces possible errors and contingencies in the evaluation process, thereby more accurately reflecting the actual ADA of the indicator combination. Specifically:

[0138]

[0139] Among them, U represents the number of sample subsets divided by the cross-validation algorithm. Y is the total number of alternative classifications. TP y Indicates the total number of alternatives of the yth classification type correctly predicted by the model, that is, the alternative actually belongs to the yth type, and the RF model also accurately predicts that it belongs to the yth type. FP y is the total number of alternatives that are not actually of type y but are incorrectly predicted to be of type y by the RF model. y It indicates the total number of alternatives that actually belong to type y but are incorrectly predicted as other types by the RF model. The higher the ADA value, the stronger the indicator combination's ability to distinguish between alternatives.

[0140] Then, the indicator combinations are sorted according to their ADA values, and the indicator combinations with high ADA are selected based on the unique decision-making needs of decision makers to construct the optimal new energy development site selection indicator system, which is the final indicator system.

[0141] The above-mentioned human-machine hybrid-driven new energy site selection index system construction method, the application of the random forest model can automatically calculate the weight and ranking of each key index, comprehensively consider the synergistic influence between the indicators, and overcome the limitation of the traditional method that only relies on the subjective judgment of experts. By optimizing the indicator combination, the differences between different alternative development locations can be more accurately identified, and the ability to distinguish between alternatives (ADA) is significantly improved, ensuring the accuracy and reliability of new energy development site selection decisions. At the same time, by quantitatively measuring the ability to distinguish between alternatives of each indicator combination, the relationship between the indicator system and the decision-making results is clarified, providing a direction for quantitative optimization of the indicator system, breaking through the bottleneck of traditional qualitative analysis, effectively eliminating the difficulties of quantitative analysis, improving the scientificity and systematicness of the optimization process, and enhancing the practicality and flexibility of the indicator system in practical applications.

[0142] In some embodiments, in order to more intuitively and comprehensively illustrate the above-mentioned method for constructing a human-machine hybrid-driven new energy site selection index system, the following is an application example of the method for constructing a human-machine hybrid-driven new energy site selection index system. It should be noted that the implementation cases given in this specification are only illustrative and are not the only limitation for the specific implementation cases of the present invention. Those skilled in the art can adopt the above-mentioned method for constructing a human-machine hybrid-driven new energy site selection index system under the guidance of the implementation cases provided by the present invention to improve the scientific rigor and automation and intelligence level of new energy site selection.

[0143] First, a decision-making committee is composed of a large number of experts in different fields such as wind energy, social economy, environment, geology and hydrology, and engineering construction management. For the convenience of management, these experts are divided into three decision-making groups (which can be regarded as three decision makers). Each decision-making group has experts in all the above fields, denoted as E1, E2, and E3. After discussion and consultation, the weights of the three decision-making groups are 0.35, 0.25, and 0.4 respectively.

[0144] In addition, the decision-making committee preliminarily identified several alternative development locations for offshore wind farms based on the collected meteorological data, ocean data and other information, and eliminated some locations based on four veto factors: engineering infrastructure, port channel areas, natural retrograde areas, and military activity areas. Finally, 32 locations were identified as alternative options, denoted as A1, A2, ..., A32.

[0145] Specific implementation process: Step 1: Relying on decision makers and experts, referring to existing research literature on new energy development site selection, relevant reports and sustainable development requirements, combined with own experience and systematic judgment, determine the preliminary index system for offshore wind power site selection from a sustainable perspective. The preliminary offshore wind power site selection index system is shown in Table 2:

[0146] Table 2

[0147]

[0148]

[0149] Step 2: Obtain preliminary indicator evaluation information. Decision maker E k (k = 1, 2, 3) Use language terms to provide each secondary indicator in Option A i Then, the trapezoidal fuzzy number individual decision matrix is ​​obtained according to Table 2 Due to space limitations, only part of the language individual decision matrix of decision maker E1 is given as an example. The language individual decision matrix of decision maker E1 is shown in Table 3:

[0150] Table 3

[0151]

[0152] Step 3: For all decision matrices X of decision makers k (k=1,2,3) to obtain the collective decision matrix containing the collective wisdom of decision makers Then, the defuzzified collective decision matrix X is obtained. Due to space limitations, only part of the information of the matrix X is shown. The defuzzified collective decision matrix X is shown in Table 4:

[0153] Table 4

[0154]

[0155] Step 4: Construct a PC-ANOVA (partial correlation variance analysis) indicator redundancy analysis model to remove redundant indicators in the preliminary indicator system for offshore wind power site selection. The results of the partial correlation coefficients between all secondary indicators under indicator C1 are shown in Table 5:

[0156]

[0157] From Table 5, we can see that C 11 and C 12 , C 15 and C 19 , C 16 and C 18, the absolute value of the partial correlation coefficient between them is greater than 0.6, and their F value is further calculated as: F(C 11 )=23.356,F(C 12 )=7.825,F(C 15 )=16.266,F(C 16 )=18.642,F(C 18 )=12.408,F(C 19 )=15.199. For redundant indicator pairs with high partial correlation coefficients, delete the indicator with smaller F value. In other words, delete indicator C 12 , C 18 , C 19 Similarly, a redundancy analysis was conducted on the secondary indicators under indicators C2, C3, C4 and C5, and 12 indicators were deleted from the preliminary indicator system for offshore wind power site selection. 12 , C 18 , C 19 , C 26 , C 27 , C 33 , C 36 , C 44 , C 45 , C 46 , C 53 , C 55}.

[0158] Step 5: Classify the alternatives and determine the classification feature matrix X Id Due to space limitations, only some results are given. The classification feature matrix X Id As shown in Table 6:

[0159]

[0160]

[0161] Step 6: Use the categorical feature matrix X Id Construct a random forest model, calculate the weight of each indicator, and sort them. The weight and sort of the indicators are shown in Table 7:

[0162] Table 7:

[0163]

[0164]

[0165] Step 7: Evaluate the alternative discrimination ability (ADA) of all offshore wind power indicator combinations. The alternative discrimination ability (ADA) of different offshore wind power site selection indicator combinations is shown in Table 8:

[0166] Table 8:

[0167]

[0168]

[0169] Step 8: Determine the optimal offshore wind power site selection index system. According to the ADA coefficient, the index combination {C 14 , C 16 , C 17 , C 21 , C 22 , C 23 , C 31 , C 34 , C 37 , C 41 , C 42 , C 43 , C 51 , C 52 The ADA value of} is the highest (0.876), covering five primary indicators of resources, technology, economy, society and environment, and meeting the current decision-making needs of decision makers. Therefore, it is determined to be the optimal offshore wind power site selection indicator system.

[0170] Comparative analysis: Scenario 1 relies only on decision makers and experts to construct a new energy site selection index system, while Scenario 2 applies the human-machine hybrid driven new energy site selection index system construction method proposed in this embodiment. Both cases use ADA to evaluate the quality of the constructed index system. In Scenario 1, the ADA of the preliminary new energy development site selection index system is evaluated by executing steps 1-3 and steps 5-7 of this embodiment; Scenario 2 executes all the steps of this embodiment. The comparison results of Scenario 1 and Scenario 2 are shown in Table 9:

[0171] Table 9:

[0172]

[0173] According to Table 9, compared with the ADA value of 0.782 in Scenario 1, the ADA value of the method in this embodiment reaches 0.876, an increase of 10.731%. The higher the ADA value, the more the index system can distinguish the alternatives at a finer granularity, thereby improving the accuracy and reliability of the offshore wind power site selection decision results. In addition, the number of indicators in the index system in this embodiment is 14, which is significantly less than the 35 indicators in Scenario 1 that only rely on the subjective judgment of decision makers and experts, and the reduction in indicators is 60%. The index system constructed by the human-machine hybrid drive method not only has a small number of indicators, but also has a high ability to distinguish alternatives. It effectively reduces the complexity of offshore wind power site selection, and significantly improves the objectivity, automation and intelligence of decision-making, ensuring that the results are more accurate and reliable.

[0174] Sensitivity experiment: This section provides a detailed sensitivity analysis to verify the robustness of the proposed model. Since the defuzzified collective decision matrix X is the intermediate data of the proposed model, its change is equivalent to the change of the entire input data, so the decision matrix X is selected as the object of sensitivity analysis. Specifically, the decision matrix X is perturbed within the range of ±30% to observe the changes in the output results. The sensitivity analysis results are shown in Table 10:

[0175] Table 10:

[0176]

[0177] According to Table 10, in the ±30% disturbance of matrix X, the ADA value and the optimal index system are 0.876 and {C 14 ,C 16 ,C 17 ,C 21 ,C 22 ,C 23 ,C 31 ,C 34 ,C 37 ,C 41 ,C 42 ,C 43 ,C 51 ,C 52}. Only twice did the ADA value change (i.e., perturbations of -15% and -20%), and only once did the indicator system change (i.e., perturbation of -20%). This stability shows that the method can maintain its performance in the presence of ±30% noise or uncertain data, has strong robustness, and can provide reliable and reasonable results.

[0178] It should be understood that although Figure 1 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 1 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0179] In one embodiment, a human-machine hybrid driven new energy site selection index system construction system is also provided, including:

[0180] The information acquisition module is used to obtain the preliminary indicator system and evaluation information for the site selection of new energy development; the preliminary indicator system is constructed and improved by the new energy site selection experts, and the evaluation information is the information on the evaluation of each indicator of the preliminary indicator system by the decision makers of the new energy site selection under each alternative plan;

[0181] An information processing module is used to obtain a decision matrix using trapezoidal fuzzy numbers according to the evaluation information, integrate and defuzzify the decision matrix, and obtain a defuzzified collective decision matrix;

[0182] The indicator system optimization module is used to optimize and analyze the preliminary indicator system based on the defuzzified collective decision matrix using machine intelligence, and obtain the final indicator system based on the analysis results.

[0183] The above-mentioned human-machine hybrid-driven new energy site selection index system construction system generates a defuzzified collective decision matrix through the preliminary index system constructed by new energy site selection experts and the evaluation information of decision makers. Based on this matrix, the index system is optimized by machine intelligence to obtain the final index system for new energy development site selection. Combining the human intelligence of new energy site selection experts and decision makers with machine intelligence not only solves the inherent strong subjectivity problem in the process of constructing the new energy development site selection index system, but also improves the scientific rigor and automation and intelligence level of the construction of the new energy development site selection index system.

[0184] In one embodiment, the above-mentioned human-machine hybrid driven new energy site selection index system construction system, the index system optimization module includes: a redundancy analysis submodule, which is used to analyze the preliminary index system according to the defuzzified collective decision matrix and use the index redundancy analysis model to remove redundant indicators in the preliminary index system.

[0185] In one embodiment, the human-machine hybrid driven new energy site selection index system construction system, the index system optimization module further includes:

[0186] The evaluation submodule is used to obtain the evaluation results by using a machine learning algorithm to quantitatively measure the ability of differentiating between alternatives for each key indicator combination based on the defuzzified collective decision matrix and redundant indicators; each key indicator is the remaining indicator except the redundant indicator;

[0187] The optimization submodule is used to obtain the final indicator system based on the evaluation results.

[0188] In one embodiment, the above-mentioned human-machine hybrid driven new energy site selection index system construction system, the redundancy analysis submodule includes:

[0189] The first calculation submodule is used to calculate the correlation coefficient between any two secondary indicators under the same primary indicator;

[0190] A solution submodule is constructed to construct a correlation coefficient matrix according to each correlation coefficient and solve the inverse matrix of the correlation coefficient matrix;

[0191] The second calculation submodule is used to calculate the partial correlation coefficient between any two secondary indicators under the same primary indicator according to the inverse matrix;

[0192] The quantification submodule is used to quantify the ability of each secondary indicator to distinguish between alternatives according to each partial correlation coefficient and obtain a quantitative result;

[0193] The deletion submodule is used to delete redundant indicators in the preliminary indicator system according to the quantification results.

[0194] In one embodiment, the human-machine hybrid driven new energy site selection index system construction system, the evaluation submodule includes:

[0195] The classification matrix submodule is used to obtain the classification feature matrix according to the defuzzified collective decision matrix and each key indicator;

[0196] The construction submodule is used to construct a random forest model based on the classification feature matrix;

[0197] The third calculation submodule is used to calculate the weight and ranking of each key indicator according to the random forest model;

[0198] The execution submodule is used to obtain all key indicator combinations according to weights and rankings, quantitatively measure the ability of each indicator combination to distinguish between alternative solutions, and obtain evaluation results.

[0199] In one embodiment, the above-mentioned human-machine hybrid driven new energy site selection index system construction system, the classification matrix submodule includes:

[0200] A refinement matrix submodule is used to obtain a refinement decision data matrix based on the defuzzified collective decision matrix and each key indicator;

[0201] The classification submodule is used to calculate the performance coefficient of each alternative plan according to the refined decision data matrix, classify each alternative plan according to each performance coefficient, and obtain the classification type of each alternative plan;

[0202] The synthesis submodule is used to obtain the classification feature matrix according to the refined decision data matrix and each classification type.

[0203] For the specific limitations of the human-machine hybrid driven new energy site selection index system construction system, please refer to the limitations of the human-machine hybrid driven new energy site selection index system construction method above, which will not be repeated here. Each module in the above-mentioned human-machine hybrid driven new energy site selection index system construction system can be implemented in whole or in part through software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0204] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0205] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for constructing a new energy site selection index system driven by human-machine hybrid, characterized in that: Includes steps: Obtaining a preliminary indicator system and evaluation information for new energy development site selection; the preliminary indicator system is constructed and improved by new energy site selection experts, and the evaluation information is information on the evaluation of each indicator of the preliminary indicator system by new energy site selection decision makers under various alternative plans; According to the evaluation information, a decision matrix is ​​obtained by using trapezoidal fuzzy numbers, and the decision matrix is ​​integrated and defuzzified to obtain a defuzzified collective decision matrix; According to the defuzzified collective decision matrix, the preliminary indicator system is optimized and analyzed using machine intelligence, and the final indicator system is obtained according to the analysis results.

2. The method for constructing a new energy site selection index system for human-machine hybrid drive according to claim 1, characterized in that: The step of optimizing and analyzing the preliminary indicator system using machine intelligence according to the defuzzified collective decision matrix includes: According to the defuzzified collective decision matrix, the preliminary indicator system is analyzed using an indicator redundancy analysis model to remove redundant indicators in the preliminary indicator system.

3. The method for constructing a new energy site selection index system for human-machine hybrid drive according to claim 2, characterized in that: In the step of optimizing and analyzing the preliminary indicator system by using machine intelligence according to the defuzzified collective decision matrix, after removing the redundant indicators, the step further includes: According to the defuzzified collective decision matrix and the redundant indicators, a machine learning algorithm is used to quantitatively measure the ability of each key indicator combination to distinguish between alternative solutions to obtain an evaluation result; each key indicator is the remaining indicator except the redundant indicator; According to the evaluation results, the final indicator system is obtained.

4. The method for constructing a new energy site selection index system for human-machine hybrid drive according to claim 2 or 3, characterized in that: The step of analyzing the preliminary indicator system according to the defuzzified collective decision matrix using an indicator redundancy analysis model to remove redundant indicators in the preliminary indicator system includes: Calculate the correlation coefficient between any two secondary indicators under the same primary indicator; According to the correlation coefficients, a correlation coefficient matrix is ​​constructed, and an inverse matrix of the correlation coefficient matrix is ​​solved; According to the inverse matrix, the partial correlation coefficient between any two secondary indicators under the same primary indicator is calculated; According to the partial correlation coefficients, the ability of each secondary indicator to distinguish between alternatives is quantified to obtain a quantitative result; According to the quantification results, redundant indicators in the preliminary indicator system are deleted.

5. The method for constructing a new energy site selection index system for human-machine hybrid drive according to claim 4 is characterized in that: The step of obtaining the evaluation result by using a machine learning algorithm to quantitatively measure the ability of distinguishing between alternatives of each key indicator combination according to the defuzzified collective decision matrix and the redundant indicator comprises: Obtaining a classification feature matrix according to the defuzzified collective decision matrix and each of the key indicators; Constructing a random forest model according to the classification feature matrix; According to the random forest model, the weight and ranking of each of the key indicators are calculated; According to the weights and rankings, all key indicator combinations are obtained, and the ability to distinguish between alternative solutions of each indicator combination is quantitatively measured to obtain the evaluation result.

6. The method for constructing a new energy site selection index system for human-machine hybrid drive according to claim 5, characterized in that: The step of obtaining a classification feature matrix according to the defuzzified collective decision matrix and each of the key indicators comprises: Obtaining a refined decision data matrix according to the defuzzified collective decision matrix and each of the key indicators; Calculating the performance coefficient of each of the alternative solutions according to the refined decision data matrix, classifying each of the alternative solutions according to the performance coefficients, and obtaining the classification type of each of the alternative solutions; The classification feature matrix is ​​obtained according to the refined decision data matrix and each of the classification types.

7. A human-machine hybrid driven new energy site selection index system construction system, characterized in that: include: Information acquisition module, used to obtain the preliminary indicator system and evaluation information for new energy development site selection; The preliminary indicator system is constructed and improved by new energy site selection experts, and the evaluation information is the information on the evaluation of each indicator of the preliminary indicator system by the new energy site selection decision makers under each alternative plan; An information processing module, used to obtain a decision matrix using trapezoidal fuzzy numbers according to the evaluation information, integrate and defuzzify the decision matrix, and obtain a defuzzified collective decision matrix; The indicator system optimization module is used to optimize and analyze the preliminary indicator system based on the defuzzified collective decision matrix using machine intelligence, and obtain the final indicator system based on the analysis results.

8. The human-machine hybrid driven new energy site selection index system construction system according to claim 7 is characterized in that: The indicator system optimization module includes: The redundancy analysis submodule is used to optimize and analyze the preliminary indicator system according to the defuzzified collective decision matrix using an indicator redundancy analysis model to remove redundant indicators in the preliminary indicator system.

9. The human-machine hybrid driven new energy site selection index system construction system according to claim 8 is characterized in that: The indicator system optimization module also includes: An evaluation submodule, for obtaining an evaluation result by using a machine learning algorithm to quantitatively measure the ability of differentiating between alternatives of each key indicator combination according to the defuzzified collective decision matrix and the redundant indicators; each key indicator is an indicator remaining except the redundant indicators; The optimization submodule is used to obtain the final indicator system according to the evaluation results.

Images