An offshore floating photovoltaic site suitability evaluation method and system fusing XAI and MCDA methods, a computer device and a storage medium

Abstract

The application discloses a kind of offshore floating photovoltaic site suitability evaluation method, system, computer equipment and storage medium fused with XAI and MCDA method.Collect historical operation data, use multi-tree-based regression model and SHAP to calculate the contribution of each index to direct-current output power, obtain the ratio and build XAI-FAHP weight;Collect regional marine monthly environmental data, and calculate the weight based on monthly data using CRITIC;The comprehensive weight is obtained by combining the weights of XAI-FAHP and CRITIC using game theory;The grid is sorted using TOPSIS;Superimpose spatial constraints such as exclusive economic zone, marine protected area, tropical cyclone frequency, water depth, offshore distance and sea ice concentration, etc., to eliminate non-deployable areas, and obtain the evaluation result.The application reduces the influence of artificial subjective weighting when the sample and field experts are limited, and improves the objectivity and repeatability of the evaluation.

Description

Technical Field

[0001] This invention relates to new energy, marine engineering and intelligent decision-making technology, specifically to a method, system, computer equipment and storage medium for assessing the suitability of offshore floating photovoltaic sites by integrating interpretable artificial intelligence (XAI) and multi-criteria decision analysis (MCDA) methods. Background Technology

[0002] With the large-scale development of clean energy, offshore floating photovoltaic (PV) systems have attracted attention due to their lack of land-based resource occupation, proximity to coastal load centers, and compatibility with offshore wind power and marine ranching. Compared to inland floating PV systems, offshore floating PV systems can utilize a wider sea area, but they also face complex marine environmental risks such as salt spray corrosion, wave loads, strong winds, tropical cyclones, sea ice, and deep-water anchoring.

[0003] Existing research on photovoltaic or floating photovoltaic site selection mostly employs multi-criteria decision analysis methods or machine learning methods. Traditional multi-criteria decision-making methods often rely on human experience to determine the weights of indicators, which is highly subjective and difficult to maintain consistency on a global or cross-regional scale. Pure machine learning methods, on the other hand, require a large number of existing and publicly available offshore floating photovoltaic samples, but such projects are still in the demonstration stage and have limited long-term operational data.

[0004] Therefore, there is a need for a site suitability assessment method that can fully utilize limited real-world data on floating marine photovoltaic systems while adapting to the spatiotemporal heterogeneity of global or regional marine environments. This method should reduce subjective weighting and avoid relying solely on objective weighting based on the dispersion of single data points. It should also incorporate legal, ecological, and engineering constraints that preclude actual deployment as the final spatial elimination step, thereby obtaining suitability assessment results that more closely approximate engineering applications. Summary of the Invention

[0005] The purpose of this invention is to provide a method, system, computer equipment, and storage medium for assessing the suitability of offshore floating photovoltaic sites by integrating XAI and MCDA methods, in order to solve the problems of strong subjectivity of human weighting in existing site assessments, scarcity of offshore floating photovoltaic samples, and difficulty in characterizing the spatiotemporal heterogeneity of the marine environment.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a method for assessing the suitability of offshore floating photovoltaic sites by integrating XAI and MCDA methods, comprising the following steps:

[0007] Step 1: Collect historical operational data of offshore floating photovoltaic systems The historical operational dataset includes solar radiation indices, temperature indices, significant wave height, wind speed at 10 meters altitude, and DC output power; [The last sentence appears to be incomplete and possibly refers to a separate topic:] ...the historical operational dataset... Preprocessing is performed, including removing outliers, missing values, invalid nighttime data, and extreme outliers; aligning the time dimension; and normalizing or standardizing the data to obtain the training set. and test set ;

[0008] Step 2: Construct multiple tree-based regression models. Each tree-based regression model uses DC output power as the output index and solar radiation, temperature, significant wave height, and wind speed at 10 meters as input indexes. Train the multiple tree-based regression models using the training set and verify their performance using the test set. Calculate the contribution of each input index to DC output power using the SHAP interpretation method for each trained model. Further calculate the global importance of each input index and calculate the relative importance ratio between each input index based on the global importance.

[0009] Step 3: Construct an XAI-enhanced fuzzy hierarchical analysis judgment matrix based on the relative importance ratios obtained in Step 2; map the minimum, average, and maximum values ​​of the relative importance ratios of the same indicator pair in multiple models to triangular fuzzy numbers to form a fuzzy judgment matrix; calculate the fuzzy weights of each indicator using the fuzzy geometric mean method and perform normalization processing, and then use the centroid method to defuzzify the fuzzy weights to obtain the XAI-FAHP weights.

[0010] Step 4: Collect multi-source marine environmental datasets for marine grid cells within the assessment area. and spatially constrained datasets The multi-source marine environment dataset The spatially constrained dataset includes solar radiation indices, temperature indices, significant wave height, and wind speed at a height of 10 meters. This includes exclusive economic zones, marine protected areas, tropical cyclone frequency, water depth, offshore distance, and sea ice concentration; and the collection of multi-source marine environmental datasets for the assessment area. Monthly-scale statistics, spatial registration, and dimensionless processing were performed. The dynamic objective weights of each indicator were calculated using the CRITIC method. The CRITIC method simultaneously uses the standard deviation of the indicators to characterize the contrast intensity and the correlation coefficient between the indicators to characterize the degree of conflict, thereby obtaining the CRITIC weights based on monthly-scale data.

[0011] Step 5: Use game theory combinatorial weighting method to optimize and merge the XAI-FAHP weights obtained in Step 3 and the CRITIC weights obtained in Step 4, establish an optimization model that minimizes the deviation between the combined weight vector and each basic weight vector, solve and normalize the combination coefficients to obtain the comprehensive weights.

[0012] Step 6: Based on the multi-source marine environment dataset processed in Step 4 and the comprehensive weights obtained in Step 5, the TOPSIS (Topological Solution Approximation) method is used to calculate the comprehensive suitability assessment score of each marine grid cell in the assessment area; the marine grid cells are sorted according to the comprehensive suitability assessment scores to form the suitability assessment results before constraint removal.

[0013] Step 7: Based on the suitability assessment results output in Step 6, perform spatial constraint elimination processing. Overlay the suitability assessment results with spatial layers of Exclusive Economic Zone (EEZ), Marine Protected Area (MRA), Tropical Cyclone Frequency, Water Depth, Offshore Distance, and Sea Ice Concentration. First, retain only grid cells located within the EEA (i.e., grid cells with marine jurisdiction conditions). Then, based on this, eliminate grid cells that intersect with any of the following constraints: Marine Protected Area, Extreme Climate Risk Zone with Tropical Cyclone Frequency exceeding a preset threshold, Excessively Deep Water with Water Depth exceeding a preset threshold, Excessively Farshore Water with Offshore Distance exceeding a preset threshold, or Sea Ice Concentration exceeding a preset threshold. The preset thresholds are all configurable parameters that can be adjusted according to engineering requirements. This yields the final suitability assessment results for offshore floating photovoltaic sites.

[0014] A system for assessing the suitability of offshore floating photovoltaic sites by integrating XAI and MCDA methods, used to implement the aforementioned method for assessing the suitability of offshore floating photovoltaic sites by integrating XAI and MCDA methods, and comprising:

[0015] The data acquisition module is used to collect historical operational data, multi-source marine environmental data, and spatial constraint data;

[0016] The data preprocessing module is used to perform outlier removal, missing value handling, time alignment, spatial registration, and dimensionless processing.

[0017] The model training module is used to train multiple tree-based regression models and perform model validation.

[0018] The XAI analysis module is used to generate metric contributions and global importance based on the SHAP algorithm;

[0019] The XAI-FAHP weighting module is used to construct a fuzzy judgment matrix based on the SHAP importance ratio of multiple models and output the XAI-FAHP weights.

[0020] The CRITIC calculation module is used to evaluate the CRITIC weights output from multi-source marine environmental datasets for a region.

[0021] The game theory combinatorial weighting module is used to optimize and merge XAI-FAHP weights and CRITIC weights and output a comprehensive weight.

[0022] The TOPSIS ranking module is used to score and rank ocean grid cells based on their suitability.

[0023] The spatial constraint removal module is used to perform overlay masking processing of exclusive economic zones, marine protected areas, tropical cyclone frequency, water depth, offshore distance, and sea ice concentration after TOPSIS sorting.

[0024] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the aforementioned method for assessing the suitability of offshore floating photovoltaic sites by integrating the XAI and MCDA methods.

[0025] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned method for assessing the suitability of offshore floating photovoltaic sites by integrating the XAI and MCDA methods.

[0026] Compared with the prior art, the present invention has the following significant advantages:

[0027] (1) The present invention uses the SHAP interpretation results to construct XAI-FAHP weights, which transforms the nonlinear physical response in the real running data into interpretable index importance, thereby reducing the influence of subjective human weighting.

[0028] (2) The present invention uses the CRITIC method to replace the information entropy weighting method based only on the discreteness, so that the objective weights can simultaneously reflect the comparative strength of indicators and the degree of conflict between indicators, and are applicable to marine environmental indicators such as wind speed, wave height, temperature and radiation that are coupled with each other.

[0029] (3) The present invention uses game theory combination weighting to replace the preset ratio of linear combination weighting, so that the fusion coefficient between XAI-FAHP weight and CRITIC weight is determined by the optimization model, and the combination weighting is performed independently for each month to obtain the monthly comprehensive weight, which enhances the robustness and repeatability of the comprehensive weight.

[0030] (4) This invention sets restrictive factors such as exclusive economic zones, marine protected areas, tropical cyclone frequency, water depth, offshore distance, and sea ice concentration after the TOPSIS ranking and removes them uniformly. This not only retains the suitability ranking results before constraints but also outputs final results oriented towards actual deployment. At the same time, the thresholds can all be preset according to engineering requirements, which improves the adaptability of the method. Attached Figure Description

[0031] Figure 1 This is a flowchart of the offshore floating photovoltaic site suitability assessment method that integrates XAI and MCDA methods according to the present invention.

[0032] Figure 2 This is a schematic diagram illustrating the SHAP contribution of each input index in a tree-based regression model.

[0033] Figure 3 This is a schematic diagram illustrating the monthly scale changes of the XAI-FAHP weights, CRITIC weights, and game theory combined weights. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of this application clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the embodiments described are only for explaining the present invention and are not intended to limit the scope of protection of the present invention.

[0035] This invention discloses a method, system, computer equipment, and storage medium for assessing the suitability of offshore floating photovoltaic (PV) sites by integrating interpretable artificial intelligence (XAI) and multi-criteria decision analysis (MCDA). In this invention, the frontal irradiance used in historical operational data and the surface-downward solar radiation used in regional marine environmental data are both solar radiation indicators. They can be converted to each other using an inclined surface radiation conversion model, and their impact on PV system power generation is consistent. Similarly, ambient temperature and sea surface temperature are both temperature indicators, showing a strong correlation in nearshore areas and can be used as substitute indicators for large-scale suitability assessments. For simplicity, this invention uses "solar radiation indicator" and "temperature indicator" to refer to the aforementioned physical quantities in each step.

[0036] like Figure 1 As shown, a method for assessing the suitability of offshore floating photovoltaic sites by integrating XAI and MCDA methods includes the following steps:

[0037] Step 1: Collect historical operational datasets of offshore floating photovoltaic systems. The historical operational dataset includes solar radiation indices, temperature indices, significant wave height, wind speed at 10 meters altitude, and DC output power; [The last sentence appears to be incomplete and possibly refers to a separate topic:] ...the historical operational dataset... Preprocessing may include removing outliers, missing values, invalid data from the previous night, and extreme outliers; aligning the time dimension; and normalizing or standardizing the data to obtain the training set. and test set ;

[0038] The historical operational dataset comes from offshore floating photovoltaic demonstration platforms, on-site monitoring systems, or climate observation stations; the historical operational dataset... Preprocessing is performed, aligning data dimensions and normalizing or standardizing each metric using Z-score or Min-Max methods. Finally, the data is randomly split into training sets according to a preset ratio. and test set .

[0039] Step 2: Construct multiple tree-based regression models. Each tree-based regression model uses DC output power as the output index and solar radiation, temperature, significant wave height, and wind speed at 10 meters as input indexes. Train the multiple tree-based regression models using the training set and verify their performance using the test set. Calculate the contribution of each input index to DC output power using the SHAP interpretation method for each trained model. Further calculate the global importance of each input index and calculate the relative importance ratio between each input index based on the global importance.

[0040] The multiple tree-based regression models include at least two of the following: random forest, extreme gradient boosting, and lightweight gradient boosting machine. During model training, one or more of the following methods are used for hyperparameter optimization: cross-validation, Bayesian optimization, grid search, and early stopping strategy. At least one of the following metrics is used to verify model performance: coefficient of determination, root mean square error, mean absolute error, or mean absolute percentage error. The model can only proceed to subsequent SHAP analysis when the performance metrics meet the preset threshold.

[0041] The sample-level Shapley values ​​are calculated using TreeExplainer or the equivalent SHAP algorithm, and the absolute Shapley values ​​of each sample in the test set are averaged to obtain the global importance of each input indicator in any model. Then, based on the global importance in different models, the relative importance ratio between any two input indicators is calculated.

[0042] The formula for calculating SHAP contribution value is as follows:

[0043] ;

[0044] Indicators The corresponding Shapley value, which is the average marginal contribution of this indicator to the model's prediction results; This represents the model value function determined by a subset of input indicators; This represents the set of all indicators involved in the model's prediction. This represents a single indicator to be explained, and ; This represents the total number of indicators, and ; express The content does not include indicators. Any subset of indicators; Representing a subset The number of indicators in; , and This is a factorial term used to weight the marginal contributions of different indices in different order of addition; Indicates a subset of existing indicators Add indicators on the basis The increment of the model output;

[0045] The formula for calculating global importance is as follows:

[0046] .

[0047] Indicators The average absolute global importance of SHAP; This represents the total number of test samples used for interpreting the analysis; Indicates the sample number. ; Indicates the first Indicators at each sample Local SHAP values; This indicates that the absolute value of the local SHAP value is taken to eliminate the cancellation of positive and negative directions; This represents the summation over all test samples; This indicates that the average value is calculated over all samples.

[0048] Step 3: Construct an XAI-enhanced fuzzy hierarchical analysis judgment matrix based on the relative importance ratios obtained in Step 2; map the minimum, average, and maximum values ​​of the relative importance ratios of the same indicator pair in multiple models to triangular fuzzy numbers to form a fuzzy judgment matrix; calculate the fuzzy weights of each indicator using the fuzzy geometric mean method and perform normalization processing, and then use the centroid method to defuzzify the fuzzy weights to obtain the XAI-FAHP weights.

[0049] The XAI-FAHP weights are constructed entirely from the importance ratios of the multi-model SHAP, without introducing any manually assigned weighting parameters. Before the weights are calculated, a consistency check is performed using a triangular fuzzy number median matrix, which is composed of the median of each triangular fuzzy number. When the consistency ratio does not meet the preset threshold, the fuzzy judgment matrix is ​​corrected to ensure that the consistency meets the requirements.

[0050] The formula for calculating the relative importance ratio is as follows:

[0051] ;

[0052] Indicates the first Indicators under a tree-based regression model relative to indicators Importance ratio; Indicates the first Indicators in each model The average absolute global importance of SHAP; Indicates the first Indicators in each model The average absolute global importance of SHAP; This is a smoothing factor used to prevent numerical instability caused by the denominator approaching zero; a value of [value missing] is preferred. ; Indicates the number of models participating in the integration; Indicates the model number. ;

[0053] The formula for constructing triangular fuzzy numbers is as follows:

[0054] ;

[0055] Indicators relative to indicators Triangular fuzzy numbers; , and These represent the lower bound, most likely value, and upper bound of the triangular fuzzy number, respectively.

[0056] The lower bound, most likely value, and upper bound are determined as follows:

[0057] ;

[0058] Represents the relative importance ratio under multiple models The minimum value; Represents the relative importance ratio under multiple models The average value; Represents the relative importance ratio under multiple models The maximum value; , and These represent the numbering of all models. The corresponding data includes the minimum, average, and maximum values;

[0059] The triangular fuzzy number for reverse comparison is determined as follows:

[0060] ;

[0061] Indicators relative to indicators The reverse triangular fuzzy number; , and They represent respectively to The inverse comparison value is obtained by taking the reciprocals of the upper bound, most likely value, and lower bound; by swapping the order of the upper and lower bounds, the inverse triangular fuzzy number remains valid;

[0062] The formula for calculating the consistency ratio is as follows:

[0063] ;

[0064] Represents the consistency ratio, used to verify the logical consistency of the value matrix in the fuzzy judgment matrix; This represents the largest eigenvalue of the median judgment matrix; Indicates the total number of indicators; Indicators and their quantities The corresponding random consistency index; when When the value is less than a preset threshold, the judgment matrix satisfies the consistency requirement;

[0065] The formula for calculating the fuzzy geometric mean is as follows:

[0066] ;

[0067] Indicates the first The fuzzy geometric mean corresponding to each indicator; Represents the fuzzy judgment matrix in which the first... Line 1 The triangular fuzzy judgment number of the column; Indicates the first Multiply all triangular fuzzy judgment numbers together; Indicates the total number of indicators; Indicates to proceed Root operation; the above multiplication and root extraction are calculated separately for the three components of the triangular fuzzy number;

[0068] The formula for calculating fuzzy weights is as follows:

[0069] ;

[0070] Indicates the first Unresolved fuzzy weights of each indicator; Indicates the first The fuzzy geometric mean corresponding to each indicator; Indicates the first The fuzzy geometric mean corresponding to each indicator; This represents triangular fuzzy number multiplication; This represents triangular fuzzy number addition; The fuzzy sum representing the fuzzy geometric mean of all indicators; This represents the inverse triangular fuzzy number of the fuzzy sum;

[0071] The calculation formula for fuzzy resolution using the centroid method is as follows:

[0072] ;

[0073] Indicates the first Clear weights of each indicator after defuzzification using the centroid method; , and They represent fuzzy weights respectively. The lower bound, median, and upper bound of the denominator; This means taking the arithmetic mean of the three components of the triangular blur number to obtain the sharpness value;

[0074] The formula for calculating the XAI-FAHP weight normalization is as follows:

[0075] .

[0076] This represents the normalized XAI-FAHP weights; Indicates the first Clear weighting of each indicator; Indicates the first Clear weighting of each indicator; This represents the sum of the clear weights of all indicators; This represents the total number of indicators; this normalization process ensures that the sum of all XAI-FAHP weights is 1.

[0077] Step 4: Collect multi-source marine environmental datasets for marine grid cells within the assessment area. and spatially constrained datasets The multi-source marine environment dataset The spatially constrained dataset includes solar radiation indices, temperature indices, significant wave height, and wind speed at a height of 10 meters. This includes exclusive economic zones, marine protected areas, tropical cyclone frequency, water depth, offshore distance, and sea ice concentration; and the collection of multi-source marine environmental datasets for the assessment area. Monthly-scale statistics, spatial registration, and dimensionless processing were performed. The dynamic objective weights of each indicator were calculated using the CRITIC method. The CRITIC method simultaneously uses the standard deviation of the indicators to characterize the contrast intensity and the correlation coefficient between the indicators to characterize the degree of conflict, thereby obtaining the CRITIC weights based on monthly-scale data.

[0078] The multi-source marine environment dataset Sources include satellite remote sensing products, global reanalysis databases, and ocean reanalysis databases; different data sources are unified to a preset spatial resolution and temporal scale before being incorporated into the CRITIC and TOPSIS methods;

[0079] Let the dimensionless monthly scale evaluation matrix be... ,in Represents ocean grid cells, Indicators; for any indicator Calculate the standard deviation And calculate the correlation coefficient between this indicator and other indicators. ;Will As an indicator The information content is processed and the CRITIC weights are obtained through normalization.

[0080] The formula for calculating the degree of indicator conflict is as follows:

[0081] ;

[0082] Indicates the first The degree of overall conflict between this indicator and the rest of the indicators; Indicates the first The first indicator and the first Correlation coefficients between the indicators; This indicates the degree of conflict or difference between two indicators; Indicates the total number of indicators; This represents the summation of all indicators;

[0083] The formula for calculating the information content of an indicator is as follows:

[0084] ;

[0085] Indicates the first The amount of information contained in each indicator; Indicates the first The standard deviation of an indicator across all ocean grid cells, i.e., the contrast intensity; Indicates the first The overall degree of conflict among the indicators; Indicates the first The first indicator and the first Correlation coefficients between the indicators; Indicates the first The degree of conflict between an individual indicator and all indicators; It is determined by both the intensity of the contrast and the degree of conflict;

[0086] The formula for calculating the CRITIC objective weight is as follows:

[0087] .

[0088] Indicates the first CRITIC objective weights for each indicator; Indicates the first The amount of information contained in each indicator; Indicates the first The amount of information contained in each indicator; This represents the sum of information content from all indicators; This represents the total number of indicators; this normalization process ensures that the sum of all CRITIC weights is 1.

[0089] Step 5: The XAI-FAHP weights obtained in Step 3 and the CRITIC weights obtained in Step 4 are optimized and fused using the game theory combination weighting method. An optimization model is established to minimize the deviation between the combined weight vector and each basic weight vector. The combination coefficients are solved and normalized to obtain the comprehensive weight.

[0090] Let the XAI-FAHP weight vector be... The CRITIC weight vector is The comprehensive weight vector is obtained by game theory optimization. By constructing an optimization problem that minimizes the total deviation between the combined weight and each basic weight vector, the combination coefficients corresponding to the basic weight vectors are solved, and the combination coefficients are normalized to obtain the comprehensive weight vector used for TOPSIS calculation.

[0091] The expression for the combined weights is as follows:

[0092] ;

[0093] This represents a combined weight vector composed of the basic weight vectors. Indicates the first There are 1 basic weight vector, where The XAI-FAHP weight vector, This is the CRITIC weight vector; This represents the transpose of the corresponding weight vector; Indicates the first The unnormalized combination coefficients of the basic weight vectors; This invention employs two types of basic weights;

[0094] The optimization objective for the combination coefficients is as follows:

[0095] ;

[0096] This indicates that the optimization objective is to minimize the objective function. Represents the Euclidean norm; This represents a combined weight vector synthesized from two basic weight vectors; Indicates the first Transpose of the basic weight vectors; Indicates the first The unnormalized combination coefficients of the basic weight vectors; This represents minimizing the deviation between the combined weights and the XAI-FAHP weights and CRITIC weights, respectively.

[0097] The system of linear equations for solving the combination coefficients is as follows:

[0098] ;

[0099] Represents the XAI-FAHP weight vector; Represents the CRITIC weight vector; and These represent the transposes of the corresponding weight vectors; Represents the weight vector and The inner product between; and These represent the unnormalized combination coefficients corresponding to the XAI-FAHP weight vector and the CRITIC weight vector, respectively; the vector on the right-hand side represents the constant term formed by the inner product of each basic weight vector.

[0100] The formula for normalizing the combination coefficients is as follows:

[0101] ;

[0102] Represents the normalized i-th Combination coefficients; Indicates the first Unnormalized combination coefficients; express The absolute value; This represents the sum of the absolute values ​​of the coefficients of two unnormalized combinations;

[0103] The final overall weight calculation formula is as follows:

[0104] ;

[0105] This represents the final integrated weight vector used in subsequent TOPSIS calculations; This represents the normalized combination coefficients corresponding to the XAI-FAHP weight vector; This represents the normalized combination coefficients corresponding to the CRITIC weight vector; Represents the XAI-FAHP weight vector; This represents the CRITIC weight vector.

[0106] Step 6: Based on the multi-source marine environment dataset processed in Step 4 and the comprehensive weights obtained in Step 5, the TOPSIS (Topological Solution Approximation) method is used to calculate the comprehensive suitability assessment score of each marine grid cell in the assessment area; the marine grid cells are sorted according to the comprehensive suitability assessment scores to form the suitability assessment results before constraint removal.

[0107] The attribute direction of each indicator is determined based on SHAP dependency or preset physical rules. Specifically, the attribute direction is set according to the average marginal contribution direction of each indicator to the output power or physical common sense. The attribute direction includes benefit-type or cost-type. In TOPSIS calculation, the evaluation matrix is ​​positiveized and standardized according to the indicator attributes. Cost-type indicators are converted into benefit-type indicators. The standardization adopts the range normalization method to construct a weighted standardized decision matrix under comprehensive weight constraints. The distance from each ocean grid cell to the positive ideal solution and the negative ideal solution is calculated, and the relative proximity score is obtained accordingly.

[0108] The calculation formulas for positive transformation and standardization of profitability indicators are as follows:

[0109] ;

[0110] The calculation formulas for positiveizing and standardizing cost indicators are as follows:

[0111] ;

[0112] in Indicates the first The ocean grid cell in the first The original values ​​of each indicator; This represents the value after normalization and orientation. and They represent the first The maximum and minimum values ​​of each indicator in all ocean grid cells;

[0113] The formula for calculating the elements of the weighted normalized matrix is ​​as follows:

[0114] ;

[0115] In the weighted standardized decision matrix, the first... The first marine grid unit, the The elements corresponding to each indicator; The first game obtained by game theory combinatorial weighting is... The final weight of each indicator; This represents the first [unit] after forwarding and standardization. The grid cell in the first... The values ​​of each indicator; Indicates the ocean grid cell number; Indicates the indicator number;

[0116] The formula for calculating the ideal solution is as follows:

[0117] ;

[0118] Represents the vector of the positive ideal solution; , to These represent the nth element in all ocean grid cells. , No. To the The maximum weighted standardized value of each indicator; Represents the elements of a weighted standardized decision matrix; Indicates the ocean grid cell number; Indicates the total number of indicators;

[0119] The formula for calculating the negative ideal solution is as follows:

[0120] ;

[0121] Represents the negative ideal solution vector; , to These represent the nth cells in all ocean grid cells. , No. To the The least weighted standardized value of each indicator; Represents the elements of a weighted standardized decision matrix; Indicates the ocean grid cell number; Indicates the total number of indicators;

[0122] The formula for calculating the distance to the ideal solution is as follows:

[0123] ;

[0124] Indicates the first From ocean grid cells to the ideal solution The Euclidean distance; Indicates the first The grid cell in the first... Weighted standardized values ​​for each indicator; In the positive ideal solution, the first The values ​​of each indicator; This represents the summation of all indicators; Indicates the total number of indicators; Represents the square root operation;

[0125] The formula for calculating the distance to the negative ideal solution is as follows:

[0126] ;

[0127] Indicates the first From a single ocean grid cell to a negative ideal solution The Euclidean distance; Indicates the first The grid cell in the first... Weighted standardized values ​​for each indicator; In the negative ideal solution, the first... The values ​​of each indicator; This represents the summation of all indicators; Indicates the total number of indicators; Represents the square root operation;

[0128] The formula for calculating the relative closeness score is as follows:

[0129] .

[0130] Indicates the first The relative proximity score of each ocean grid cell; Indicates the first The Euclidean distance from each grid cell to the positive ideal solution; Indicates the first The Euclidean distance from each grid cell to the negative ideal solution; The range of values ​​is to The larger the value, the closer it is to the ideal state and the further it is from the negative ideal state, and the higher the suitability of offshore floating photovoltaic deployment.

[0131] Step 7: Based on the suitability assessment results output in Step 6, a spatial constraint elimination process is finally performed. The suitability assessment results are overlaid with spatial layers such as Exclusive Economic Zone (EEZ), Marine Protected Area (MRA), Tropical Cyclone Frequency, Water Depth, Offshore Distance, and Sea Ice Concentration. First, only grid cells located within the EEZ (i.e., grid cells with marine jurisdiction conditions) are retained. Then, based on this, grid cells intersecting with any of the following constraints are eliminated: Marine Protected Area, Extreme Climate Risk Zone with Tropical Cyclone Frequency exceeding a preset threshold, Excessively Deep Water with Water Depth exceeding a preset threshold, Excessively Farshore Water with Offshore Distance exceeding a preset threshold, or Sea Ice Concentration exceeding a preset threshold. The preset thresholds are all configurable parameters that can be adjusted according to engineering requirements. The final suitability assessment result for the offshore floating photovoltaic site is obtained.

[0132] This invention also proposes a marine floating photovoltaic site suitability assessment system based on XAI and MCDA, used to implement the aforementioned marine floating photovoltaic site suitability assessment method integrating XAI and MCDA methods, including:

[0133] The data acquisition module is used to collect historical operational data, multi-source marine environmental data, and spatial constraint data;

[0134] The data preprocessing module is used to perform outlier removal, missing value handling, time alignment, spatial registration, and dimensionless processing.

[0135] The model training module is used to train multiple tree-based regression models and perform model validation.

[0136] The XAI analysis module is used to generate metric contributions and global importance based on the SHAP algorithm;

[0137] The XAI-FAHP weighting module is used to construct a fuzzy judgment matrix based on the SHAP importance ratio of multiple models and output the XAI-FAHP weights.

[0138] The CRITIC calculation module is used to evaluate the CRITIC weights output from multi-source marine environmental datasets for a region.

[0139] The game theory combinatorial weighting module is used to optimize and merge XAI-FAHP weights and CRITIC weights and output a comprehensive weight.

[0140] The TOPSIS ranking module is used to score and rank ocean grid cells based on their suitability.

[0141] The spatial constraint removal module is used to perform overlay masking processing of exclusive economic zones, marine protected areas, tropical cyclone frequency, water depth, offshore distance, and sea ice concentration after TOPSIS sorting.

[0142] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the offshore floating photovoltaic site suitability assessment method that integrates XAI and MCDA methods.

[0143] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned method for assessing the suitability of offshore floating photovoltaic sites by integrating the XAI and MCDA methods.

[0144] Example

[0145] To verify the effectiveness of the present invention, the present invention will be further described below with reference to specific embodiments, using historical operation data of real offshore floating photovoltaic pilot projects to assess the suitability of offshore floating photovoltaic in global sea areas.

[0146] It should be understood that this embodiment is only used to illustrate the technical solution of the present invention and is not intended to limit the scope of protection of the present invention. Without departing from the core idea of ​​the present invention, those skilled in the art can make adaptive adjustments to the data source, threshold conditions and calculation scale according to the actual sea area range, data availability, engineering constraints and spatial resolution requirements.

[0147] 1) Historical operational data collection and preprocessing

[0148] This embodiment uses a near-shore semi-submersible floating photovoltaic demonstration platform as the source of measured data. The platform is located approximately 4 km offshore, in water depths of approximately 9 to 11 meters, with an installed capacity of approximately 400 kW, and employs a combined floating body and truss structure. The on-site monitoring system collects data on solar radiation (specifically, frontal irradiance), temperature (specifically, ambient temperature), wind speed at a height of 10 meters, and DC output power, with a sampling interval of 30 seconds. Significant wave height data is supplemented from ERA5 reanalysis data and time-matched with the measured data from the demonstration platform.

[0149] Historical operational data underwent quality control to remove sensor outliers, missing values, invalid nighttime samples, and extreme outliers. Subsequently, the field monitoring data was time-aligned with the reanalysis wave data to form a complete sample set including solar radiation, temperature, 10-meter wind speed, significant wave height, and DC output power. Input indicators were Z-score standardized, while DC output power retained its original physical dimensions. The cleaned data was divided into training and testing sets using a fixed random seed. The training set was used for model fitting, and the testing set was used for model validation and SHAP interpretation analysis.

[0150] 2) Training and performance validation of tree-based regression models

[0151] Using frontal irradiance, ambient temperature, wind speed at 10 meters, and significant wave height as input metrics, and DC output power as the output metric, random forest, extreme gradient boosting (XGBoost), and lightweight gradient boosting machine (LightGBM) tree-based regression models were constructed, respectively. Parallel training of multiple models was used to reduce the influence of a single model structure on the metric contribution identification results and to improve the robustness of subsequent weight construction.

[0152] Model training employed five-fold cross-validation. Hyperparameter optimization combined Bayesian optimization with grid search; XGBoost and LightGBM models introduced early stopping mechanisms to reduce overfitting risk. After training, performance was evaluated using the coefficient of determination, root mean square error, mean absolute error, and mean absolute percentage error. Models meeting prediction performance requirements proceeded to subsequent SHAP interpretation analysis.

[0153] 3) SHAP-based indicator contribution identification

[0154] SHAP interpretation analysis was performed on the trained Random Forest, XGBoost, and LightGBM models. The SHAP method decomposes the model prediction results into the contribution values ​​of each input index, which is used to identify the degree of influence of frontal irradiance, ambient temperature, 10-meter wind speed, and significant wave height on DC output power.

[0155] For each test sample, a local SHAP value is calculated, and the absolute SHAP values ​​of all test samples are averaged to obtain the global importance of each input indicator under the corresponding model. The indicator importance results obtained from different tree-based regression models are used together to construct the subsequent XAI-FAHP judgment matrix. The differences in importance between models, as a source of uncertainty, are expressed using triangular fuzzy numbers.

[0156] SHAP dependencies are used to determine the orientation of evaluation indicator attributes. Solar radiation is used as a benefit indicator, ambient temperature as a cost indicator, and 10-meter wind speed and significant wave height are set as cost indicators based on their impact on DC output power and maritime operational risks. This processing provides a data interpretation basis and physical evidence for the indicator positiveization process in TOPSIS.

[0157] 4) Construction of XAI-FAHP basic weights

[0158] Based on the SHAP global importance results obtained from three tree-based regression models, the relative importance ratio between any two evaluation indicators is calculated. For the same indicator pair, the relative importance ratios obtained from Random Forest, XGBoost, and LightGBM are calculated respectively, and their minimum, average, and maximum values ​​are taken as the lower bound, most likely value, and upper bound of the triangular fuzzy number to construct the XAI-enhanced fuzzy judgment matrix.

[0159] Consistency checks were performed using a triangular fuzzy number median matrix (composed of the median of each triangular fuzzy number). After meeting the consistency requirements, the fuzzy weights of each evaluation index were calculated using the fuzzy geometric mean method, and then the XAI-FAHP weights were obtained after defuzzification and normalization. These weights were derived from the measured operating data of the demonstration platform and the model interpretation results, without introducing manual scoring or manual weighting parameters.

[0160] 5) Processing of multi-source marine environmental data and spatially constrained data

[0161] This embodiment constructs a global monthly-scale multi-source marine environmental dataset from 2001 to 2020. Downward solar radiation, sea surface temperature, and wind speed at 10 meters altitude were obtained using ERA5 reanalysis data, while significant wave height was obtained from Copernicus MarineService ocean data. Each indicator was first aggregated monthly, and then the average value for the same month over many years was calculated to characterize long-term stable marine environmental indicators.

[0162] Spatial constraint data includes Exclusive Economic Zones (EEZs), Marine Protected Areas (MLAs), tropical cyclone frequency, water depth, offshore distance, and sea ice concentration. EEZs define the marine areas suitable for development and management; MLAs identify ecologically sensitive areas; tropical cyclone frequency characterizes extreme weather risks; water depth characterizes the feasibility of mooring, anchoring, and submarine cable laying; offshore distance characterizes operational accessibility and grid connection difficulty; and sea ice concentration characterizes the impact of freezing in high-latitude seas.

[0163] Data from different sources were standardized to a spatial resolution of 0.25° × 0.25°. Continuous indices were processed using spatial interpolation or resampling, while boundary and Boolean constraint layers were processed using nearest neighbor resampling to avoid smoothing out the constraint boundaries. After spatial registration, the land area was masked, retaining only the marine grid cells for subsequent evaluation processes.

[0164] 6) CRITIC Dynamic Objective Weight Calculation

[0165] The XAI-FAHP weights reflect the fundamental physical response relationships in the demonstration platform's operational data. To further characterize the spatial differences and seasonal variations in global ocean areas, this embodiment uses the CRITIC method to dynamically weight monthly marine environmental data.

[0166] After dimensionless processing of the marine environmental assessment matrix for each month, the standard deviation of each assessment indicator across all marine grid cells is calculated to characterize the spatial contrast intensity. Simultaneously, the correlation between assessment indicators is calculated to characterize the degree of conflict between them. Indicators with greater spatial differences and lower repeatability with other indicators receive higher CRITIC weights.

[0167] The CRITIC weights are calculated monthly, reflecting the differences in solar radiation, sea surface temperature, wind speed, and significant wave height across different months and sea areas. This step complements the static characteristics of the XAI-FAHP weights, enabling the final evaluation system to reflect both measured operational response and the heterogeneity of the global marine environment.

[0168] 7) Game theory combinatorial weighting

[0169] XAI-FAHP weights and CRITIC weights are used as two basic weights and fused using a game-theoretic combined weighting method. XAI-FAHP weights reflect the physical response patterns in actual operational data, while CRITIC weights reflect spatial differences and index conflicts in multi-source marine environmental data.

[0170] Game theory-based combinatorial weighting automatically calculates the combinatorial coefficients by minimizing the deviation between the combinatorial weights and the two basic weights. This method does not pre-determine a fixed linear weighting ratio and can adaptively obtain the comprehensive weights based on the relationship between the two weight vectors. These comprehensive weights are then used for subsequent TOPSIS suitability ranking.

[0171] 8) TOPSIS Suitability Ranking

[0172] Based on global monthly marine environmental data and game theory-based weighted indexes, TOPSIS fitness is calculated for marine grid cells. First, each evaluation index is positiveized and standardized according to its attribute direction. Cost-based indicators are converted into benefit-based indicators. Standardization employs range normalization to transform indicators with different dimensions and directions into a unified evaluation scale.

[0173] Subsequently, a weighted standardized decision matrix was constructed, and positive and negative ideal solutions were determined. A positive ideal solution represents a hypothetical grid where all evaluation indicators are in their optimal state, while a negative ideal solution represents a hypothetical grid where all evaluation indicators are in their most unfavorable state. A relative proximity score was obtained by calculating the distance between each marine grid cell and the positive and negative ideal solutions. A higher score indicates that the grid cell is more suitable as a candidate site for offshore floating photovoltaic systems.

[0174] After all grid cells have been calculated, they are sorted in descending order based on their relative proximity scores, and then classified into suitability levels according to quantiles or preset thresholds, forming the suitability evaluation results before spatial constraint removal.

[0175] 9) Spatial constraints are finally eliminated

[0176] After TOPSIS ranking is completed, spatial constraint removal is performed. This step is independent of the weight calculation and suitability ranking process and serves as a final spatial mask to ensure that the evaluation results meet legal, ecological, and engineering feasibility requirements.

[0177] First, the exclusive economic zone layer is used to screen grid cells that meet the conditions for maritime jurisdiction. Then, a marine protected area layer is overlaid, and grid cells that intersect with marine protected areas are removed. Further layers of tropical cyclone frequency, water depth, offshore distance, and sea ice concentration are overlaid to remove grid cells with excessively high tropical cyclone risk, water depth exceeding the applicable range of the project, offshore distance exceeding the access and operation and maintenance capabilities, or significant impact from sea ice.

[0178] In this embodiment, grid cells with a water depth exceeding 300 m, a distance from shore exceeding 200 km, a sea ice concentration exceeding a preset threshold of 0, or a tropical cyclone frequency exceeding a preset threshold of 0 are designated as restricted areas. These preset thresholds can be adjusted according to engineering requirements. All spatial constraints are processed using a parallel Boolean masking method; that is, any grid cell that violates any hard constraint is removed from the final candidate areas. The remaining grid cells after removal constitute the final suitability assessment result for offshore floating photovoltaic sites.

[0179] 10) Results Analysis and Output

[0180] The output of this embodiment includes the SHAP index contribution result, the weight result of each stage, the fitness partition result before constraint removal, and the final fitness partition result after constraint removal.

[0181] like Figure 2 As shown, the Shapley contribution results of the trained example model demonstrate the direction and intensity of the contributions of frontal irradiance, ambient temperature, wind speed at 10 meters height, and significant wave height to DC output power, and provide a basis for the construction of XAI-FAHP weights and the determination of TOPSIS index attributes.

[0182] like Figure 3 As shown, the monthly weight calculation results for each assessment stage demonstrate the differences and integration relationships between the XAI-FAHP weight, CRITIC weight, and game theory combined weight, illustrating that this invention simultaneously considers the measured operational response and the spatiotemporal heterogeneity of the marine environment.

[0183] This embodiment demonstrates that the present invention can utilize limited measured operational data from floating marine photovoltaic (PV) systems, combined with multi-source marine environmental data and spatial constraint factors, to achieve an objective, interpretable, and scalable assessment of the suitability of floating PV sites. This method is applicable to macro-level site selection analysis of floating PV systems at global, regional, or specific sea area scales.

[0184] The technical features of the above embodiments can be combined in any way. For the sake of brevity, 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.

[0185] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these modifications and improvements all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for assessing the suitability of offshore floating photovoltaic sites by integrating XAI and MCDA methods, characterized in that, include: Step 1: Collect historical operational data of the offshore floating photovoltaic system, including solar radiation index, temperature index, significant wave height, wind speed at 10 meters height, and DC output power; Preprocess the historical dataset to obtain the training set and the test set; Step 2: Construct multiple tree-based regression models with DC output power as the output. After training and validation, use the SHAP interpretation method to calculate the global importance of each input indicator and calculate the relative importance ratio. Step 3: Construct a fuzzy hierarchical analysis judgment matrix based on the relative importance ratio. After mapping it to a fuzzy judgment matrix, use the fuzzy geometric mean method and centroid method to defuzzify and obtain the XAI-FAHP weights. Step 4: Collect multi-source marine environmental datasets and spatially constrained datasets for marine grid cells within the evaluation area. The multi-source marine environmental datasets include solar radiation indices, temperature indices, significant wave height, and wind speed at 10 meters above sea level. The spatially constrained datasets include exclusive economic zones, marine protected areas, tropical cyclone frequency, water depth, offshore distance, and sea ice concentration. Perform monthly-scale statistics, spatial registration, and dimensionless processing on the multi-source marine environmental datasets, and use the CRITIC method to obtain CRITIC weights. Step 5: Use a game theory-based combined weighting method to fuse the XAI-FAHP weights and CRITIC weights to obtain the comprehensive weights; Step 6: Based on the multi-source marine environment dataset and the comprehensive weight, the TOPSIS (Topological Solution Approximation) method is used to calculate and sort the comprehensive suitability assessment score of each marine grid cell to obtain the assessment results before constraint removal. Step 7: Overlay a spatial constraint layer on the results of Step 6. First, retain the grid cells located within the exclusive economic zone, and then remove the grid cells that intersect with marine protected areas, tropical cyclones exceeding the preset threshold, water depth exceeding the preset threshold, offshore distance exceeding the preset threshold, or sea ice concentration exceeding the preset threshold to obtain the final evaluation results.

2. The method for assessing the suitability of offshore floating photovoltaic sites by integrating XAI and MCDA methods according to claim 1, characterized in that, In step 1, historical operational datasets of offshore floating photovoltaic systems are collected. The historical operational dataset includes solar radiation indices, temperature indices, significant wave height, wind speed at 10 meters altitude, and DC output power; [The last sentence appears to be incomplete and possibly refers to a separate topic:] ...the historical operational dataset... Preprocessing is performed, including removing outliers, missing values, invalid nighttime data, and extreme outliers; aligning the time dimension; and normalizing or standardizing the data to obtain the training set. and test set Specifically: The historical operational dataset comes from offshore floating photovoltaic demonstration platforms, on-site monitoring systems, or climate observation stations; the historical operational dataset... Preprocessing is performed, aligning data dimensions and normalizing or standardizing each metric using Z-score or Min-Max methods. Finally, the data is randomly split into training sets according to a preset ratio. and test set .

3. The method for assessing the suitability of offshore floating photovoltaic sites by integrating XAI and MCDA methods according to claim 1, characterized in that, In step 2, multiple tree-based regression models are constructed. Each model uses DC output power as the output index and solar radiation, temperature, significant wave height, and wind speed at 10 meters as input indices. The models are trained using a training set and their performance is validated using a test set. The SHAP interpretation method is used to calculate the contribution of each input index to DC output power for each trained model. The global importance of each input index is then calculated, and the relative importance ratios between the input indices are calculated based on their global importance. Specifically: The multiple tree-based regression models include at least two of random forest, extreme gradient boosting, and lightweight gradient boosting machine; during model training, one or more of the following methods are used for hyperparameter optimization: cross-validation, Bayesian optimization, grid search, and early stopping strategy; and at least one of the following indicators is used for model performance verification: coefficient of determination, root mean square error, mean absolute error, or mean absolute percentage error. The model can only proceed to subsequent SHAP analysis when the performance indicator meets the preset threshold. The sample-level Shapley values ​​are calculated using TreeExplainer or the equivalent SHAP algorithm, and the absolute Shapley values ​​of each sample in the test set are averaged to obtain the global importance of each input indicator in any model. Then, based on the global importance in different models, the relative importance ratio between any two input indicators is calculated. The formula for calculating SHAP contribution value is as follows: ； in Indicators The corresponding Shapley value, which is the average marginal contribution of this indicator to the model's prediction results; This represents the model value function determined by a subset of input indicators; This represents the set of all indicators involved in the model's prediction. This represents a single indicator to be explained, and ; This represents the total number of indicators, and ; express The content does not include indicators. Any subset of indicators; Representing a subset The number of indicators in; , and This is a factorial term used to weight the marginal contributions of different indices in different order of addition; Indicates a subset of existing indicators Add indicators on the basis The increment of the model output; The formula for calculating global importance is as follows: ； in Indicators The average absolute global importance of SHAP; This represents the total number of test samples used for interpreting the analysis; Indicates the sample number. ; Indicates the first Indicators at each sample Local SHAP values; This indicates that the absolute value of the local SHAP value is taken to eliminate the cancellation of positive and negative directions; This represents the summation over all test samples; This indicates that the average value is calculated over all samples.

4. The method for assessing the suitability of offshore floating photovoltaic sites by integrating XAI and MCDA methods according to claim 1, characterized in that, In step 3, an XAI-enhanced fuzzy hierarchical analysis (FAHP) judgment matrix is ​​constructed based on the relative importance ratios obtained in step 2. The minimum, average, and maximum values ​​of the relative importance ratios of the same indicator pair in multiple models are mapped to triangular fuzzy numbers to form a fuzzy judgment matrix. The fuzzy weights of each indicator are calculated using the fuzzy geometric mean method and normalized. Then, the centroid method is used to defuzzify the fuzzy weights to obtain the XAI-FAHP weights, specifically: The XAI-FAHP weights are constructed entirely from the importance ratios of the multi-model SHAP, without introducing any manually assigned weighting parameters. Before the weights are calculated, a consistency check is performed using a triangular fuzzy number median matrix, which is composed of the median of each triangular fuzzy number. When the consistency ratio does not meet the preset threshold, the fuzzy judgment matrix is ​​corrected to ensure that the consistency meets the requirements. The formula for calculating the relative importance ratio is as follows: ； in Indicates the first Indicators under a tree-based regression model relative to indicators Importance ratio; Indicates the first Indicators in each model The average absolute global importance of SHAP; Indicates the first Indicators in each model The average absolute global importance of SHAP; This is a smoothing factor used to prevent numerical instability caused by the denominator approaching zero. ; Indicates the number of models participating in the integration; Indicates the model number. ; The formula for constructing triangular fuzzy numbers is as follows: ； in Indicators relative to indicators Triangular fuzzy numbers; , and These represent the lower bound, most likely value, and upper bound of the triangular fuzzy number, respectively. The lower bound, most likely value, and upper bound are determined as follows: ； in Represents the relative importance ratio under multiple models The minimum value; Represents the relative importance ratio under multiple models The average value; Represents the relative importance ratio under multiple models The maximum value; , and These represent the numbering of all models. The corresponding data includes the minimum, average, and maximum values; The triangular fuzzy number for reverse comparison is determined as follows: ； in Indicators relative to indicators The reverse triangular fuzzy number; , and They represent respectively to The inverse comparison value is obtained by taking the reciprocals of the upper bound, most likely value, and lower bound; by swapping the order of the upper and lower bounds, the inverse triangular fuzzy number remains valid; The formula for calculating the consistency ratio is as follows: ； in Represents the consistency ratio, used to verify the logical consistency of the value matrix in the fuzzy judgment matrix; This represents the largest eigenvalue of the median judgment matrix; Indicates the total number of indicators; Indicators and their quantities The corresponding random consistency index; when When the value is less than a preset threshold, the judgment matrix satisfies the consistency requirement; The formula for calculating the fuzzy geometric mean is as follows: ； in Indicates the first The fuzzy geometric mean corresponding to each indicator; Represents the fuzzy judgment matrix in which the first... Line number The triangular fuzzy judgment number of the column; Indicates the first Multiply all triangular fuzzy judgment numbers together; Indicates the total number of indicators; Indicates to proceed Root operation; the above multiplication and root extraction are calculated separately for the three components of the triangular fuzzy number; The formula for calculating fuzzy weights is as follows: ； in Indicates the first Unresolved fuzzy weights of each indicator; Indicates the first The fuzzy geometric mean corresponding to each indicator; Indicates the first The fuzzy geometric mean corresponding to each indicator; This represents triangular fuzzy number multiplication; This represents triangular fuzzy number addition; The fuzzy sum representing the fuzzy geometric mean of all indicators; This represents the inverse triangular fuzzy number of the fuzzy sum; The calculation formula for fuzzy resolution using the centroid method is as follows: ； in Indicates the first Clear weights of each indicator after defuzzification using the centroid method; , and They represent fuzzy weights respectively. The lower bound, median, and upper bound of the denominator; This means taking the arithmetic mean of the three components of the triangular blur number to obtain the sharpness value; The formula for calculating the XAI-FAHP weight normalization is as follows: ； in This represents the normalized XAI-FAHP weights; Indicates the first Clear weighting of each indicator; Indicates the first Clear weighting of each indicator; This represents the sum of the clear weights of all indicators; This represents the total number of indicators; this normalization process ensures that the sum of all XAI-FAHP weights is 1.

5. The method for assessing the suitability of offshore floating photovoltaic sites by integrating XAI and MCDA methods according to claim 1, characterized in that, In step 4, multi-source marine environmental datasets of marine grid cells within the assessment area are collected. and spatially constrained datasets The multi-source marine environment dataset The spatially constrained dataset includes solar radiation indices, temperature indices, significant wave height, and wind speed at a height of 10 meters. This includes exclusive economic zones, marine protected areas, tropical cyclone frequency, water depth, offshore distance, and sea ice concentration; and the collection of multi-source marine environmental datasets for the assessment area. Monthly-scale statistics, spatial registration, and dimensionless processing were performed. The CRITIC method was used to calculate the dynamic objective weights of each indicator. The CRITIC method simultaneously uses the standard deviation of the indicators to characterize the contrast intensity and the correlation coefficient between the indicators to characterize the degree of conflict, thus obtaining the CRITIC weights based on monthly-scale data. The multi-source marine environment dataset Sources include satellite remote sensing products, global reanalysis databases, and ocean reanalysis databases; different data sources are unified to a preset spatial resolution and temporal scale before being incorporated into the CRITIC and TOPSIS methods; Let the dimensionless monthly scale evaluation matrix be... ,in Represents ocean grid cells, Indicators; for any indicator Calculate the standard deviation And calculate the correlation coefficient between this indicator and other indicators. ;Will As an indicator The information content is processed and the CRITIC weights are obtained through normalization. The formula for calculating the degree of indicator conflict is as follows: ； in Indicates the first The degree of overall conflict between this indicator and the rest of the indicators; Indicates the first The first indicator and the first Correlation coefficients between the indicators; This indicates the degree of conflict or difference between two indicators; Indicates the total number of indicators; This represents the summation of all indicators; The formula for calculating the information content of an indicator is as follows: ； in Indicates the first The amount of information contained in each indicator; Indicates the first The standard deviation of an indicator across all ocean grid cells, i.e., the contrast intensity; Indicates the first The overall degree of conflict among the indicators; Indicates the first The first indicator and the first Correlation coefficients between the indicators; Indicates the first The degree of conflict between an individual indicator and all indicators; It is determined by both the intensity of the contrast and the degree of conflict; The formula for calculating the CRITIC objective weight is as follows: ； in Indicates the first CRITIC objective weights for each indicator; Indicates the first The amount of information contained in each indicator; Indicates the first The amount of information contained in each indicator; This represents the sum of information content from all indicators; This represents the total number of indicators; this normalization process ensures that the sum of all CRITIC weights is 1.

6. The method for assessing the suitability of offshore floating photovoltaic sites by integrating XAI and MCDA methods according to claim 1, characterized in that, In step 5, a game-theoretic combinatorial weighting method is used to optimize and fuse the XAI-FAHP weights obtained in step 3 and the CRITIC weights obtained in step 4, establishing an optimization model that minimizes the deviation between the combined weight vector and each basic weight vector. The combination coefficients are then solved and normalized to obtain the comprehensive weights, specifically: Let the XAI-FAHP weight vector be... The CRITIC weight vector is The comprehensive weight vector is obtained by game theory optimization. By constructing an optimization problem that minimizes the total deviation between the combined weight and each basic weight vector, the combination coefficients corresponding to the basic weight vectors are solved, and the combination coefficients are normalized to obtain the comprehensive weight vector used for TOPSIS calculation. The expression for the combined weights is as follows: ； This represents a combined weight vector composed of the basic weight vectors. Indicates the first There are 1 basic weight vector, where The XAI-FAHP weight vector, This is the CRITIC weight vector; This represents the transpose of the corresponding weight vector; Indicates the first The unnormalized combination coefficients of the basic weight vectors; This invention employs two types of basic weights; The optimization objective for the combination coefficients is as follows: ； in This indicates that the optimization objective is to minimize the objective function. Represents the Euclidean norm; This represents a combined weight vector synthesized from two basic weight vectors; Indicates the first Transpose of the basic weight vectors; Indicates the first The unnormalized combination coefficients of the basic weight vectors; This represents minimizing the deviation between the combined weights and the XAI-FAHP weights and CRITIC weights, respectively. The system of linear equations for solving the combination coefficients is as follows: ； in Represents the XAI-FAHP weight vector; Represents the CRITIC weight vector; and These represent the transposes of the corresponding weight vectors; Represents the weight vector and The inner product between; and These represent the unnormalized combination coefficients corresponding to the XAI-FAHP weight vector and the CRITIC weight vector, respectively; the vector on the right-hand side represents the constant term formed by the inner product of each basic weight vector. The formula for normalizing the combination coefficients is as follows: ； in Represents the normalized i-th Combination coefficients; Indicates the first Unnormalized combination coefficients; express The absolute value; This represents the sum of the absolute values ​​of the coefficients of two unnormalized combinations; The final overall weight calculation formula is as follows: ； in This represents the final integrated weight vector used in subsequent TOPSIS calculations; This represents the normalized combination coefficients corresponding to the XAI-FAHP weight vector; This represents the normalized combination coefficients corresponding to the CRITIC weight vector; Represents the XAI-FAHP weight vector; This represents the CRITIC weight vector.

7. The method for assessing the suitability of offshore floating photovoltaic sites by integrating XAI and MCDA methods according to claim 1, characterized in that, In step 6, based on the multi-source marine environment dataset processed in step 4 and the comprehensive weights obtained in step 5, the Topology-Optimal Solution Ranking (TOPSIS) method is used to calculate the comprehensive suitability assessment score for each marine grid cell within the assessment area. The marine grid cells are then ranked according to their comprehensive suitability assessment scores to form the suitability assessment results before constraint removal. Specifically: The attribute direction of each indicator is determined based on SHAP dependency or preset physical rules. Specifically, the attribute direction is set according to the average marginal contribution direction of each indicator to the output power or physical common sense. The attribute direction includes benefit-type or cost-type. In TOPSIS calculation, the evaluation matrix is ​​positiveized and standardized according to the indicator attributes. Cost-type indicators are converted into benefit-type indicators. The standardization adopts the range normalization method to construct a weighted standardized decision matrix under comprehensive weight constraints. The distance from each ocean grid cell to the positive ideal solution and the negative ideal solution is calculated, and the relative proximity score is obtained accordingly. The calculation formulas for positive transformation and standardization of profitability indicators are as follows: ； The calculation formulas for positiveizing and standardizing cost indicators are as follows: ； in Indicates the first The ocean grid cell in the first The original values ​​of each indicator; This represents the value after normalization and orientation. and They represent the first The maximum and minimum values ​​of each indicator in all ocean grid cells; The formula for calculating the elements of the weighted normalized matrix is ​​as follows: ； in In the weighted standardized decision matrix, the first... The first marine grid unit, the The elements corresponding to each indicator; The first game obtained by game theory combinatorial weighting is... The final weight of each indicator; This represents the first [unit] after forwarding and standardization. The grid cell in the first... The values ​​of each indicator; Indicates the ocean grid cell number; Indicates the indicator number; The formula for calculating the ideal solution is as follows: ； Represents the vector of the positive ideal solution; , to These represent the nth element in all ocean grid cells. , No. To the The maximum weighted standardized value of each indicator; Represents the elements of a weighted standardized decision matrix; Indicates the ocean grid cell number; Indicates the total number of indicators; The formula for calculating the negative ideal solution is as follows: ； Represents the negative ideal solution vector; , to These represent the nth element in all ocean grid cells. , No. To the The least weighted standardized value of each indicator; Represents the elements of a weighted standardized decision matrix; Indicates the ocean grid cell number; Indicates the total number of indicators; The formula for calculating the distance to the ideal solution is as follows: ； Indicates the first From ocean grid cells to the ideal solution The Euclidean distance; Indicates the first The grid cell in the first... Weighted standardized values ​​for each indicator; In the positive ideal solution, the first The values ​​of each indicator; This represents the summation of all indicators; Indicates the total number of indicators; Represents the square root operation; The formula for calculating the distance to the negative ideal solution is as follows: ； Indicates the first From a single ocean grid cell to a negative ideal solution The Euclidean distance; Indicates the first The grid cell in the first... Weighted standardized values ​​for each indicator; In the negative ideal solution, the first... The values ​​of each indicator; This represents the summation of all indicators; Indicates the total number of indicators; Represents the square root operation; The formula for calculating the relative closeness score is as follows: ； Indicates the first The relative proximity score of each ocean grid cell; Indicates the first The Euclidean distance from each grid cell to the positive ideal solution; Indicates the first The Euclidean distance from each grid cell to the negative ideal solution; The range of values ​​is to The larger the value, the closer it is to the ideal state and the further it is from the negative ideal state, and the higher the suitability of offshore floating photovoltaic deployment.

8. A marine floating photovoltaic site suitability assessment system integrating XAI and MCDA methods, characterized in that, A method for assessing the suitability of offshore floating photovoltaic sites for implementing the fusion of XAI and MCDA methods as described in any one of claims 1 to 7, comprising: The data acquisition module is used to collect historical operational data, multi-source marine environmental data, and spatial constraint data; The data preprocessing module is used to perform outlier removal, missing value handling, time alignment, spatial registration, and dimensionless processing. The model training module is used to train multiple tree-based regression models and perform model validation. The XAI analysis module is used to generate metric contributions and global importance based on the SHAP algorithm; The XAI-FAHP weighting module is used to construct a fuzzy judgment matrix based on the SHAP importance ratio of multiple models and output the XAI-FAHP weights. The CRITIC calculation module is used to evaluate the CRITIC weights output from multi-source marine environmental datasets for a region. The game theory combinatorial weighting module is used to optimize and merge XAI-FAHP weights and CRITIC weights and output a comprehensive weight. The TOPSIS ranking module is used to score and rank ocean grid cells based on their suitability. The spatial constraint removal module is used to perform overlay masking processing of exclusive economic zones, marine protected areas, tropical cyclone frequency, water depth, offshore distance, and sea ice concentration after TOPSIS sorting.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method for assessing the suitability of offshore floating photovoltaic sites that integrates the XAI and MCDA methods as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the method for assessing the suitability of offshore floating photovoltaic sites by integrating the XAI and MCDA methods as described in any one of claims 1 to 7.

Images