Multi-objective decision analysis method based on fuzzy logic

Abstract

The invention relates to the technical field of energy management, in particular to a fuzzy logic-based multi-objective decision analysis method, which comprises the following steps of: firstly, dividing an evaluation criterion into quantitative and qualitative index sets; constructing an interval type-2 fuzzy set capable of describing data uncertainty according to quantitative indexes such as a full-life investment payback period; for qualitative indexes such as technology maturity, a spherical fuzzy set is introduced, and an intelligent operator for perceiving hesitance is used for aggregation; calculating a group utility value, an individual regret value and a comprehensive sorting index of each energy configuration scheme by applying a hybrid fuzzy VIKOR algorithm, outputting an optimal compromise sorting result, and identifying a reference optimal strategy and a risk avoidance strategy; according to the method, the explainable enhanced fuzzy rule is extracted, the robustness of the decision method is improved by analyzing different types of fuzzy logic, and the transparency, the credibility and the compliance of the energy configuration decision process are remarkably enhanced by the finally generated explainable decision basis, so that a feasible configuration decision reasoning model is provided for energy planning.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of energy management, in particular to a multi-objective decision analysis method based on fuzzy logic. BACKGROUND

[0002] In the field of energy system planning and management, especially in strategic decision-making activities with multi-dimension, long cycle and high uncertainty characteristics such as regional integrated energy system construction and major energy infrastructure project selection, existing decision support analysis tools have significant limitations in dealing with complex information; When quantifying the uncertainty of the energy system operation environment, the existing reasoning model usually adopts a single-dimensional modeling framework. For example, some methods rely on historical operation data to make probability prediction of system risk, but it is difficult to effectively integrate and quantify the forward-looking judgment and industry experience from energy field experts or planning decision-makers. On the contrary, other qualitative analysis methods that rely on expert scoring are also difficult to fully digest and utilize the vast amount of power grid operation data, equipment working condition parameters and energy market indicators. This improper combination or separation of system risk derived from objective quantitative data and cognitive uncertainty derived from subjective expert judgment leads to distortion of input information before it enters the decision reasoning model, thereby reducing the accuracy of energy planning and prediction and the operational robustness of the final energy configuration scheme. In terms of the explainability and transparency of decision logic, many modern data-driven decision support tools that incorporate advanced algorithms have complex internal operation mechanisms that make them a "black box" in management. This lack of transparency in the decision-making process seriously affects the decision-maker's trust in the analysis results, hindering the effective adoption of the model. Especially in decision-making scenarios that require project feasibility demonstration, meet industry regulatory requirements or conduct major scheme review, a reasoning model or semantic network that cannot explain its internal logic will greatly limit its practical engineering and management application value.

[0003] Therefore, a multi-objective decision analysis method based on fuzzy logic is proposed. SUMMARY

[0004] The purpose of the present application is to provide a multi-objective decision analysis method based on fuzzy logic, which addresses the complex scenario of both quantitative data noise and expert judgment hesitation in the decision-making process. By integrating interval type-2 and spherical fuzzy sets and VIKOR algorithm, a completely interpretable decision analysis method is achieved. To achieve the above purpose, the present application provides the following technical scheme: a multi-objective decision analysis method based on fuzzy logic, comprising: obtaining initial evaluation values and dividing the evaluation criteria into a quantitative index set and a qualitative index set, for the quantitative index set, taking the initial evaluation value as the center value, and applying resampling to generate an interval type-2 fuzzy set; For the set of qualitative indicators, receive a judgment data stream given in the form of a spherical fuzzy set, the spherical fuzzy set containing membership, non-membership and hesitancy; apply a perception-hesitancy-aware intelligent aggregation operator to aggregate the judgment data stream, and generate an aggregated spherical fuzzy set for each criterion in the set of qualitative indicators; Construct the interval bivariate fuzzy set and the aggregated spherical fuzzy set into a heterogeneous fuzzy decision matrix, extract decision feature values, apply a hybrid fuzzy VIKOR algorithm to process the heterogeneous fuzzy decision matrix, calculate the group utility value, individual regret value and comprehensive ranking index of each energy configuration scheme, and output the compromise ranking result of the energy configuration scheme; Train an adaptive neuro-fuzzy inference network to fit the mapping relationship between the decision feature values and the compromise ranking result; and extract enhanced fuzzy rules to generate interpretable decision basis.

[0005] Preferably, the specific steps of dividing the evaluation criteria into the set of quantitative indicators and the set of qualitative indicators include: receiving an energy configuration scheme and evaluation criteria for evaluating the energy configuration scheme; dividing the full-life payback period, comprehensive energy conversion efficiency, park load matching degree and annual carbon emissions into the set of quantitative indicators; and dividing the technology maturity, system energy supply reliability, future system expandability and regional energy policy compliance into the set of qualitative indicators.

[0006] Preferably, the specific steps for the set of quantitative indicators include: using a historical project data set and an energy system simulation data set, applying statistical resampling to process the set of quantitative indicators, and determining the numerical fluctuation range of each criterion; and taking the initial evaluation value as the center value and the numerical fluctuation range as the upper and lower membership boundaries to obtain an interval bivariate fuzzy set.

[0007] Preferably, the historical project data set includes equipment expenditure, operation and maintenance expenditure, price fluctuation data, investment cost overrun rate, project construction delay records of historical energy projects, and corresponding energy conversion efficiency, annual average equipment failure rate and available hours of these projects in actual operation; The energy system simulation data set includes annual income and expenditure data, degree electricity cost, unit output carbon emissions, grid acceptance capacity and dependence on external energy input of the energy system under market price fluctuation scenarios.

[0008] Preferably, the specific steps for the set of qualitative indicators include: receiving a spherical fuzzy set formed by an expert knowledge base under the same criterion for the same energy configuration scheme, the expert knowledge base being a structured database of knowledge, experience and reasoning rules in the energy field; verifying whether the membership, non-membership and hesitancy of the spherical fuzzy set satisfy the numerical constraint conditions of the spherical fuzzy set, and eliminating invalid data; For the verified spherical fuzzy set, the hesitancy value is extracted, and a preset monotone decreasing nonlinear trust function is applied to generate a trust weight; a dynamic weighted spherical arithmetic average aggregation operator is called to weight and aggregate the effective spherical fuzzy set using the trust weight, and the membership, non-membership and hesitancy of the aggregated spherical fuzzy set are calculated respectively to generate an aggregated spherical fuzzy set; Preferably, the specific steps of the compromise ranking result of the output energy configuration scheme include: filling the interval type-2 fuzzy set and the aggregated spherical fuzzy set into a matrix with energy configuration schemes as rows and evaluation criteria as columns to form a heterogeneous fuzzy decision matrix, and normalizing the matrix to eliminate the dimension effect; the calculation steps of the decision characteristic value: for the interval type-2 fuzzy set, the iterative center point method is used to determine the center value, and for the aggregated spherical fuzzy set, the scoring function is used to integrate into a single evaluation value; The hybrid fuzzy VIKOR algorithm is applied to compare each column in the normalized matrix to determine a hybrid fuzzy positive ideal solution composed of optimal evaluation values and a hybrid fuzzy negative ideal solution composed of worst evaluation values; the hybrid distance measure is applied to calculate the separation degree between the energy configuration scheme and the positive / negative ideal solution, calculate the group utility value representing the overall performance of the energy configuration scheme, and calculate the individual regret value representing the maximum short board under a specific single criterion; The group utility value and the individual regret value are combined by weighting to calculate the comprehensive ranking index of the energy configuration scheme, and all energy configuration schemes are arranged in ascending order according to the comprehensive ranking index to obtain the compromise ranking result.

[0009] Preferably, the method further includes: generating an energy strategy report for the decision maker based on the calculated group utility value, individual regret value and comprehensive ranking index, the energy configuration scheme with the first ranking of the comprehensive ranking index is determined as the benchmark optimal strategy, and the energy configuration scheme with the lowest individual regret value is determined as the risk-averse strategy; the energy strategy report also includes decision trigger point analysis, which analyzes the sensitivity of key decision indicators, including: full-life investment payback period, annual average carbon emissions, regional energy policy compliance and technology maturity; identifies the critical change condition that causes the ranking between the risk-averse strategy and the benchmark optimal strategy to reverse; and quantifies the critical change condition to determine the decision reversal singularity.

[0010] Preferably, the steps of training the adaptive neural fuzzy inference network to generate an interpretable decision basis include: The adaptive neuro-fuzzy inference network is applied to take the characteristic value as input and the comprehensive ranking index as learning target, to fit the mapping relationship between the compromise ranking result and the evaluation value, and to construct an inference model for reproducing the mapping relationship; an initial fuzzy rule is automatically generated from the inference model, and is refined and induced to extract an enhanced fuzzy rule, which is combined into a decision logic path and output as an interpretable decision basis.

[0011] Compared with the prior art, the application has the following beneficial effects: 1. By innovatively combining interval type-2 fuzzy sets and spherical fuzzy sets, the measurement noise in objective data and the hesitancy degree in expert subjective judgment can be accurately modeled respectively; this method effectively solves the problem that traditional decision inference models are difficult to compatibly process heterogeneous uncertainty, so that the evaluation of a complex system (such as a comprehensive energy system) is closer to reality, and the rigor of the decision analysis process and the reliability of the final result are significantly improved.

[0012] 2. The hybrid fuzzy VIKOR algorithm not only focuses on the overall comprehensive performance (group utility) of the energy configuration scheme, but also particularly considers the worst performance (individual regret) that may occur under a single criterion; this balanced decision-making mechanism that takes into account both “collective satisfaction” and “individual risk” can effectively avoid selecting a scheme with fatal shortcomings, thereby screening out a more robust and lower-risk compromise ranking result, and comprehensively improving the quality of multi-objective trade-off decision-making.

[0013] 3. By training ANFIS and extracting core fuzzy rules with confidence scores, the final ranking result can be automatically translated into a set of human-readable and structured decision logic; this high degree of interpretability not only significantly improves the transparency of the decision-making process, but also enhances the understanding and trust of decision-makers in the model results, providing a solid logical support for the implementation and promotion of the scheme. BRIEF DESCRIPTION OF DRAWINGS

[0014] Figure 1 A step flowchart of the multi-objective decision analysis method based on fuzzy logic of the application; Figure 2 A flowchart of the multi-objective decision analysis method based on fuzzy logic of the application; Figure 3 A flowchart of the ANFIS rule extraction of the application. DETAILED DESCRIPTION

[0015] With reference to the accompanying drawings, the technical solutions in the embodiments of the present application will be clearly and completely described below, obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments; based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor belong to the protection scope of the present application.

[0016] Please refer to Figures 1 to 3 , the present application provides a multi-objective decision analysis method based on fuzzy logic, referring to Figure 1 step flow chart and Figure 2 flow chart; a multi-objective decision analysis method based on fuzzy logic, comprising: obtaining initial evaluation value and dividing evaluation criteria into quantitative index set and qualitative index set, for the quantitative index set, taking the initial evaluation value as the center value, applying resampling to generate interval type-2 fuzzy set; for the qualitative index set, receiving judgment data stream given in the form of spherical fuzzy set, the spherical fuzzy set contains membership, non-membership and hesitation; applying intelligent aggregation operator with perception of hesitation to aggregate the judgment data stream, generating aggregated spherical fuzzy set for each criterion in the qualitative index set; constructing the interval type-2 fuzzy set and the aggregated spherical fuzzy set into heterogeneous fuzzy decision matrix, extracting decision feature value, applying hybrid fuzzy VIKOR algorithm to process the heterogeneous fuzzy decision matrix, calculating group utility value, individual regret value and comprehensive ranking index of each energy configuration scheme, and outputting the compromise ranking result of the energy configuration scheme; training adaptive neural fuzzy inference network to fit the mapping relationship between decision feature value and compromise ranking result; and extracting enhanced fuzzy rules to generate interpretable decision basis.

[0017] Embodiment one: This embodiment is based on the optimization configuration of an industrial park energy system, which is planning to upgrade and transform the existing energy system, facing a variety of feasible energy configuration schemes, and the park decision-making layer hopes to select the optimization scheme that best meets the development needs of the park in terms of economy, efficiency, environmental protection and sustainability, etc.

[0018] Firstly, the following three energy configuration schemes are determined: Energy configuration scheme one: using high-efficiency gas distributed energy system combined with small-scale rooftop photovoltaic; Energy configuration scheme two: vigorously developing centralized photovoltaic and wind power, and supporting large-scale energy storage system; Energy configuration scheme three: introducing ground source heat pump technology, cooperating with distributed photovoltaic and energy storage system to realize park cold-heat-electricity multi-generation.

[0019] Further, the criteria for evaluating these energy configuration schemes are classified into two categories: quantitative indicators, which can be accurately measured by data, including total investment cost, overall energy conversion efficiency, load matching degree, and total carbon emissions per year; and qualitative indicators, which require experience to judge, including the maturity of the technology used, the reliability of the energy supply system, the expansion potential of the future system, and the degree of compliance with the existing regional energy policy. By systematically dividing the evaluation criteria into data-driven quantitative indicators and expert-driven qualitative indicators, the method achieves targeted processing of different types of data. This division allows the method to match the most suitable fuzzification and aggregation techniques for each type of indicator, avoiding information loss and precision decline when a single model processes heterogeneous data, thereby ensuring the accuracy of the initial data modeling and laying a solid foundation for subsequent precise ranking and analysis.

[0020] Further, for the quantitative indicator set, a data resampling method is used to quantify the inherent uncertainty and construct an interval type-2 fuzzy set. Specifically, taking the "full life investment payback period" as an example, a large amount of historical project data and energy system simulation data are collected. The historical project data includes initial investment, operation and maintenance costs, and past investment overruns for similar projects. For example, it is found that there is a 10% to 25% investment overrun rate for similar projects. Energy system simulation data simulates the annual income and expenditure of each energy configuration scheme under the fluctuation scenarios of key energy factors such as natural gas prices and electricity prices in the next ten years.

[0021] Next, statistical resampling techniques are applied to process the above data to determine the numerical fluctuation range of the payback period. The cost overrun rate, delay records, and operation and maintenance cost fluctuations in historical data are combined with future market price fluctuations in simulation data. Through statistical resampling methods such as Monte Carlo, thousands of possible future scenarios are simulated. In each scenario, a corresponding "full life investment payback period" is calculated. Through statistical analysis of the thousands of results, a numerical fluctuation range with high confidence (e.g., 90%) is ultimately determined. For example, the analysis results show that the investment payback period of "Energy Configuration Scheme One" has a 90% probability of falling between 7.5 years and 10.5 years.

[0022] With the initial evaluation value as the center, combined with the numerical fluctuation range, an interval two-type fuzzy set is generated; the initial evaluation value "8.5 years" is taken as the center value of the constructed interval two-type fuzzy set; at the same time, the calculated numerical fluctuation range "[7.5, 10.5] years" is taken as the basis for defining the upper and lower membership boundaries of the fuzzy set; the finally generated interval two-type fuzzy set is no longer a isolated value "8.5", but a fuzzy number that can fully express its uncertainty; this fuzzy number will be placed in the heterogeneous fuzzy decision matrix as the final quantitative evaluation result of "energy configuration scheme one" in the "full life investment return period" criterion, which is used for subsequent comprehensive sorting calculation; in this way, the decision reasoning model fully considers the risks brought by cost, construction period and market price fluctuations in the real world.

[0023] By applying statistical resampling to analyze the cost overrun rate and market price fluctuation data, the core risk factors in project implementation and market environment are endogenously integrated into the calculation model in the form of "risk exposure interval"; this makes the analysis results no longer based on idealized static cost, but fully considers the volatility of the real economic environment, significantly enhances the real guiding significance and risk prediction ability of the evaluation results, and improves the robustness of the final scheme.

[0024] Further, for the set of qualitative indicators, the judgment data stream is directly extracted from the expert knowledge base; the expert knowledge base is a structured database that is pre-constructed and contains a large amount of historical project evaluation data, industry standards and coded rules; The construction process mainly includes the following three key steps: First step: collection and transformation of historical project evaluation data, this step aims to dataize past experience.

[0025] A large amount of historical project data in the energy field is collected, such as historical project evaluation reports, feasibility studies, expert review opinions, post-project evaluation reports, industry best practice rules, etc. These materials contain evaluation conclusions on various energy configuration schemes in "technical maturity", "system energy supply reliability" and other qualitative indicators. The collected unstructured or semi-structured evaluation conclusions are transformed into a unified spherical fuzzy set format. This transformation process requires the participation of field experts, who assign three-dimensional evaluation values to each evaluation case according to the description of historical data: recognition degree (membership degree) represents the support degree of the historical evaluation to the current indicator; non-recognition degree (non-membership degree) represents the degree of opposition or non-support; hesitation degree (hesitation degree) represents the uncertainty, controversy or insufficient information in the original evaluation opinion.

[0026] Second step: Coding of industry standards and best practices, which aims to formalize the recognized industry guidelines, comb the various industry standards, design specifications, safety guidelines and recognized best practice guidelines in the energy field, and "code" these standards and guidelines into IF-THEN rules that can be executed by machines.

[0027] Third step: Building a structured database The above two types of data are stored in a structured database, and each record of the database is a judgment data stream, which should include at least the following core fields: Evaluation criteria ID: such as "technology maturity", "regional energy policy compliance" and so on.

[0028] Energy configuration scheme ID: the specific scheme related to the judgment.

[0029] Data source: indicates whether the data is derived from "historical project evaluation" or "coded industry rules".

[0030] Spherical fuzzy set data: contains three values of membership, non-membership and hesitation.

[0031] These judgment data are given in the form of spherical fuzzy sets, each of which contains "approval" (membership), "disapproval" (non-membership) and "hesitation" (hesitation) for a certain indicator; For example, in the evaluation of strategy one "technology maturity", for example, two sets of relevant historical evaluation data are retrieved from the knowledge base: Data one (derived from historical evaluation of similar projects), approval = 0.8, disapproval = 0.1, hesitation = 0.3; Data two (derived from coded industry best practice rules), approval = 0.6, disapproval = 0.3, hesitation = 0.5; Specifically, first, the effectiveness of the spherical fuzzy set extracted from the knowledge base is checked to ensure that it meets the mathematical definition of a spherical fuzzy set, i.e. the sum of the squares of the membership, disapproval and hesitation is not greater than 1, and the data that does not meet the requirements is marked or removed.

[0032] Further, for the spherical fuzzy set that passes the verification, the "hesitation" information is extracted; a pre-set monotonic decreasing nonlinear trust function is used, which adopts a Gaussian decay function, to convert the hesitation of each data stream into a trust weight; This weight is completely determined by the hesitation of the data itself, the higher the hesitation, the greater the uncertainty of the data, and the lower the trust weight obtained.

[0033] A weighted spherical arithmetic mean aggregation operator is called to aggregate the multiple valid spherical fuzzy sets using the generated trust weights; the weighted spherical arithmetic mean aggregation operator is an operator for fusing multiple spherical fuzzy set information, calculating a comprehensive spherical fuzzy set that can represent all independent judgments through weighted averaging, fusing multiple independent fuzzy judgments into a comprehensive conclusion, and intelligently amplifying judgments with high certainty and suppressing judgments with strong uncertainty through weights based on "hesitation degree", thereby improving the reliability of the final result; the specific steps are as follows: Prepare input data: the input of the operator is N valid spherical fuzzy sets verified (for example, S1, S2,..., SN) and N trust weights (W1, W2,..., WN) corresponding to them; Aggregate membership degree: multiply the membership degree (approval degree) of each spherical fuzzy set by its corresponding trust weight, then add all the products, and finally divide by the sum of all trust weights to obtain the aggregated comprehensive membership degree; Aggregate non-membership degree: use the same weighted averaging method as in the previous step to calculate the non-membership degree (disapproval degree) of all spherical fuzzy sets to obtain the aggregated comprehensive non-membership degree; Aggregate hesitation degree: again use the same weighted averaging method to calculate the hesitation degree (hesitation degree) of all spherical fuzzy sets to obtain the aggregated comprehensive hesitation degree; combine the obtained comprehensive membership degree, comprehensive non-membership degree, and comprehensive hesitation degree into a new, single spherical fuzzy set.

[0034] Further, two dimensions can be added: "historical accuracy" (the prediction accuracy of the expert knowledge in similar past decisions) and "group consensus degree" (the degree of deviation of the judgment from the center of group opinion); the aggregation operator will no longer be simply weighted, but will dynamically adjust the weights of the three dimensions, for example, in the initial decision-making stage with greater disagreement, the weight of "historical accuracy" is increased; when the opinions need to be unified, the weight of "group consensus degree" is increased.

[0035] Through aggregation, a comprehensive spherical fuzzy set for each qualitative indicator is obtained, which fuses the judgments of multiple related data sources in the knowledge base and reflects the collective, uncertainty-adjusted judgment result; this aggregated spherical fuzzy set will be placed in the heterogeneous fuzzy decision matrix as the quantitative evaluation result of the qualitative indicator.

[0036] Further, after completing the fuzzification of all quantitative and qualitative indicators, the results are filled into a matrix with energy configuration schemes as rows and evaluation criteria as columns, forming a heterogeneous fuzzy decision matrix; the matrix contains interval type-2 fuzzy sets generated from the quantitative indicator set and spherical fuzzy sets aggregated from the qualitative indicator set; then, the matrix is normalized to map all different types of fuzzy evaluation values into a uniform interval to eliminate the influence of different dimensions of criteria.

[0037] Further, according to the benefit or cost properties of each criterion, the columns of the normalized matrix are compared one by one. For benefit-type criteria, such as overall energy conversion efficiency, the fuzzy number with the optimal evaluation value in the column is selected; for cost-type criteria, such as full-life payback period, the fuzzy number with the worst evaluation value in the column is selected; by filtering all criteria one by one, a mixed fuzzy positive ideal solution composed of optimal evaluation values and a mixed fuzzy negative ideal solution composed of worst evaluation values are determined.

[0038] Specifically, the calculation of distance needs to be done in two steps using a pre-set mixed distance measure; the first step is to independently calculate the distance between the evaluation value of the scheme and the ideal evaluation value under each evaluation criterion, and the second step is to combine all the independently calculated distance components into a total separation degree; the separation degree between each energy configuration scheme and the above-mentioned positive and negative ideal solutions is calculated; through this calculation, two key indicators are generated for each energy configuration scheme: The group utility value represents the overall closeness of the scheme to the positive ideal solution, and the smaller the value, the closer the comprehensive performance to the optimal level; the individual regret value represents the maximum gap between the scheme and the positive ideal solution under all single criteria, and the smaller the value, the less obvious the fatal shortcoming of the scheme; after the distance calculation with the positive and negative ideal solutions, the evaluation values of the three schemes are as follows: first, collect and organize the "group utility value (S)" and "individual regret value (R)" of all schemes to be evaluated; Secondly, among all the S and R values of the schemes, the optimal value (i.e. the minimum value S and R) and the worst value (i.e. the maximum value S⁻ and R⁻) are determined; these values will be used as a reference for subsequent calculations.

[0039] Then, for each scheme, the core normalization calculation is performed, which is as follows: Calculate the standardization performance of S value: First, calculate the individual gap between the S value of the current scheme and the optimal S value (S*). Secondly, calculate the maximum possible gap between the worst S value (S⁻) and the optimal S value (S*). Finally, the individual gap is divided by the maximum possible gap to obtain a ratio between 0 and 1, which is the normalized performance of S value; the same method as S value is used to calculate the gap between the current R value and the optimal R value (R*), and then divide it by the maximum possible gap of R value (R- minus R*), so as to obtain the "normalized performance" of R value; Finally, by weighting the normalized performance of the group utility value S and the normalized performance of the individual regret value R of each scheme, the final comprehensive ranking index is calculated by weighting the normalized performance of S and the normalized performance of R, and the weights of S and R are both set to 0.5; the index comprehensively reflects the overall performance and stability of the scheme; the calculation result is as follows: The comprehensive ranking index Q of the first energy configuration scheme is 0.88, the comprehensive ranking index Q of the second energy configuration scheme is 0.03, and the comprehensive ranking index Q of the third energy configuration scheme is 0.35. According to the comprehensive ranking index Q value, all energy configuration schemes are arranged in ascending order (the smaller the Q value, the better the scheme), and the compromise ranking result of the embodiment is obtained; the final ranking result is: the second energy configuration scheme > the third energy configuration scheme > the first energy configuration scheme.

[0040] The mixed fuzzy VIKOR algorithm is used to calculate the group utility value and the individual regret value, and the energy configuration schemes are comprehensively ranked; the advantage of this ranking mechanism is that it realizes the balance between "overall optimization" and "local worst avoidance", which not only can screen out the best option in comprehensive performance, but also can identify and warn the option with major short board in any single key criterion; this provides a more comprehensive and more robust perspective for users to evaluate different schemes.

[0041] Further, based on the group utility value, the individual regret value and the comprehensive ranking index calculated in the foregoing, a detailed energy strategy report is generated for the park decision maker; Specifically, in the report, the "second energy configuration scheme" ranked first (Q=0.03) according to the comprehensive ranking index is determined as the benchmark optimal strategy; at the same time, the "third energy configuration scheme" with the lowest individual regret value (R=0.33) is determined as the risk avoidance strategy, which means that it has the most balanced performance in all single criteria and does not have the most serious short board.

[0042] Further, the energy strategy report also includes a decision trigger point analysis to reveal the possible change of the current optimal choice when the key external conditions change; this work is completed by sensitivity disturbance analysis on the preset key decision indicators; in this embodiment, the selected key decision indicators are full life investment payback period, annual average carbon emission, regional energy policy compliance and technology maturity.

[0043] Specifically, the start of sensitive disturbance analysis process to identify the risk-averse strategy (Scheme III) and the benchmark optimal strategy (Scheme II) between the ranking of critical change conditions resulting in reverse; with the index of "full life investment payback period" as an example, will simulate a scenario: due to the key energy storage equipment supply chain problems, resulting in the initial investment cost of scheme II rises; with its initial evaluation value as the base point, gradually increase the "full life investment payback period" of scheme II, and after each step increment re-run the complete sorting calculation process; analysis process continues until the comprehensive ranking index Q value of scheme II exceeds the Q value of scheme III.

[0044] Finally, it is identified that when the "full life investment payback period" of "energy configuration scheme II" is extended beyond 2.5 years based on its initial evaluation value, its comprehensive ranking index will be inferior to "energy configuration scheme III", resulting in a fundamental reversal of decision ranking; this critical change condition is quantified and determined as a decision reversal singularity in this analysis; the singularity provides a clear risk threshold for the decision maker, indicating the key external variables that need to be monitored and their safety boundaries.

[0045] The calculated group utility value, individual regret value and comprehensive ranking index of each energy configuration scheme are input parameters of a preset fuzzy portfolio optimization model; the fuzzy portfolio optimization model aims to achieve the Pareto optimum of maximizing overall investment benefit and minimizing risk exposure under the preset resource constraints; and according to the solving result of the model, an optimal resource allocation strategy for multiple energy configuration schemes is output, which includes specific resource input ratio, to replace single scheme selection; the optimal resource allocation strategy at least determines the resource input percentage of the benchmark optimal strategy and the risk-averse strategy, and through portfolio optimization, it realizes the active hedging of potential risks while pursuing high expected returns, providing a mathematically optimal balance point.

[0046] The decision trigger point analysis is introduced, which can quantitatively identify the critical conditions that cause the fundamental reversal of energy configuration scheme ranking by sensitive disturbance of key indicators; this function goes beyond simple static sorting, revealing the vulnerability and potential risks hidden behind the current optimal solution, providing dynamic and forward-looking deep insights, greatly improving the practical value and decision support of the analysis result in complex and changing environment.

[0047] Further, the step of training an adaptive neuro-fuzzy inference system ANFIS to generate interpretable decision basis is implemented, and the specific parameter configuration of the constructed ANFIS is as follows: the model type adopts first-order Sugeno type fuzzy reasoning; The network comprises 6 network layers, namely, an input layer, a fuzzification layer, a rule layer, a normalization layer, a defuzzification layer and an output layer. The input layer has 4 nodes corresponding to the full-life investment payback period, the annual average carbon emission, the technology maturity and the regional energy policy compliance. The fuzzification layer adopts 3 generalized bell-shaped membership functions for fuzzy processing of each input variable, and each function has 3 nonlinear premise parameters to be learned. The rule layer has 81 nodes, each node representing a fuzzy rule. The defuzzification layer is responsible for calculating the specific output value of each independent fuzzy rule. The normalized trigger intensity, i.e., the relative weight, of the rule is multiplied by the calculation result of the linear function of the rule conclusion part to obtain the weighted output of the rule as the output layer. The learning algorithm adopts a hybrid learning algorithm for training, i.e., the premise parameters are updated using the gradient descent method, and the conclusion parameters are identified using the least squares method. The iteration period of training is set to 200 times, and the initial step size is set to 0.01.

[0048] Referring to Figure 3 For the flowchart of ANFIS rule extraction of the present application, specifically, first, an inference model capable of reproducing the decision ranking result is constructed and fitted; from the heterogeneous fuzzy decision matrix formed in the foregoing step, a set of decision feature values is extracted for each energy configuration scheme; for the interval type-2 fuzzy set representing the quantitative indicators in the heterogeneous fuzzy decision matrix, the iterative center point method is applied for defuzzification to calculate a deterministic scalar value that can uniquely represent the center position; for the aggregated spherical fuzzy set representing the qualitative indicators in the matrix, a scoring function is used, specifically, the square of the membership degree is subtracted from the square of the non-membership degree, and then the square of the hesitancy degree is further subtracted to calculate a single evaluation value; finally, the deterministic scalar values of all quantitative indicators corresponding to each energy configuration scheme are combined with the single evaluation values of all qualitative indicators in order to form a standardized decision feature vector.

[0049] Further, the above data (and more data from historical or simulation cases) is submitted to the adaptive neuro-fuzzy inference network, and the system fits the mapping relationship between the compromise ranking result and the evaluation value through iterative learning until the model is constructed; For the three energy configuration schemes, the corresponding partial input evaluation values and learning goals are shown in Table 1: Table 1: Partial input evaluation values and learning goals corresponding to each energy configuration scheme Energy configuration scheme Full-life investment payback period Annual average carbon emissions Technical maturity Regional energy policy compliance Learning target (Q value) Scheme 1 8.5 2.5 0.72 0.65 0.88 Scheme 2 11.2 0.3 0.75 0.98 0.03 Scheme 3 9.5 1.0 0.80 0.90 0.35 Further, from the constructed inference model, an initial fuzzy rule set containing dozens of rules is automatically generated; then, the system refines and induces the initial rule base, by evaluating the contribution of each rule to distinguishing the optimal strategy, and eliminating redundant or weakly affected rules, to finally refine a core rule combination composed of a few enhanced fuzzy rules; specifically, the rule combination with the highest contribution to the compromise ranking result is evaluated and retained; for example, one of the retained high-contribution enhanced fuzzy rules is: "if (low annual carbon emissions) and (high regional energy policy compliance) are met, then (comprehensive ranking index is optimal)" ; finally, the enhanced fuzzy rule combination is combined into a decision logic path, which is output as the interpretable decision basis.

[0050] When it is necessary to explain why "energy allocation scheme two" is rated as optimal, its input vector is matched with the core rules, and the final decision logic path is output, which presents as: "the core decision logic for why energy allocation scheme two is optimal is that the annual carbon emissions evaluation value (0.3) of the scheme is extremely low, and the regional energy policy compliance evaluation value (0.98) is extremely high, which triggers the core decision rule pointing to the optimal level, thus winning in the comprehensive ranking." The adaptive neural fuzzy inference network is used to fit the evaluation values and ranking results, and enhanced fuzzy rules are extracted therefrom; the complex nonlinear mapping relationship is converted into a logic path understandable by humans; thus, the inference model can express complex fuzzy reasoning as reasoning with stronger logic, and understand the key factor combination and its internal logic that affect the final result.

[0051] Embodiment Two The application scenario of this embodiment is the same as that of Embodiment One, however, to meet the higher requirements of the park decision-making layer for deep mining of risk assessment and decision basis, the analysis process adopted by this embodiment is as follows: First, in the step of fuzzy processing of the evaluation criteria, a higher-order fuzzy set representation and data-driven modeling method is adopted in the specific implementation; specifically, for the quantitative index set, including full-life investment payback period, comprehensive energy conversion efficiency, park load matching degree, and annual carbon emissions, this embodiment introduces an interval type-2 fuzzy set construction method based on probability density function and a-cut set; the step of generating the interval type-2 fuzzy set specifically includes: kernel density estimation is performed on the resampled data to fit the probability density function thereof; and based on the probability density function, the interval type-2 fuzzy set is constructed through multi-level a-cut set and weighted integration, so that the membership function boundary can reflect the distribution characteristics of the original data; specifically as follows: Firstly, the historical project dataset and the energy system simulation dataset are used not only to determine the fluctuation range of the numerical value, but also to fit the probability density function of each index evaluation value; for example, by kernel density estimation on a large number of resampled data of the full-life payback period, it is found that the distribution presents an asymmetric long-tail shape; by setting multiple confidence levels on the fitted probability density function, a- cut sets such as (a = 0.8, 0.9, 0.95) are generated for each confidence level; then, these nested intervals generated by different a- cut sets are weighted integrated, and finally an interval-valued fuzzy set is constructed which can more accurately describe the uncertainty of data distribution; the fuzzy set generated by this method has upper and lower boundary shapes of the membership function that are more consistent with the real distribution characteristics of the original data, and compared with simply taking the fluctuation range as the boundary, it can more finely depict the risk.

[0052] Further, in the ranking decision stage, an improved VIKOR algorithm considering the risk preference of the decision maker is adopted; specifically, after the construction of the heterogeneous fuzzy decision matrix containing high-order interval-valued fuzzy sets and aggregated interval spherical fuzzy sets and normalization, on the basis of determining the mixed fuzzy positive and negative ideal solutions, the risk preference coefficient m of the decision maker is introduced.

[0053] Further, when calculating the separation degree between each scheme and the ideal solution, the risk preference coefficient m is used to adjust the distance measure function; when the decision maker prefers risk aversion (m > 0.5), the distance measure will disproportionately magnify the short board of the scheme in any single criterion, i.e. the distance from the positive ideal solution, so that the scheme with fatal defects will be more severely punished; on the contrary, when the decision maker prefers risk seeking (m < 0.5), the distance measure will focus more on the overall average performance of the scheme; Specifically, this risk preference adjustment mechanism directly affects the calculation of group utility value S and individual regret value R, so that the final comprehensive ranking index Q not only reflects the pros and cons of the scheme, but also embeds the risk attitude of the decision maker; for example, even if scheme two has the lowest Q value in the conventional calculation, but if it performs extremely poorly in a key indicator and the decision maker chooses a high risk aversion (m = 0.8), then its individual regret value R will be significantly magnified, which may lead to its final Q value worse than that of scheme three which performs more balanced in all indicators.

[0054] Further, in the decision explanation stage, the method uses an ensemble learning model to mine deeper decision logic; specifically, after generating the energy strategy report and identifying the benchmark optimal strategy and risk-averse strategy, the embodiment does not use ANFIS, but calls the gradient boosting decision tree GBDT model to fit the mapping relationship between the evaluation value and the comprehensive ranking index; as a powerful ensemble learning model, GBDT can capture more complex, high-order nonlinear relationships and interaction effects between input features.

[0055] Further, after the model training is completed, instead of directly extracting fuzzy rules, SHAP analysis is used to generate interpretable decision basis; for each evaluation criterion of each scheme, an accurate contribution value is calculated, quantifying to what extent the criterion will push up or pull down the final ranking of the scheme; finally, the SHAP analysis results are integrated into a visual decision force waterfall chart and output; the chart clearly shows that for the optimal scheme, how its evaluation criteria accumulate its advantages step by step, and how its disadvantage criteria weaken its total score; this decision logic path not only tells the decision maker "what", but also reveals "why" and "how important" in quantitative form, providing more profound and comprehensive decision insights than single fuzzy rules.

[0056] The embodiment is suitable for scenarios with higher requirements for risk and decision depth; when dealing with quantitative indicators, it uses a method based on probability density function and alpha-cut, which can more accurately construct interval type-2 fuzzy sets that conform to the true distribution of data; for qualitative indicators, it introduces interval spherical fuzzy sets to better express expert uncertainty, and through consensus measurement and feedback mechanism ensures high consistency of expert opinions; in the ranking stage, it uses an improved VIKOR algorithm that introduces a decision maker risk preference coefficient, so that the ranking result can dynamically reflect the decision maker's risk attitude; finally, in the decision explanation link, it uses a more powerful gradient boosting decision tree model, and combines SHAP analysis to provide a more profound and quantitative attribution analysis of the decision result in the form of decision force waterfall chart.

[0057] Although embodiments of the present application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made therein without departing from the principles and spirit of the application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method of multi-objective decision analysis based on fuzzy logic, characterized by, The method comprises the following steps: obtaining initial evaluation values and dividing evaluation criteria into a quantitative index set and a qualitative index set, for the quantitative index set, taking the initial evaluation values as the center values, and applying resampling to generate interval two-type fuzzy sets; for the qualitative index set, receiving a judgment data stream in the form of a spherical fuzzy set, the spherical fuzzy set containing membership, non-membership and hesitation; applying an intelligent aggregation operator that perceives hesitation to aggregate the judgment data stream, and generating an aggregated spherical fuzzy set for each criterion in the qualitative index set; constructing the interval two-type fuzzy sets and the aggregated spherical fuzzy sets into a heterogeneous fuzzy decision matrix, extracting decision characteristic values, applying a hybrid fuzzy VIKOR algorithm to process the heterogeneous fuzzy decision matrix, calculating the group utility value, individual regret value and comprehensive ranking index of each energy configuration scheme, and outputting the compromise ranking result of the energy configuration scheme; training an adaptive neural fuzzy inference network to fit the mapping relationship between the decision characteristic values and the compromise ranking result; and extracting enhanced fuzzy rules to generate an interpretable decision basis.

2. The method of claim 1, wherein the method is based on fuzzy logic. The specific steps of dividing the evaluation criteria into the quantitative index set and the qualitative index set comprise: receiving energy configuration schemes and evaluation criteria for evaluating the energy configuration schemes; dividing the full-life payback period, comprehensive energy conversion efficiency, park load matching degree and annual average carbon emissions into the quantitative index set; and dividing the technology maturity, system energy supply reliability, future system expandability and regional energy policy compliance into the qualitative index set.

3. The method of claim 1, wherein the method is based on fuzzy logic. The specific steps for the quantitative index set comprise: using a historical project data set and an energy system simulation data set, applying statistical resampling to process the quantitative index set, determining the numerical fluctuation range of each criterion; and taking the initial evaluation values as the center values and the numerical fluctuation range as the upper and lower membership boundaries to obtain the interval two-type fuzzy sets.

4. The multi-objective decision analysis method based on fuzzy logic according to claim 3, wherein the historical project data set includes equipment expenditure, operation and maintenance expenditure, price fluctuation data, investment cost overrun rate, project construction delay records of historical energy projects, and the corresponding energy conversion efficiency, annual average equipment failure rate and available hours of these projects in actual operation; the energy system simulation data set includes annual income and expenditure data, degree electricity cost, unit output carbon emissions, grid accommodation capacity and dependence on external energy input of the energy system under market price fluctuation scenarios.

5. The method of claim 1, wherein the method is based on fuzzy logic. The specific steps for the qualitative index set comprise: receiving the spherical fuzzy sets formed by the expert knowledge base under the same criterion for the same energy configuration scheme, the expert knowledge base being a structured database of knowledge, experience and reasoning rules in the energy field; verifying whether the membership, non-membership and hesitation of the spherical fuzzy sets meet the numerical constraint conditions of the spherical fuzzy sets, and eliminating invalid data; For the verified spherical fuzzy set, the hesitancy value is extracted, and a preset monotone decreasing nonlinear trust function is applied to process to generate a trust weight; a dynamic weighted spherical arithmetic average aggregation operator is called to weight aggregate the effective spherical fuzzy set using the trust weight, to respectively calculate the membership, non-membership and hesitancy after aggregation, and to generate an aggregated spherical fuzzy set.

6. The method of claim 1, wherein the method is based on fuzzy logic. The specific steps of the compromise ranking result of the output energy configuration scheme include: The interval type-2 fuzzy set and the aggregated spherical fuzzy set are filled into a matrix with energy configuration schemes as rows and evaluation criteria as columns to form a heterogeneous fuzzy decision matrix, and the matrix is normalized to eliminate the dimension effect; the calculation steps of the decision characteristic value include: for the interval type-2 fuzzy set, the iterative center point method is used to determine the center value, and for the aggregated spherical fuzzy set, a scoring function is used to integrate into a single evaluation value; The hybrid fuzzy VIKOR algorithm is applied to compare column by column in the normalized matrix to respectively determine a hybrid fuzzy positive ideal solution composed of optimal evaluation values and a hybrid fuzzy negative ideal solution composed of worst evaluation values; the hybrid distance measure is applied to calculate the separation degree between the energy configuration scheme and the positive / negative ideal solution, to calculate the group utility value representing the overall performance of the energy configuration scheme, and the individual regret value representing the maximum short board under a specific single criterion; The group utility value and the individual regret value are combined by weighting to calculate the comprehensive ranking index of the energy configuration scheme, and all the energy configuration schemes are arranged in ascending order according to the comprehensive ranking index to obtain the compromise ranking result.

7. The method of claim 6, wherein the method further comprises: The method further includes: Based on the calculated group utility value, individual regret value and comprehensive ranking index, an energy strategy report is generated for the decision maker, the energy configuration scheme with the first ranking of the comprehensive ranking index is determined as the benchmark optimal strategy, and the energy configuration scheme with the lowest individual regret value is determined as the risk-averse strategy; the energy strategy report also includes decision trigger point analysis, through sensitivity disturbance analysis on key decision indicators including full-life investment payback period, annual average carbon emission, regional energy policy compliance and technology maturity, the critical change condition causing the ranking reversal between the risk-averse strategy and the benchmark optimal strategy is identified, and the decision reversal singularity is quantified.

8. The method of claim 1, wherein the method is based on fuzzy logic. The steps of training the adaptive neuro-fuzzy inference network to generate an interpretable decision basis include: The adaptive neuro-fuzzy inference network is applied to take the decision characteristic value as input and the comprehensive ranking index as learning target, to fit the mapping relationship between the compromise ranking result and the evaluation value, to construct an inference model to reproduce the mapping relationship; initial fuzzy rules are automatically generated from the inference model, and are refined and induced to extract enhanced fuzzy rules, which are combined into a decision logic path as an interpretable decision basis for output.

Images