Performance test optimization method and device for wind and wave hybrid energy obtaining system

Abstract

The invention discloses a performance test optimization method and device for a wind and wave hybrid energy obtaining system, and relates to the field of wind and wave hybrid energy obtaining simulation optimization and the field of data processing. The method comprises the following steps: generating an initial database based on a full-coupling virtual test model, and determining a six-dimensional comprehensive performance test matrix and a comprehensive scoring mode of a core index; performing intelligent testing on the target parameter space by using a Bayesian algorithm, and screening a recommendation scheme; carrying out feature importance evaluation and weight optimization on the core indexes of the recommendation schemes by adopting a LightBGM model, and calculating to obtain a comprehensive score of each recommendation scheme; and carrying out optimization decision making by combining visualization and intelligent diagnosis results, selecting a recommendation scheme with the highest comprehensive score as a target design scheme, and if the scheme does not meet requirements, carrying out iterative optimization until a final scheme is selected. According to the scheme provided by the invention, efficient optimization and accurate evaluation of the system design scheme can be realized, the research and development period is remarkably shortened, and the test cost is reduced.

Description

Technical Field

[0001] This application relates to the field of data processing, and in particular to a method and apparatus for performance testing and optimization of a wind and wave hybrid energy harvesting system. Background Technology

[0002] As a complex multi-body, multi-physics coupled system, the core of the design optimization of wind and wave hybrid energy harvesting systems lies in how to maximize power generation performance while ensuring its safety performance throughout its entire life cycle (especially under extreme sea conditions). Performance testing is a key means to achieve this goal, including physical model tests and numerical simulation tests. These systems typically involve the integration of wind turbines and wave energy converters (WECs), and their performance is affected by the complex interactions of numerous design parameters (such as PTO system parameters, mooring system configuration, and floating body geometry).

[0003] In related technologies, a parametric scanning test method based on a single target can be used for performance testing. The typical process of this method is as follows: In order to optimize a certain performance (e.g., total power generation), engineers will set a series of discrete values ​​for one or two key parameters (e.g., PTO damping coefficient and spring stiffness) to form a parameter grid. Then, each parameter combination in the grid is tested one by one through numerical simulation, and finally the parameter combination that makes the single performance index optimal is found. For safety, it is usually done after the design is completed, by selecting one or two extreme conditions (e.g., a once-in-a-century storm) to conduct an independent ultimate load verification test.

[0004] However, existing performance testing methods heavily rely on human experience to set evaluation indicators, normalization formulas, and weighting coefficients. The evaluation dimensions are singular, and there is a lack of systematic consideration of various fault conditions such as cable breakage, resulting in highly subjective and unreliable evaluation results. At the same time, existing methods usually adopt an "exhaustive" parameter scanning approach, lacking an intelligent test point selection mechanism, and cannot scientifically balance conflicting performance aspects such as "power generation" and "safety." When the evaluation results are unsatisfactory, they cannot automatically diagnose performance bottlenecks and provide actionable optimization suggestions.

[0005] Therefore, overcoming the aforementioned technical problems and defects has become a key issue that needs to be addressed. Summary of the Invention

[0006] To address the aforementioned technical issues, this application provides a performance testing and optimization method and apparatus for a wind-wave hybrid energy harvesting system. By constructing a closed-loop optimization process that combines virtual testing with intelligent algorithms, it is possible to achieve efficient optimization and accurate evaluation of system design schemes, significantly shortening the R&D cycle and reducing testing costs.

[0007] According to one aspect of this application, a performance testing and optimization method for a wind-wave hybrid energy harvesting system is provided, the method comprising:

[0008] Establish a parameterized, fully coupled virtual test model to form a design parameter database; substitute the test object to form a performance index database;

[0009] A six-dimensional comprehensive performance test matrix and a candidate scheme scoring method were determined; the six-dimensional comprehensive performance test matrix includes six core indicators across two dimensions: power generation performance and safety performance.

[0010] Intelligent parameter testing is performed using Bayesian optimization to select recommended solutions;

[0011] LightGBM is used to extract the model quantification of the comprehensive performance evaluation results, optimize the weight allocation of core indicators, and evaluate the comprehensive score of the recommended solution.

[0012] Optimization decisions are made based on visualization and intelligent diagnostics to determine whether direct optimization is needed.

[0013] If not required, the target design scheme is determined based on the comprehensive score of the recommended scheme; if required, the design is improved to form a new candidate scheme, and the intelligent parameter test is carried out again.

[0014] In the above scheme, the design improvement forms a new candidate scheme, including:

[0015] The LightGBM diagnostic model was used to predict and evaluate each recommended solution, and the evaluation results were obtained.

[0016] The evaluation results of each recommended solution are analyzed and diagnosed using model interpretation tools to identify the performance bottleneck information of each recommended solution;

[0017] Based on the performance bottleneck information, the configuration of the corresponding recommended solution is optimized to obtain the corresponding optimization solution;

[0018] All optimized solutions are considered as new candidate solutions.

[0019] In the above scheme, the candidate scheme scoring method is configured to include the normalized score of each core indicator in the corresponding scheme and the comprehensive performance score of the corresponding scheme; wherein, the comprehensive performance score is obtained by weighting the normalized scores of each core indicator.

[0020] In the above scheme, the step of using model interpretation tools to analyze and diagnose the evaluation results of each recommended scheme and determine the performance bottleneck information of each recommended scheme includes:

[0021] The SHAP interpreter is invoked to perform attribution analysis and contribution quantification on the evaluation results of each recommended solution, thereby identifying the corresponding performance bottleneck information.

[0022] Based on the performance bottleneck information, the bottleneck indicators with performance bottlenecks in the core indicators corresponding to each recommendation scheme are identified, and a visual diagnostic report for each recommendation scheme is generated; wherein, the visual diagnostic report includes the basic information of the bottleneck indicators in the corresponding scheme and the causes of their occurrence.

[0023] In the above scheme, optimizing the configuration of the corresponding recommended scheme based on the performance bottleneck information to obtain the corresponding optimized scheme includes:

[0024] Using the SHAP interpreter, optimization strategies for each bottleneck metric are generated based on the visualized diagnostic report;

[0025] Based on the preset priority of each core indicator, the optimization strategies corresponding to each bottleneck indicator are sorted, and optimization suggestions for the corresponding recommended solutions are generated.

[0026] In the above scheme, the core indicators corresponding to the power generation performance dimension include total output power, platform motion stability, and motion synergy effect; the core indicators corresponding to the safety performance dimension include maximum mooring tension, platform extreme deviation, and fault robustness index.

[0027] In the above scheme, the fault robustness index is determined by quantifying the degree of performance degradation after failure; the process of determining the fault robustness index includes:

[0028] Determine the motion deterioration ratio and the mooring deterioration ratio;

[0029] Based on the motion degradation ratio and the mooring degradation ratio, the degree of system performance degradation under failure conditions is determined, and the fault robustness index is obtained.

[0030] In the above scheme, the step of using Bayesian optimization for intelligent parameter testing to select recommended schemes includes:

[0031] Define the objective function and parameter space;

[0032] Within the parameter space, an initial sample set containing n test points is generated by Latin hypercube sampling, and the comprehensive performance score is calculated to construct the initial GPR proxy model.

[0033] Select the next test point based on the acquisition function;

[0034] Perform the simulation and update the agent model;

[0035] The algorithm iteratively selects the next test point based on the updated proxy model until it converges and finds the optimal solution.

[0036] In the above scheme, the step of using LightGBM to extract the model quantification comprehensive performance evaluation results, optimizing the weight allocation of core indicators, and evaluating the comprehensive score of the recommended scheme includes:

[0037] Based on the proportion of the benefits of each recommended solution in the cost expenditure, the corresponding benefit-cost ratio is determined, and the benefit-cost ratio is used as the ultimate evaluation index of the corresponding recommended solution.

[0038] The normalized scores of the six core metrics in each recommendation scheme are used as input features, and the final evaluation metrics of the corresponding recommendation scheme are used as output labels. The gradient boosting decision tree model is trained under supervision to generate the LightGBM extraction model.

[0039] Based on the feature importance of each core indicator in the LightGBM extraction model, the weight allocation of the core indicators is optimized.

[0040] The overall score of the recommended solution is evaluated based on the optimized weight allocation of the core indicators.

[0041] According to a second aspect of this application, a performance testing and optimization device for a wind-wave hybrid energy harvesting system is provided. The device comprises: a performance testing and optimization device for a wind-wave hybrid energy harvesting system, characterized in that the device includes:

[0042] The first processing unit is used to establish a parameterized fully coupled virtual test model to form a design parameter database; substitute the test object to form a performance index database; and determine the six-dimensional comprehensive performance test matrix and the candidate scheme scoring method; the six-dimensional comprehensive performance test matrix includes six core indicators in two dimensions: power generation performance and safety performance.

[0043] The testing and evaluation unit is used to perform intelligent parameter testing using Bayesian optimization to select recommended solutions; and to use LightGBM to extract the model quantification comprehensive performance evaluation results, optimize the weight allocation of core indicators, and evaluate the comprehensive score of the recommended solutions.

[0044] The second processing unit is used to make optimization decisions based on visualization and intelligent diagnosis, and to determine whether the solution needs to be optimized. If not, the target design solution is determined based on the comprehensive score of the recommended solution. If it is needed, the design is improved to form a new candidate solution, and the intelligent parameter test is carried out again.

[0045] The performance testing and optimization method and apparatus for a wind-wave hybrid energy harvesting system provided in this application constructs a parallel testing framework encompassing multiple dimensions and operating conditions. It innovatively introduces machine learning models for objective performance evaluation, intelligent bottleneck diagnosis, and accelerated optimization processes. This integrates the previously independent and contradictory performance testing and optimization processes into a systematic, closed-loop, and intelligent scientific decision-making process, improving the accuracy and reliability of test evaluation results. Furthermore, the introduction of Bayesian optimization methods effectively improves test evaluation efficiency. Furthermore, the introduction of a six-dimensional comprehensive performance testing matrix enables performance evaluation of core indicators across multiple dimensions, thereby obtaining the optimal design scheme while balancing high power generation efficiency and high safety, improving the accuracy and reliability of evaluation and optimization results. Finally, the introduction of machine learning models for objective performance evaluation and SHAP for intelligent bottleneck diagnosis enhances the objectivity and accuracy of performance test evaluation results, further improving the reliability of the evaluation results.

[0046] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0047] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application, and do not constitute an undue limitation of this application.

[0048] Figure 1 A schematic diagram of a performance testing and optimization method for a wind-wave hybrid energy harvesting system provided in this application embodiment;

[0049] Figure 2 This is a schematic diagram of the fully coupled virtual test model of the wind-wave hybrid energy harvesting system in the performance test optimization method of the wind-wave hybrid energy harvesting system in the embodiments of this application;

[0050] Figure 3 This is a flowchart illustrating step S103 of the performance testing and optimization method for the wind-wave hybrid energy harvesting system in this application embodiment.

[0051] Figure 4 This is a schematic diagram of the mean prediction results of the GP model in the performance testing and optimization method of the wind-wave hybrid energy harvesting system in the embodiments of this application;

[0052] Figure 5 This is a schematic diagram of the BO convergence curve in the performance testing and optimization method of the wind-wave hybrid energy harvesting system in this application embodiment;

[0053] Figure 6 This is a schematic diagram of the parameter space exploration trajectory in the performance testing and optimization method of the wind-wave hybrid energy harvesting system in this application embodiment;

[0054] Figure 7 This is a schematic diagram showing the efficiency comparison between the Bayesian optimization method and the grid method in the performance testing and optimization method of the wind-wave hybrid energy harvesting system in the embodiments of this application;

[0055] Figure 8 This is a flowchart illustrating step S104 of the performance testing and optimization method for the wind-wave hybrid energy harvesting system in this application embodiment.

[0056] Figure 9 This is a schematic diagram of the objective weight allocation results in the performance testing and optimization method of the wind-wave hybrid energy harvesting system in this application embodiment;

[0057] Figure 10 This is a schematic diagram showing the performance comparison results of six core indicators of three candidate schemes (A, B, and C) in the performance testing and optimization method of the wind-wave hybrid energy harvesting system in the embodiments of this application.

[0058] Figure 11 This is a flowchart illustrating the optimization decision-making process based on visualization and intelligent diagnosis in the performance testing and optimization method of the wind and wave hybrid energy harvesting system according to an embodiment of this application.

[0059] Figure 12 This is a flowchart illustrating step S1102 of the performance testing and optimization method for the wind-wave hybrid energy harvesting system in an embodiment of this application.

[0060] Figure 13 This is a radar chart showing the performance index of scheme A in the performance testing and optimization method for the wind-wave hybrid energy harvesting system in this application embodiment;

[0061] Figure 14 This is a SHAP waterfall diagram in the performance testing and optimization method of the wind-wave hybrid energy harvesting system in the embodiments of this application;

[0062] Figure 15 This is a radar chart showing the optimized performance indicators of Scheme A in the performance testing and optimization method for the wind-wave hybrid energy harvesting system in this application embodiment.

[0063] Figure 16 A schematic diagram of a performance testing and optimization device for a wind-wave hybrid energy harvesting system provided in this application embodiment. Detailed Implementation

[0064] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0065] Existing performance testing methods have many problems in their implementation.

[0066] First, the performance evaluation is fragmented and cannot handle conflicts between multiple objectives. Specifically, this method treats "power generation performance testing" and "safety performance testing" as two independent and sequential processes, which cannot effectively reveal and handle the strong coupling and mutual constraints between the two. At the same time, the parameters optimized with the goal of maximizing power generation may lead to the system exceeding the limits of mooring tension or platform motion under survivable conditions, and vice versa.

[0067] Secondly, there is a lack of testing for fault conditions, which poses a significant safety hazard. Specifically, sudden failures such as "cable breakage" are key to determining the final safety of the platform, but existing testing procedures rarely include them in the routine performance evaluation system, which may lead to undiscovered "brittle" failure modes in the design.

[0068] Third, the optimization efficiency is low and limited by the "curse of dimensionality". Specifically, a simple parameter scan is an "exhaustive" test. When there are more than three design variables, the workload of the test increases exponentially (i.e., the "curse of dimensionality"). With limited computing resources, only a rough and low-precision search can be performed, which can easily miss the true optimal solution.

[0069] Fourth, the weighting is subjective, and the evaluation results lack credibility. Specifically, when a comprehensive evaluation is required, the weighting coefficients of different performance indicators (such as power, cost, and safety) are usually set subjectively by the design team based on experience. This makes the evaluation results dependent on the decision-maker's preferences and lacks objectivity and repeatability.

[0070] Based on this, in various embodiments of this application, a parallel testing framework encompassing multiple dimensions and operating conditions is constructed. Machine learning models are innovatively introduced to objectively evaluate performance, intelligently diagnose bottlenecks, and accelerate the optimization process. This integrates the originally independent and contradictory performance testing and optimization processes into a systematic, closed-loop, and intelligent scientific decision-making process, improving the accuracy and reliability of test evaluation results. Furthermore, the introduction of Bayesian optimization methods effectively improves test evaluation efficiency. Furthermore, the introduction of a six-dimensional comprehensive performance testing matrix enables performance evaluation of core indicators across multiple dimensions, thereby obtaining the optimal design scheme while balancing high power generation efficiency and high safety, improving the accuracy and reliability of evaluation and optimization results. Furthermore, the introduction of machine learning models for objective performance evaluation and SHAP for intelligent bottleneck diagnosis enhances the objectivity and accuracy of performance test evaluation results, further improving the reliability of the evaluation results.

[0071] This application provides a performance testing and optimization method for a wind-wave hybrid energy harvesting system, such as... Figure 1 As shown, the method may include steps S101 to S107. Steps S101 to S107 will be described in detail below with reference to specific embodiments.

[0072] S101: Establish a parameterized, fully coupled virtual test model to form a design parameter database; substitute the test object to form a performance index database.

[0073] In practical applications, multiple candidate schemes can be generated based on the data in the initial database and the preset six-dimensional comprehensive performance test matrix. The initial database includes a design parameter database and a performance index database. The design parameter data in the design parameter database is used to train the LightGBM diagnostic model, and the time history data of the performance index in the performance index database is used to train the LightGBM extraction model. The six-dimensional comprehensive performance test matrix includes six core indicators, which are divided into two categories: power generation performance dimension and safety performance dimension.

[0074] In practical applications, a high-fidelity, fully coupled virtual test model of a wind-wave hybrid energy harvesting system can be established. This model can be configured as a parameter-driven system, meaning that key design parameters of the system (such as WEC dimensions, PTO damping, mooring configuration, etc.) and environmental parameters (wind, waves, currents, etc.) can be modified as external input variables. This model can also serve as a "data generator" for subsequent steps to build the initial database.

[0075] Figure 2 The structure of a fully coupled virtual test model of a wind-wave hybrid energy harvesting system provided in an embodiment of this application is shown, as follows: Figure 2 As shown, the fully coupled data model may include an input module C101, an external load calculation module C105, an internal interactive calculation module C106, a unified time-domain dynamics solver C104, and an output module C107. The functions of each module in the fully coupled data model are explained below.

[0076] Input module C101 is the entry point for the entire fully coupled virtual test model, used to define the initial state and boundary conditions of the system; input module C101 includes design parameter input module C102 and environmental condition input module C103.

[0077] The design parameter input module C102 is used to define the physical characteristics of the system under test, including key design parameters such as mooring system type, WEC geometry, and PTO damping coefficient.

[0078] The environmental condition input module C103 is used to define the marine environment in which the simulation takes place; specifically, the simulated marine environmental parameters include wave spectrum, wind speed profile, water depth, ocean current, etc.

[0079] The external load calculation module C105 is used to calculate various loads that are directly applied to the system by the external environment. The external load calculation module C105 may include three parallel sub-modules: aerodynamic module C1051, mooring dynamics module C1052 and hydrodynamics module C1053.

[0080] The aerodynamics module C1051 is used to calculate the aerodynamic loads exerted by wind on the wind turbine.

[0081] The mooring dynamics module C1052 is used to calculate the restoring force and damping force provided by the mooring system, i.e., the mooring tension.

[0082] The hydrodynamics module C1053 is used to calculate the hydrodynamic forces of waves and ocean currents acting on a floating body. It employs a hybrid fidelity strategy, combining potential flow theory with viscous correction.

[0083] The internal interactive computing module C106 is used to calculate the interaction forces between various components within the system. The internal interactive computing module C106 contains two parallel sub-modules: the multi-connection and constraint module C1061 and the PTO module C1062.

[0084] The multi-connection and constraint module C1061 is used to calculate the connection forces or constraint forces between components.

[0085] The PTO module C1062 is used to calculate the damping force generated by the PTO according to the control strategy, and simultaneously calculate its power generation.

[0086] The unified time-domain dynamics solver C104 is the core of the entire fully coupled virtual test model. It is used to solve the system's equations of motion in the time domain, achieving tight coupling of various physical fields.

[0087] The unified time-domain dynamics solver C104 can be specifically used for:

[0088] The solver receives force signals from each module; specifically, at each time step, the solver first gathers all force signals from the external load calculation module C105 and the internal interactive calculation module C106.

[0089] Solve for the motion state; specifically, based on Newton's second law, solve for the acceleration, velocity, and displacement of the system at that time step;

[0090] The motion state is fed back to each module; specifically, the newly calculated motion state is fed back to the external load calculation module C105 and the internal interactive calculation module C106 so that they can calculate the force and load for the next time step, thus forming a calculation closed loop.

[0091] The output module C107 is used to collect and output simulation results, providing raw data for subsequent performance evaluation.

[0092] In practical applications, the simulation results output by output module C107 may include:

[0093] Platform six-degree-of-freedom motion time history;

[0094] Mooring cable tension time history;

[0095] Instantaneous wind power generation time history;

[0096] WEC instantaneous power generation time history.

[0097] In practical applications, the simulation results output by the output module C107 may also include other performance indicators such as structural stress. The specific simulation results to be output can be configured according to the actual application scenario, and this application embodiment does not limit this.

[0098] In practical applications, at each time step, the external load module C101 and the internal interactive calculation module C106 can calculate the current forces based on the motion state of the previous moment and transmit these force signals to the unified temporal dynamics solver C104. The unified temporal dynamics solver C104 solves the motion state at the current moment, outputs the result to the output module C107 for recording, and feeds back the new motion state to the external load module C101 and the internal interactive calculation module C106 for calculation of the next time step. In this embodiment, this iterative process of "state-force-new state" constitutes the core loop of the entire temporal coupled simulation.

[0099] After establishing the fully coupled virtual test model, the initial database can be constructed using the fully coupled virtual test model.

[0100] Specifically, batch simulation of the fully coupled virtual test model can be performed. For example, hundreds of design parameter combinations can be generated by using Latin hypercube sampling, thereby generating an initial database containing design parameters X and corresponding original performance index time history data Y; this initial database is the basis for subsequent machine learning model training.

[0101] In practical applications, multiple candidate solutions can be generated based on the core indicators in the predefined six-dimensional comprehensive performance test matrix and the data in the initial database.

[0102] In practical applications, the inclusion of both power generation performance and safety performance indicators in the pre-defined six-dimensional comprehensive performance test matrix ensures that the evaluated scheme can take into account both safety and performance, thereby achieving a balance between power generation and safety in the wind and wave hybrid energy system, and improving system safety while ensuring power generation capacity.

[0103] In one embodiment, the core indicators corresponding to the power generation performance dimension include total output power, platform motion stability, and motion synergy; the core indicators corresponding to the safety performance dimension include maximum mooring tension, platform extreme deviation, and fault robustness index.

[0104] Here, the total output power can be defined as dimension 1, the platform motion stability as dimension 2, the motion synergy effect as dimension 3, the maximum mooring tension as dimension 4, the platform limit offset as dimension 5, and the fault robustness index as dimension 6.

[0105] S102: Determine the six-dimensional comprehensive performance test matrix and the candidate scheme scoring method; the six-dimensional comprehensive performance test matrix includes six core indicators in two dimensions: power generation performance and safety performance.

[0106] In practical applications, the candidate solution scoring method is configured to include the normalized score of each core indicator in the corresponding solution and the comprehensive performance score of the corresponding solution.

[0107] In practical applications, the original test indicators of each core dimension can be converted into normalized scores between 0 and 1 using a preset scoring function, where 1 is the best and 0 is the worst. The calculation process of the six core indicators is explained in detail below.

[0108] Dimension 1 scoring, i.e., the normalized score for total output power, can be expressed as:

[0109] P_score = (P_avg - P_benchmark) / (P_target - P_benchmark);

[0110] Wherein, P_score represents the normalized score of the total output power, with a value range of [0, 1], P_avg represents the average total output power (kW) obtained from the numerical simulation, P_benchmark represents the benchmark power value (kW), representing the industry average level or the typical value of existing technology, and P_target represents the target power value (kW), representing the ideal target value designed for this project.

[0111] The scoring logic for total output power is as follows:

[0112] When P_avg ≤ P_benchmark, P_score = 0 (benchmark not reached);

[0113] When P_avg ≥ P_target, P_score = 1 (target achieved);

[0114] Linear interpolation is used when P_benchmark < P_avg < P_target.

[0115] The physical meaning of this core indicator is that the function maps power to the interval [0, 1], and the closer it is to or exceeds the target value, the higher the score.

[0116] For example, if P_benchmark = 5000 kW and P_target = 7000 kW, and the actual measured P_avg = 6000 kW, then P_score = (6000-5000) / (7000-5000) = 0.50.

[0117] The normalized score for platform motion stability can be expressed as:

[0118] S_motion_score = 1.0 - (σ_pitch / σ_limit);

[0119] Wherein, S_motion_score represents the normalized score, with a value range of [0, 1], σ_pitch represents the standard deviation of the pitch angle (°) extracted from the time-domain simulation, reflecting the severity of the platform's sway, and σ_limit represents the preset pitch angle safety threshold (°). When the pitch angle exceeds this value, it will lead to the wind turbine cutting out or structural fatigue.

[0120] Dimension 2 scoring, namely the scoring logic for platform motion stability, is as follows:

[0121] When σ_pitch ≥ σ_limit, S_motion_score = 0 (motion exceeds limit);

[0122] When σ_pitch = 0, S_motion_score = 1 (ideal state);

[0123] When 0 < σ_pitch < σ_limit, it decreases linearly.

[0124] The physical meaning of this core indicator is that the smaller the platform movement, the better the stability and the higher the score; this indicator directly affects the effective working time and fatigue life of the wind turbine.

[0125] For example, if σ_limit = 6.0°, and the actual measured σ_pitch = 4.0°, then S_motion_score = 1.0 - 4.0 / 6.0 = 0.333.

[0126] The normalized score of the motion synergy effect can be expressed as:

[0127] C_motion_score = (σ_without_WEC - σ_with_WEC) / σ_without_WEC;

[0128] Wherein, C_motion_score represents the normalized score, i.e. the roll reduction rate of WEC, with a value range of [0, 1], σ_without_WEC represents the standard deviation of pitch (°) when only the wind turbine platform is used (without WEC), and σ_with_WEC represents the standard deviation of pitch (°) after integrating WEC.

[0129] Dimension 3 scoring, namely the scoring logic for motion synergy, is as follows:

[0130] When σ_with_WEC ≥ σ_without_WEC, then C_motion_score = 0 (WEC has no anti-shake effect or may even worsen the motion).

[0131] When σ_with_WEC = 0, then C_motion_score = 1 (complete suppression of motion, ideal state);

[0132] When 0 < σ_with_WEC < σ_without_WEC, the calculation is based on the reduction rate.

[0133] The physical significance of this core indicator lies in the fact that it quantifies the "cooperative inhibition" effect of WEC on platform motion; the higher the score, the greater the stabilization contribution of WEC.

[0134] For example, if σ_without_WEC = 5.0° without WEC and σ_with_WEC = 3.5° after integrating WEC, then C_motion_score = (5.0-3.5) / 5.0 / target motion reduction rate = 0.30 (motion reduction rate 30%) / target motion reduction rate (40%) = 0.75.

[0135] The normalized score of the maximum mooring tension can be represented by a piecewise function, specifically as follows:

[0136]

[0137] Where T_score represents the normalized score, with a value range of [0, 1], T_max represents the maximum instantaneous tension (kN) of the mooring cable extracted from the numerical simulation, T_safe represents the safety threshold (kN), when the tension is below this value the system has sufficient safety margin, and T_break represents the minimum breaking load (kN) of the mooring cable, when the tension exceeds this value the cable will break.

[0138] Dimension 4 scoring, namely the scoring logic for maximum mooring tension, is as follows:

[0139] First segment (T_max ≤ T_safe): The tension is within the safe range, and the score decreases slowly;

[0140] The second segment (T_safe < T_max ≤ T_break): Tension enters the risk zone, and the score drops rapidly and penalized.

[0141] The third segment (T_max > T_break): mooring failure, score 0.

[0142] The physical significance of this core indicator lies in the fact that piecewise linearity is used to reflect the nonlinear risk accumulation characteristics in engineering; the marginal benefit of reducing tension within the safe zone is not significant; once entering the risk zone, the risk rises sharply, and the score should decrease rapidly and punitively.

[0143] For example, if T_safe = 6000 kN and T_break = 10000 kN, and the actual measured T_max = 7500 kN (in the second segment), then T_score = 0.8*(10000-7500) / (10000-6000) = 0.8*0.625 = 0.50.

[0144] The Dimension 5 scoring, i.e., the normalized score of the platform's extreme offset, can be expressed as:

[0145] O_score = 1.0 - (Offset_max / Offset_limit);

[0146] Where O_score represents the normalized score, with a value range of [0, 1], Offset_max represents the maximum horizontal offset distance (m) of the platform extracted from the numerical simulation, and Offset_limit represents the preset offset limit (m), which is determined by the geometric constraints of the mooring system and the cable tension limit.

[0147] The scoring logic for platform extreme offset is as follows:

[0148] When Offset_max ≥ Offset_limit, O_score = 0 (offset exceeds limit);

[0149] When Offset_max = 0, O_score = 1 (no offset, ideal state);

[0150] When 0 < Offset_max < Offset_limit, it decreases linearly.

[0151] The physical significance of this core indicator is that excessive platform offset will lead to increased dynamic cable tension and deterioration of mooring angle, affecting system safety.

[0152] For example, if Offset_limit = 80 m, and the actual measured Offset_max = 45 m, then O_score = 1.0 - 45 / 80 = 0.438.

[0153] The 6-dimensional scoring means that the fault robustness index can be represented by quantifying the degree of performance degradation after failure.

[0154] Based on this, in one embodiment, the fault robustness index is determined by quantifying the degree of performance degradation after failure; the process of determining the fault robustness index includes:

[0155] Determine the motion deterioration ratio and the mooring deterioration ratio;

[0156] Based on the motion degradation ratio and the mooring degradation ratio, the degree of system performance degradation under failure conditions is determined, and the fault robustness index is obtained.

[0157] In practical applications, the motion degradation ratio can be quantified using six degrees of freedom: surge, sway, heave, roll, pitch, and yaw. Specifically, for the i-th degree of freedom:

[0158] The formula for calculating the amplitude of motion in the i-th degree of freedom under normal operating conditions is as follows:

[0159] Amp_normal_i = (max_motion_i - min_motion_i) / 2;

[0160] The motion amplitude of the i-th degree of freedom after failure is calculated using the following formula:

[0161] Amp_post_i = (max_motion_post_i - min_motion_post_i) / 2;

[0162] The deterioration ratio of the i-th degree of freedom is calculated using the following formula:

[0163] R_motion_i = Amp_post_i / Amp_normal_i;

[0164] Take the most severe deterioration ratio as the motion deterioration ratio R_motion:

[0165] R_motion = max(R_motion_1, R_motion_2, ..., R_motion_6).

[0166] The physical meaning of the motion degradation ratio is that it can reflect the degree of deterioration of the system's motion attitude after failure; specifically, R_motion=1 means no degradation, and R_motion=2 means the motion amplitude doubles.

[0167] The formula for calculating the mooring deterioration ratio is expressed as:

[0168] R_mooring = T_max_post / T_max_normal;

[0169] Where R_mooring represents the mooring deterioration ratio, T_max_normal represents the maximum mooring tension (kN) under normal operating conditions, and T_max_post represents the maximum mooring tension (kN) of the remaining mooring line under failure conditions.

[0170] The physical meaning of the mooring degradation ratio is that it can reflect the degree of increase in the load borne by the remaining mooring system after failure.

[0171] Once the motion degradation ratio and mooring degradation ratio are determined, the comprehensive failure robustness index is calculated based on these two parameters; the specific calculation formula is as follows:

[0172] FRI = 1 / (0.6 * R_motion + 0.4 * R_mooring);

[0173] Here, FRI represents the robustness index, with a value range of (0, 1]. The closer the FRI value is to 1, the smaller the performance degradation after a failure and the stronger the robustness. Here, the weights of 0.6 and 0.4 are set based on engineering experience, with 0.6 corresponding to motion safety and 0.4 corresponding to structural safety.

[0174] Here, since the robustness index itself is a normalized index, FRI can be directly used as FRI_score, that is, the robustness index can be directly used as the corresponding normalized score.

[0175] The evaluation criteria for the robustness index are as follows:

[0176] FRI > 0.7: Excellent (performance degradation <30% after failure);

[0177] 0.5 < FRI ≤ 0.7: Pass (30-50% degradation);

[0178] FRI ≤ 0.5: Unacceptable (degradation > 50%, posing a safety hazard).

[0179] For example, the test data under normal operating conditions are: Pitch amplitude = 5.0°, maximum mooring tension = 5000 kN; under the condition of cable breakage, the test data after cable breakage are: Pitch amplitude = 12.0° (the degree of freedom with the most severe deterioration), maximum tension of the remaining cable = 8000 kN; at this time, R_motion = 12.0 / 5.0 = 2.4 (pitch amplitude increases by 140%), R_mooring = 8000 / 5000 = 1.6 (tension increases by 60%), FRI = 1 / (0.6×2.4 + 0.4×1.6) = 1 / (1.44 + 0.64) = 1 / 2.08 = 0.481; therefore, it can be concluded that: FRI = 0.481 < 0.5, the scheme is "unqualified", there is a safety hazard, and the design needs to be optimized.

[0180] After calculating the normalized score of each core indicator, the normalized scores of all core indicators can be weighted to obtain the overall performance score of the system.

[0181] S_total = Σ(W_i * Score_i), i=1~6;

[0182] Where S_total is the final comprehensive performance score, with a value range of [0, 1], W_1 ~ W_6 are the weight coefficients of the six dimensions (i.e., the six core indicators), respectively, satisfying ΣW_i=1; Score_i is the normalized score of the i-th dimension.

[0183] S103: Uses Bayesian optimization for intelligent parameter testing to select recommended solutions.

[0184] Specifically, based on the candidate solution scoring results and the preset objective function, Bayesian optimization methods can be used to select multiple recommended solutions from all candidate solutions.

[0185] Here, by employing Bayesian optimization methods to intelligently select test parameter points, the problem of low efficiency in traditional parameter scanning in multidimensional parameter space can be solved, thereby improving the efficiency of the entire test evaluation process.

[0186] In one embodiment, such as Figure 3 As shown, the step of using Bayesian optimization for intelligent parameter testing and selecting recommended solutions, i.e., S103, can include S301 to S305. The following detailed description of S301 to S305 is provided in conjunction with specific embodiments.

[0187] S301: Define the objective function and parameter space.

[0188] In practical applications, the optimization parameters of the objective function include the damping coefficient and the mooring cable pretension.

[0189] In practical applications, defining a parameter space can be understood as defining N design parameters that need to be optimized and their value ranges. In this embodiment, the parameter space is represented by X, and the preset key optimization parameters include x1 and x2, where x1 is the PTO damping coefficient with a value range of [10000, 50000] Ns / m, and x2 is the mooring cable pretension with a value range of [500, 2000] kN.

[0190] In practical applications, the objective function f(X) can be understood as encapsulating the complete process of "inputting a set of design parameters X = {x1, x2}, running a fully coupled numerical simulation, and calculating the final comprehensive performance score S_total according to the method of S102" into a "black box" objective function; the goal is to find the parameter combination X* that maximizes f(X).

[0191] S302: Within the parameter space, an initial sample set containing n test points is generated by Latin hypercube sampling, and the comprehensive performance score is calculated to construct the initial GPR proxy model.

[0192] In practical applications, n is an integer greater than 0, and the specific value can be configured according to the actual application scenario. For example, n can be configured as 5.

[0193] In practical applications, S302 is the process of initializing the proxy model.

[0194] In this process, an initial sample set containing 5 points was generated in the parameter space by Latin hypercube sampling, and its S_total value was calculated to build the initial GPR proxy model. The initial test data is shown in Table 1, and the initial optimal score is 0.8616 (Test 3).

[0195] Table 1:

[0196] Test No. PTO damping (Ns / m) Pretension (kN) Overall score Testing methods Test 1 18000 650 0.7423 Initial sampling (LHS) Test 2 42000 1450 0.6824 Initial sampling (LHS) Test 3 28000 1100 0.8616 Initial sampling (LHS) Test 4 22000 1700 0.7245 Initial sampling (LHS) Test 5 38000 825 0.7891 Initial sampling (LHS)

[0197] S303: Select the next test point based on the acquisition function.

[0198] In practical applications, during the 6th iteration, the algorithm uses the "Upper Confidence Interval (UCB) sampling function" to decide the next test point. The specific calculation formula is as follows:

[0199] UCB(X) = μ(X) + κ*σ(X);

[0200] Where μ(X) represents the mean of the Gaussian process model prediction (representing the expected score), σ(X) represents the standard deviation of the Gaussian process model prediction (representing uncertainty), and κ represents the balance factor; in this embodiment, κ is configured as 1.5.

[0201] During the screening process, the GPR model evaluates two candidate points A and B in the space based on the data from the first 5 points.

[0202] Candidate point A (using an exploitation strategy):

[0203] Parameters: X_A = {PTO=26000, Pretension=1200}; This point is located near the known highest score (0.862);

[0204] GP model prediction results:

[0205] μ(X_A) = 0.870 (high prediction score);

[0206] σ(X_A) = 0.015 (low uncertainty, due to the availability of surrounding data points);

[0207] Therefore, UCB(X_A) = 0.870 + 1.5×0.015 = 0.8925.

[0208] Candidate point B (using an exploratory strategy):

[0209] Parameters: X_B = {PTO=32000, Pretension=850}; This point is in a sparse data region;

[0210] GP model prediction results:

[0211] μ(X_B) = 0.855 (prediction score slightly lower than A);

[0212] σ(X_B) = 0.040 (High uncertainty, not yet fully explored);

[0213] Therefore, UCB(X_B) = 0.855 + 1.5×0.040 = 0.9150.

[0214] Decision Result: Since UCB(X_B) > UCB(X_A), candidate point B is selected as the 6th test point based on the Bayesian algorithm calculation results. This demonstrates the intelligence of Bayesian optimization: although the expected score of point B is slightly lower, its high uncertainty means that a better solution may be hidden in this region, which is worth exploring. (See [link to relevant documentation]). Figure 4 The GP model's mean prediction results are shown.

[0215] S304: Perform simulation and update the agent model.

[0216] Specifically, a fully coupled numerical simulation is performed on the selected new test point to obtain the true comprehensive score of the test point, and the test point and the corresponding true score are added to the dataset to update the GPR proxy model.

[0217] For example, a fully coupled numerical simulation is performed on X_B = {32000, 850}, and its true comprehensive score f(X_B) = 0.898 is obtained, which is a new optimal point; the data pair ({32000, 850}, 0.898) is added to the dataset, and the GPR surrogate model is updated.

[0218] S305: Iteratively select the next test point based on the updated surrogate model until the algorithm converges and finds the optimal solution.

[0219] Specifically, S303 and S304 are repeated until the algorithm converges. For example, after 30 iterations (including the initial 5 points), the algorithm converges and determines the optimal solution. The key test points are shown in Table 2.

[0220] Table 2:

[0221] Test number PTO damping (Ns / m) Pretension (kN) Overall score Cumulative Optimum Testing methods 1 18000 650 0.7423 0.7423 Initial sampling 5 38000 825 0.7891 0.8616 Initial sampling 12 26500 1080 0.8502 0.8616 BO Recommendation 16 24200 1150 0.8723 0.8723 BO Recommendation 21 27800 1025 0.8891 0.8891 BO Recommendation 26 28100 1075 0.9125 0.9125 BO Recommendation 30 27950 1090 0.9089 0.9125 BO Recommendation

[0222] As can be seen from Table 2, the optimal solution is:

[0223] Overall score: 0.9125;

[0224] PTO damping: 28,100 Ns / m;

[0225] Pretension: 1,075 kN;

[0226] Comparison method: Traditional grid search, if the two parameters are divided into 10 steps, requires 10*10 = 100 fully coupled numerical simulations;

[0227] The key performance indicator is the number of tests required to reach the target threshold (0.85 points):

[0228] According to simulation data, the traditional grid method requires about 61 simulations, while the Bayesian optimization method can reach the optimal solution in only 3 simulations (Test 3 is the first time the target is met), which improves efficiency by 20.3 times and the improvement effect is obvious. When finding a solution close to the global optimum (>0.91), the traditional grid method requires 100 complete scans, while the Bayesian optimization only requires 26, reducing the number of tests by 74% and the reduction effect of test volume is obvious.

[0229] Figure 5 The convergence curves shown clearly demonstrate the fast convergence characteristics of the BO method. Figure 6The parameter space exploration trajectory is shown, demonstrating how the BO algorithm intelligently focuses on high-resolution regions. Figure 7 This paper presents a comparison of the efficiency of the Bayesian optimization method and the grid method used in the embodiments of this application; from Figures 5 to 7 As can be seen, by employing the Bayesian optimization method, the embodiments of this application can find an excellent design scheme that is close to the global optimum with less than 1 / 3 of the computational cost of the traditional method, which greatly improves the optimization efficiency.

[0230] S104: Use LightGBM to extract the model quantification comprehensive performance evaluation results, optimize the weight allocation of core indicators, and evaluate the comprehensive score of the recommended scheme.

[0231] In practical applications, the LightGBM extraction model, built on the gradient boosting decision tree model framework, can be used to comprehensively evaluate the feature importance of core indicators, so as to optimize the weight allocation of the core indicators corresponding to each recommendation scheme and obtain the corresponding weight allocation information.

[0232] In one embodiment, such as Figure 8 As shown, the step of using LightGBM to extract the model quantification comprehensive performance evaluation results, optimizing the weight allocation of core indicators, and evaluating the comprehensive score of the recommended scheme, i.e., S104, can include S801 to S804. The following detailed description of S801 to S804 is provided in conjunction with specific embodiments.

[0233] S801: Based on the proportion of the benefits of each recommended solution in the cost expenditure, determine the corresponding benefit-cost ratio, and use the benefit-cost ratio as the ultimate evaluation index of the corresponding recommended solution.

[0234] In practical applications, LightGBM can be introduced to address the subjectivity of weight allocation in traditional evaluation methods.

[0235] In practical applications, data preprocessing and the construction of final evaluation indicators can be carried out first.

[0236] Here, the original performance data matrix generated by S101, which contains multiple candidate design schemes, is first used as input data and preprocessed.

[0237] In practical applications, to effectively supervise model training when constructing the ultimate evaluation index, a semi-objective "ultimate evaluation index" y_ultimate can be calculated for each design scheme in the database based on the first principles of engineering economics. Specifically, this ultimate evaluation index can be defined as the benefit-cost ratio, abbreviated as BCR, and its calculation formula is as follows:

[0238] BCR = Benefit / (Cost_CAPEX / L + Cost_OPEX);

[0239] Where L represents the system design life (e.g., preset to 20 years), Benefit represents the quantitative result of the benefit item, Cost_CAPEX / L represents the quantitative result of annualized capital expenditure, and Cost_OPEX represents the quantitative result of operation and maintenance expenditure.

[0240] In practical applications, during the quantification of benefits, benefits are defined as primarily derived from annual power generation; the specific calculation formula is as follows:

[0241] Benefit = k_p * P_avg * (1 + k_sy * η_synergy);

[0242] Among them, P_avg (Dimension 1 indicator) represents the average total output power, which directly determines the basic power generation revenue; η_synergy (Dimension 3 indicator) represents the motion synergy effect (sway reduction rate), which serves as a gain coefficient to improve long-term benefits; here, more stable motion means a longer effective working time; k_p represents the power economy coefficient, which can be configured to 10,000 yuan / (kW·year); k_sy represents the synergy effect gain coefficient, which can be configured to = 0.10.

[0243] In practical applications, during the annualized capital expenditure quantification process, i.e., the Cost_CAPEX quantification process, capital expenditure (initial investment) is defined as being primarily determined by the construction costs of the structure, mooring, and anchoring systems; the calculation formula is as follows:

[0244] Cost_CAPEX = C_base_capex + k_t * T_max + k_o * O_max;

[0245] Among them, T_max (dimensional 4 index) represents the maximum mooring tension, which directly determines the specifications and cost of mooring cables, connectors, and anchoring systems; O_max (dimensional 5 index) represents the platform's ultimate offset, which affects the length of the dynamic cable and the coverage of the mooring system; C_base_capex represents the basic investment, which can be configured to 50 × 10^6 yuan; k_t represents the tension cost coefficient, which can be configured to 5000 yuan / kN; k_o represents the offset cost coefficient, which can be configured to 100,000 yuan / m.

[0246] In practical applications, during the quantification of operation and maintenance expenses, i.e., the Cost_OPEX quantification process, operation and maintenance expenses can be defined as being related to the system's daily wear and tear, fatigue damage, and failure risk; the calculation formula is as follows:

[0247] Cost_OPEX = C_base_opex + k_s * σ_pitch + k_f / FRI;

[0248] Among them, σ_pitch (dimensional 2 metric) represents the platform's motion stability (pitch standard deviation). The more intense the motion, the higher the structural fatigue and maintenance costs. FRI (dimensional 6 metric) represents the fault robustness index. The lower the FRI, the greater the potential major maintenance costs or downtime losses. Specifically, Cost_OPEX is inversely proportional to FRI. C_base_opex represents the basic operating cost, which can be configured to 5 × 10^6 yuan / year. k_s represents the motion cost coefficient, which can be configured to 500,000 yuan / °. k_f represents the fault risk cost coefficient, which can be configured to 2 × 10^6 yuan.

[0249] The above model integrates all six performance metrics into the BCR calculation, ensuring that the ultimate evaluation metric fully reflects the overall performance of the design scheme; while the intermediate evaluation metric will serve as a label for subsequent machine learning model training.

[0250] S802: The normalized scores of the six core indicators in each recommendation scheme are used as input features, and the final evaluation indicators of the corresponding recommendation scheme are used as output labels. The gradient boosting decision tree model is trained under supervision to generate the LightGBM extraction model.

[0251] Specifically, the original test metrics of the six performance dimensions are used as the input features X_feature of the machine learning model, as follows:

[0252] X1 = P_avg (power);

[0253] X2 = σ_pitch (motion stability);

[0254] X3 = η_synergy (synergistic effect);

[0255] X4 = T_max (mooring tension);

[0256] X5 = O_max (limit offset);

[0257] X6 = FRI (Fault Robustness).

[0258] Then, each original feature value is converted into a normalized score Score_i between 0 and 1 through the scoring function preset by S102, and the gradient boosting decision tree model is trained, that is, the scoring model is trained.

[0259] Specifically, X_feature is used as the input to the decision tree model, and the ultimate evaluation metric y_ultimate (i.e. BCR) is used as the output label of the model. Supervised learning training is performed on the gradient boosting decision tree model, that is, the LightGBM extraction model is generated.

[0260] It should be noted that the results of each test in S103 can be used to optimize the LightGBM extraction model with weight assignment in S104.

[0261] S803: Based on the feature importance of each core indicator in the LightGBM extraction model, optimize the weight allocation of the core indicators.

[0262] Here, after the scoring model has been trained, its built-in feature importance analysis function can be directly called from the trained model to analyze its objective weight coefficients.

[0263] Table 3:

[0264] Performance dimension Feature importance (original) Normalized weights W_i Physical explanation Dimension 1: Total Output Power 0.284 0.28 The greatest impact on the economy Dimension 4: Maximum mooring tension 0.218 0.22 Determines the main costs of CAPEX Dimension 2: Platform Motion Stability 0.162 0.16 Affecting OPEX and fatigue Dimension 5: Platform Limit Offset 0.138 0.14 Impact on CAPEX and safety margin Dimension 3: Motion Synergy 0.104 0.10 Long-term benefits Dimension 6: Fault Robustness 0.094 0.10 Risk cost contribution

[0265] S804: Evaluate the overall score of the recommended solution based on the weight allocation of the optimized core indicators.

[0266] Here, the objective weight allocation results obtained in Table 3 are as follows: Figure 9 As shown.

[0267] In practical applications, the overall score of the recommendation scheme can be obtained by weighting the normalized scores of each core indicator based on the weight allocation of the optimized core indicators.

[0268] Here, the overall score is calculated based on the weighting results:

[0269] S_total = 0.28*P_score + 0.22*T_score + 0.16*S_motion_score +

[0270] 0.14*O_score + 0.10*C_motion_score + 0.10*FRI_score;

[0271] Here, for any design scheme to be evaluated, its final comprehensive performance score S_total is calculated using this weighted summation formula.

[0272] The above weight allocation results are verified and evaluated below with reference to specific application examples.

[0273] The following is a complete evaluation of these three candidate solutions; the specific evaluation process includes S1 to S5.

[0274] S1: Raw performance data acquisition; the data here is obtained through fully coupled numerical simulation.

[0275] The original performance data of the three candidate schemes are shown in Table 4:

[0276] Table 4:

[0277] plan P_avg(kW) σ_pitch(°) Reduced speed (%) T_max(kN) O_max(m) R_motion R-mooring FRI Option A 6800 5.8 25 8500 55 2.8 1.9 0.41 Option B 5500 3.2 35 4800 38 1.3 1.2 0.77 Option C 6500 4.0 32 5800 45 1.5 1.4 0.74

[0278] Solution Features Description:

[0279] Option A: High-power type (highest P_avg), but with serious safety defects (T_max exceeds the limit, FRI fails to meet the standard);

[0280] Option B: High safety type (excellent safety indicators), but low power generation capacity;

[0281] Option C: Balanced optimization (balanced performance across all dimensions).

[0282] S2: Normalized scoring; specifically, the scoring function of S102 is applied for scoring.

[0283] The scoring process is illustrated using Scheme A as an example:

[0284] P_score_A = (6800-5000) / (7000-5000) = 0.900;

[0285] S_motion_score_A = 1.0 - 5.8 / 6.0 = 0.033 (close to failing);

[0286] C_motion_score_A = 25 / 40 = 0.625;

[0287] T_score_A (segmented calculation):

[0288] T_max=8500 kN, in the second segment (6000 < 8500 ≤ 10000);

[0289] T_score_A = 0.8*(10000-8500) / (10000-6000) = 0.300 (significantly low);

[0290] O_score_A = 1.0 - 55 / 80 = 0.313;

[0291] FRI_A = 1 / (0.6*2.8 + 0.4*1.9) = 1 / 2.44 = 0.410;

[0292] FRI_score_A = 0.410 (Unsatisfactory).

[0293] Using the same method, the normalized scores of the three schemes were calculated respectively, and the score matrices are shown in Table 5:

[0294] Table 5:

[0295] plan P_score) S_motion C_motion T_score O_score FRI_score Option A 0.900 0.033 0.625 0.300 0.313 0.410 Option B 0.250 0.467 0.875 0.900 0.525 0.770 Option C 0.750 0.333 0.800 0.700 0.438 0.740

[0296] Figure 10 The performance comparison results of the three candidate schemes A, B, and C across six core metrics are shown.

[0297] S3: Calculation of BCR ultimate evaluation index.

[0298] Taking Option A as an example:

[0299] Benefit_A = 10000 * 6800 * (1 + 0.10 * 0.25) = 69,700,000 yuan / year;

[0300] Cost_CAPEX_A = 50,000,000 + 5000*8500 + 100000*55 = 98,000,000 yuan;

[0301] Annualized CAPEX_A = 98,000,000 / 20 = 4,900,000 yuan / year;

[0302] Cost_OPEX_A = 5,000,000 + 500,000*5.8 + 2,000,000 / 0.41 = 12,778,048 yuan / year;

[0303] BCR_A = 69,700,000 / (4,900,000 + 12,778,048) = 4.00.

[0304] Using the same method, the BCR of the three schemes was calculated respectively, and the comparison of the calculation results is shown in Table 6:

[0305] Table 6:

[0306] plan Annual income (ten thousand yuan) Annual CAPEX (in ten thousand yuan) Annual OPEX (in ten thousand yuan) BCR Option A 6970 488 1207 4.00 Option B 5693 379 1060 3.96 Option C 6708 419 902 5.08

[0307] S4: Weighted composite score calculation; that is, weighting using the weight allocation results obtained from S104.

[0308] The calculation formula is expressed as follows:

[0309] S_total_A = 0.28*0.900 + 0.22*0.300 + 0.16*0.033 +0.14*0.313 + 0.10*0.625 + 0.10*0.410 = 0.4765;

[0310] S_total_B = 0.28*0.250 + 0.22*0.900 + 0.16*0.467 +0.14*0.525 + 0.10*0.875 + 0.10*0.770 = 0.6311;

[0311] S_total_C = 0.28*0.750 + 0.22*0.700 + 0.16*0.333 +0.14*0.438 + 0.10*0.800 + 0.10*0.740 = 0.6388;

[0312] The final evaluation results and rankings are generated based on the combined scores of the three options, as shown in Table 7:

[0313] Table 7:

[0314] Serial Number plan Total score S_total BCR Rating Recommended level 1 Option C 0.6388 5.08 qualified Highly recommended 2 Option B 0.6311 3.96 qualified Acceptable 3 Option A 0.4765 4.00 Unqualified Not recommended

[0315] S5: Analysis of Evaluation Conclusions

[0316] Option C (balanced optimization type) is the optimal option, with the highest overall score (0.6388), the best BCR (5.08), and the best economic efficiency.

[0317] It exhibits balanced performance across six dimensions with no obvious weaknesses and excellent fault robustness (FRI=0.74).

[0318] Although Option B (high safety type) has excellent safety performance (T_score=0.90, FRI=0.77), its power generation is relatively low (only 5500 kW), and there is room for improvement in economic efficiency. It is suitable for special scenarios with extremely high safety requirements.

[0319] Although Option A (high-power type) has the highest power output (6800 kW), its mooring tension score is only 0.30 (T_max=8500kN, which is close to the breaking load), resulting in serious safety deficiencies and unqualified fault robustness (FRI=0.41 < 0.5). The platform motion stability is also extremely poor (σ_pitch=5.8°, which is almost at the limit). Therefore, this option sacrifices unacceptable safety risks for high power generation, posing significant safety hazards and should not be adopted.

[0320] S105: Make optimization decisions based on visualization and intelligent diagnosis to determine whether the solution needs to be optimized directly.

[0321] In practical applications, it can be determined whether the overall score of the recommended solution obtained by S104 meets the standard, specifically whether it reaches the overall score threshold (e.g., 0.6), and whether the performance of each dimension is balanced; if it does not meet the standard, it is determined that the solution needs to be optimized.

[0322] In practical applications, whether the performance of each dimension is balanced can be determined by plotting the scores of the six dimensions on the same radar chart and by determining how close the shape of the radar chart is to a regular hexagon.

[0323] S106: If not required, determine the target design scheme based on the comprehensive score of the recommended scheme.

[0324] In practical applications, when the system is configured to directly select a solution, i.e., when iterative optimization is not required, the recommended solution with the highest comprehensive score can be directly selected as the target design solution.

[0325] S107: If necessary, design improvements to form new candidate solutions and retest the smart parameters.

[0326] In practical applications, when a solution needs to be iteratively optimized, a new design solution can be formed based on the visualization and intelligent diagnostic results, and then returned to S103 as a candidate solution for intelligent parameter testing.

[0327] In one embodiment, such as Figure 11 As shown, the design improvements that form new candidate solutions may include:

[0328] S1101: Use the LightGBM diagnostic model to predict and evaluate each recommended solution, and obtain the evaluation results;

[0329] S1102: Analyze and diagnose the evaluation results of each recommended solution using model interpretation tools to determine the performance bottleneck information of each recommended solution;

[0330] S1103: Based on the performance bottleneck information, optimize the configuration of the corresponding recommended solution to obtain the corresponding optimization solution;

[0331] S1104: Treat all optimized solutions as new candidate solutions and return to S103.

[0332] Here, we return to S103, that is, return to S103 and retest with the new candidate solution.

[0333] In one embodiment, such as Figure 12As shown, the step of using model interpretation tools to analyze and diagnose the evaluation results of each recommended solution and determine the performance bottleneck information of each recommended solution, i.e., S1102, may include:

[0334] S1201: Call the SHAP interpreter to perform attribution analysis and contribution quantification on the evaluation results of each recommended solution, and determine the corresponding performance bottleneck information;

[0335] S1202: Based on the performance bottleneck information, identify the bottleneck indicators with performance bottlenecks in the core indicators corresponding to each recommended solution, and generate a visual diagnostic report for each recommended solution; wherein, the visual diagnostic report includes the basic information of the bottleneck indicators in the corresponding solution and the reasons for their occurrence.

[0336] In practical applications, the recommended solutions can be ranked based on the comprehensive score S_total of each solution in the evaluation results, and a radar chart can be drawn for visualization.

[0337] In practical applications, the six-dimensional scores of all candidate solutions are plotted on the same radar chart, such as... Figure 10 As shown, radar charts can be used to visually compare the performance balance of different schemes (e.g., whether the shape is close to a regular hexagon), the performance advantages and disadvantages of different schemes (e.g., area size comparison), and identify performance shortcomings (e.g., concave areas of the shape).

[0338] In practical applications, in S1202, a LightGBM diagnostic model can be trained using a database containing 500 design schemes and their S_total scores, with the following configuration:

[0339] Input: Design parameters (such as PTO damping, mooring type, WEC diameter, number of buoys, etc.);

[0340] Output: Predicted S_total composite score.

[0341] Model performance: R²=0.92 on the test set, indicating that it can accurately learn the relationship between design parameters and final performance, and can be used for diagnosis.

[0342] Then, select the diagnostic solution and call the SHAP interpreter.

[0343] In practical applications, you can first select the solution with the lowest overall performance score among all recommended solutions as the diagnostic target;

[0344] For example, the solution A (S_total=0.4765) with the lowest overall score in the aforementioned embodiments is selected as the diagnostic target.

[0345] Complete design parameters for Scheme A:

[0346] PTO damping = 35,000 Ns / m;

[0347] Mooring configuration = chain-polyester-chain;

[0348] WEC diameter = 8.0 m;

[0349] Number of pontoons = 3.

[0350] The diagnostic process includes:

[0351] First, input this set of parameters into the trained LightGBM diagnostic model;

[0352] Secondly, the model's prediction score was 0.476 (which is very close to the true value of 0.4765, verifying the model's reliability).

[0353] Third, call "shap.TreeExplainer" to interpret the model's prediction for this time.

[0354] The SHAP interpreter calculates the average prediction score (benchmark) for all schemes in the database: E[f(x)] = 0.810 (benchmark).

[0355] For the prediction of Scheme A, SHAP calculates the specific contribution value (SHAP Value) of each design parameter; the SHAP value decomposition is shown in Table 8:

[0356] Table 8:

[0357] Design parameters SHAP value Contribution type degree of impact Physical explanation Mooring type = chain - polyester - chain -0.178 Negative contribution (pulling down) high This mooring configuration performs extremely poorly under cable breakage conditions, resulting in a severely low FRI (Free Rate Indicator). PTO damping = 35000 -0.111 Negative contribution (pulling down) high Excessive damping reduces power output and motion stability. WEC diameter = 8.0m -0.054 Negative contribution (pulling down) middle The diameter is too small, resulting in insufficient synergistic effect. Number of pontoons = 3 0.026 Positive contribution (driving up) Low This configuration has a slight positive contribution. Other parameters -0.017 Negative contribution (pulling down) Low Combined effect of other minor parameters

[0358] Here, the physical meaning of the SHAP value is:

[0359] Positive value: This parameter contributes to improving the score;

[0360] Negative value: This parameter lowers the score and is a performance bottleneck;

[0361] Absolute value: Degree of influence.

[0362] Figure 14 The SHAP waterfall diagram is shown, according to Figure 14 It can be seen how each parameter gradually increases / decreases from the baseline value of 0.810 to the final predicted value of 0.476.

[0363] Based on the above SHAP value decomposition results, the three design parameters with the lowest SHAP values ​​are taken as the main negative contribution parameters, the corresponding key negative contribution indicators are determined, and performance bottleneck information is generated.

[0364] For example, let's take solution A as an example:

[0365] Option A (high-power type) current score: 0.4765;

[0366] The three design parameters with the lowest SHAP values ​​are identified as the main negative contributors, i.e., performance bottlenecks; the main negative contributors include:

[0367] Inappropriate mooring configuration: -0.178;

[0368] PTO damping is too high: -0.111;

[0369] WEC diameter is too small: -0.054.

[0370] Based on the main negative contributions, key negative contribution indicators are determined, that is, the core indicators corresponding to each design parameter. Figure 13 The radar chart shown also reveals that the performance shortcomings of Scheme A are:

[0371] Mooring tension score: 0.300 (unacceptable);

[0372] Fault robustness score: 0.410 (unacceptable, FRI=0.41 < 0.5);

[0373] Movement stability score: 0.033 (severely low).

[0374] In S1202, based on performance bottleneck information, the corresponding key negative contribution indicators, i.e. bottleneck indicators, can be determined, and corresponding visual diagnostic reports can be generated.

[0375] In one embodiment, optimizing the configuration of the corresponding recommended solution based on the performance bottleneck information to obtain the corresponding optimized solution, i.e., S1103, may include:

[0376] Using the SHAP interpreter, optimization strategies for each bottleneck metric are generated based on the visualized diagnostic report;

[0377] Based on the preset priority of each core indicator, the optimization strategies corresponding to each bottleneck indicator are sorted, and optimization suggestions for the corresponding recommended solutions are generated.

[0378] In practical applications, the visual diagnostic report can also include optimization suggestions, that is, the diagnostic report can include both problem analysis and optimization suggestions.

[0379] The following example, using Solution A, illustrates the visualization diagnostic report.

[0380] Option A generates an intelligent diagnostic report, which includes two parts: analysis and optimization.

[0381] The analysis section includes basic information, SHAP value diagnostic results, performance bottleneck analysis, root cause analysis of optimized distribution, and optimization suggestions.

[0382] (1) Basic information includes:

[0383] Solution Name: Solution A (High Power Type);

[0384] Current score: 0.4765;

[0385] Diagnostic method: SHAP value attribution analysis.

[0386] (2) Diagnostic results of SHAP value include:

[0387] Database average score: 0.810;

[0388] Major negative contribution (performance bottleneck):

[0389] Inappropriate mooring configuration: -0.178;

[0390] PTO damping is too high: -0.111;

[0391] WEC diameter is too small: -0.054.

[0392] (3) Performance bottleneck analysis; see Figure 13 The radar image shown;

[0393] Mooring tension score: 0.300 (unacceptable);

[0394] Fault robustness score: 0.410 (unacceptable, FRI=0.41 < 0.5);

[0395] Movement stability score: 0.033 (severely low).

[0396] (4) Root cause analysis:

[0397] a. Mooring configuration (-0.178, maximum bottleneck):

[0398] The current method uses a "chain-polyester-chain" combination;

[0399] In the event of a cable breakage, the remaining cable's load-bearing capacity is insufficient, leading to:

[0400] R_mooring = 1.9 (tension increased by 90%)

[0401] R_motion = 2.8 (motion amplitude increased by 180%)

[0402] The final FRI was 0.41 (unacceptable).

[0403] b. PTO damping (-0.111, second bottleneck):

[0404] The current value of 35000 Ns / m is too high;

[0405] This results in high power generation, but a shift in the system's resonant frequency and a deterioration in motion response.

[0406] The σ_pitch is as high as 5.8°, almost reaching the safety limit of 6.0°.

[0407] c. WEC diameter (-0.054, third bottleneck):

[0408] The current value of 8.0m is too small;

[0409] The synergistic anti-sway effect is insufficient, with an anti-sway rate of only 25% (target 40%).

[0410] (5) Optimization suggestions (sorted by priority)

[0411] High priority: Change the mooring type to "semi-tensioned".

[0412] Theoretical basis: Historical data shows that the average FRI of semi-tensioned cable under cable breakage conditions is 0.72 (far exceeding the current 0.41).

[0413] Expected improvement: +0.119 points;

[0414] Implementation difficulty: Medium (requires redesign of mooring layout).

[0415] Medium priority: Reduce PTO damping to the 25000-30000 range.

[0416] Theoretical basis: SHAP analysis shows that the average SHAP value of historical schemes within this interval is +0.03 (positive contribution).

[0417] Expected improvement: +0.051 points;

[0418] Implementation difficulty: Low (only control parameters need to be adjusted).

[0419] Low priority: Increase the WEC diameter to 10-12m.

[0420] Theoretical basis: A larger WEC can provide a stronger anti-roll effect;

[0421] Expected improvement: +0.025 points;

[0422] Implementation difficulty: High (affects structural design).

[0423] Figure 15 The radar chart showing the optimized performance metrics of scheme A is presented. Figure 15It can be seen that the optimized solution A is expected to have a comprehensive score of 0.6716, according to Figure 13 and Figure 15 It can be seen that the optimized scheme A improved the overall score by +0.1951, and the FRI score was greater than 0.6, reaching the qualified level.

[0424] Furthermore, this application embodiment also simulates the impact of different mooring forms on the score using a diagnostic proxy model to verify parameter sensitivity; the verification results are as follows:

[0425] Chain-Polyester-Chain (current): 0.476;

[0426] Semi-tensioned (recommended): 0.825 (increase by 0.349);

[0427] Gravity anchor: 0.810 (lift 0.334).

[0428] Here, by revealing the performance bottlenecks of the solution, making targeted design improvements, forming new candidate solutions, and returning to S103 for another round of testing and evaluation, we can improve efficiency while achieving closed-loop iterative optimization of the design, thereby improving the accuracy and reliability of the test and evaluation results.

[0429] This application embodiment constructs a closed-loop intelligent "test-evaluation-diagnosis" system architecture that organically combines Bayesian optimization, gradient boosting decision trees, and SHAP value analysis. This integrates the traditionally separate steps into a complete technical closed loop from data generation to objective evaluation and then to intelligent diagnosis.

[0430] Secondly, by introducing an objective weight allocation method based on machine learning, this application can automatically learn from simulation data and output the "feature importance" that reflects the contribution of each performance index to the final value of the system, and use this as the weight coefficient, thus solving the subjective problem of traditional weight allocation and improving the objectivity and rationality of weight allocation.

[0431] Third, the embodiments of this application introduce model interpretation tools (such as SHAP) to perform attribution analysis on the prediction results of the performance evaluation model, thereby quantifying the contribution of each input design parameter to the performance bottleneck and providing precise guidance for optimization.

[0432] Fourth, by constructing a six-dimensional parallel testing framework that includes cable breakage fault conditions, performance testing is expanded from a single dimension to a comprehensive evaluation system covering six dimensions: power generation efficiency, motion stability, synergistic effect, structural safety, position maintenance, and fault redundancy.

[0433] Fifth, a quantitative comprehensive performance evaluation model based on normalized scoring and weighted summation was constructed for scientific optimization decision-making. By transforming multiple test indicators with different dimensions and importance into a single, quantifiable comprehensive score and supplementing it with radar charts for visualization, the subjective problem in multi-objective decision-making was solved.

[0434] Sixth, by introducing a quantitative test of the "synergistic effect of WEC and platform motion", based on the test of total power generation, an innovative quantitative test of the "synergistic vibration suppression" effect of WEC is proposed, which is used as an important dimension for evaluating the internal coupling benefits of the system. Specifically, by comparing the platform motion response with and without wave energy devices, the suppression effect of wave energy devices on platform motion is quantified, and it is used as an independent performance evaluation index to improve the reliability of system test results.

[0435] In summary, the performance testing and optimization method for the wind-wave hybrid energy harvesting system provided in this application has the following advantages:

[0436] (1) From “subjective evaluation” to “intelligent scoring”, decision-making is more scientific: The embodiments of this application replace manually set formulas and weights with machine learning models trained on historical data, thereby eliminating subjectivity and improving the objectivity and reliability of evaluation results.

[0437] (2) From “blind testing” to “intelligent optimization”, efficiency is improved exponentially: By using Bayesian optimization to replace parameter scanning, the number of tests required to find the optimal solution is reduced by more than 70% (e.g., from 100 times to 30 times in the example), which greatly shortens the R&D cycle;

[0438] (3) From “reading pictures and describing” to “automatic diagnosis”, the optimization is more accurate: By introducing the SHAP model interpretation tool, the present invention can automatically locate performance bottlenecks and provide data-driven optimization suggestions (such as the diagnostic report in the embodiment), so that the optimization process no longer depends on the personal experience of engineers, thereby improving the reliability of optimization results;

[0439] (4) More comprehensive risk control: By innovatively introducing and quantifying the "Fault Robustness Index (FRI)," the consideration of sudden failures is systematically integrated into the evaluation system, which can identify and avoid design schemes with "brittle" failure risks in advance, significantly improving system safety.

[0440] To implement the aforementioned performance testing and optimization method for wind-wave hybrid energy harvesting systems, this application also provides a performance testing and optimization device 1600 for wind-wave hybrid energy harvesting systems, such as... Figure 16 As shown, the device may include:

[0441] The first processing unit 1601 is used to establish a parameterized fully coupled virtual test model to form a design parameter database; substitute the test object to form a performance index database; and determine a six-dimensional comprehensive performance test matrix and a candidate scheme scoring method; the six-dimensional comprehensive performance test matrix includes six core indicators in two dimensions: power generation performance and safety performance.

[0442] The testing and evaluation unit 1602 is used to perform intelligent parameter testing using Bayesian optimization to select recommended solutions; and to use LightGBM to extract the model quantification comprehensive performance evaluation results, optimize the weight allocation of core indicators, and evaluate the comprehensive score of the recommended solutions.

[0443] The second processing unit 1603 is used to make optimization decisions based on visualization and intelligent diagnosis, and to determine whether the solution needs to be optimized. If not, the target design solution is determined based on the comprehensive score of the recommended solution. If it is needed, the design is improved to form a new candidate solution, and the intelligent parameter test is carried out again.

[0444] It should be noted that the performance testing and optimization device for the wind-wave hybrid energy harvesting system provided in the above embodiments is only illustrated by the division of the above-described program modules when performing performance testing and optimization of the wind-wave hybrid energy harvesting system. In practical applications, the above processing can be assigned to different program modules as needed, that is, the internal structure of the device can be divided into different program modules to complete all or part of the processing described above. In addition, the performance testing and optimization device for the wind-wave hybrid energy harvesting system provided in the above embodiments and the performance testing and optimization method embodiments for the wind-wave hybrid energy harvesting system belong to the same concept, and the specific implementation process is detailed in the method embodiments, which will not be repeated here.

[0445] It should be noted that terms such as "first" and "second" are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0446] Furthermore, the technical solutions described in the embodiments of this application can be combined arbitrarily without conflict.

[0447] The above description is merely a preferred embodiment of this application and is not intended to limit the scope of protection of this application.

Claims

1. A performance testing and optimization method for a wind-wave hybrid energy harvesting system, characterized in that, The method includes: Establish a parameterized, fully coupled virtual test model to form a design parameter database; substitute the test object to form a performance index database; A six-dimensional comprehensive performance test matrix and a candidate scheme scoring method were determined; the six-dimensional comprehensive performance test matrix includes six core indicators across two dimensions: power generation performance and safety performance. Intelligent parameter testing is performed using Bayesian optimization to select recommended solutions; LightGBM is used to extract the model quantification of the comprehensive performance evaluation results, optimize the weight allocation of core indicators, and evaluate the comprehensive score of the recommended solution. Optimization decisions are made based on visualization and intelligent diagnostics to determine whether direct optimization is needed. If not required, the target design scheme is determined based on the comprehensive score of the recommended scheme; if required, the design is improved to form a new candidate scheme, and the intelligent parameter test is carried out again.

2. The method according to claim 1, characterized in that, The design improvements lead to new candidate solutions, including: The LightGBM diagnostic model was used to predict and evaluate each recommended solution, and the evaluation results were obtained. The evaluation results of each recommended solution are analyzed and diagnosed using model interpretation tools to identify the performance bottleneck information of each recommended solution; Based on the performance bottleneck information, the configuration of the corresponding recommended solution is optimized to obtain the corresponding optimization solution; All optimized solutions are considered as new candidate solutions.

3. The method according to claim 2, characterized in that, The candidate solution scoring method is configured to include the normalized score of each core indicator in the corresponding solution and the comprehensive performance score of the corresponding solution; wherein, the comprehensive performance score is obtained by weighting the normalized scores of each core indicator.

4. The method according to claim 3, characterized in that, The process of analyzing and diagnosing the evaluation results of each recommended solution using model interpretation tools to determine the performance bottleneck information of each recommended solution includes: The SHAP interpreter is invoked to perform attribution analysis and contribution quantification on the evaluation results of each recommended solution, thereby identifying the corresponding performance bottleneck information. Based on the performance bottleneck information, the bottleneck indicators with performance bottlenecks in the core indicators corresponding to each recommendation scheme are identified, and a visual diagnostic report for each recommendation scheme is generated; wherein, the visual diagnostic report includes the basic information of the bottleneck indicators in the corresponding scheme and the causes of their occurrence.

5. The method according to claim 4, characterized in that, The step of optimizing the configuration of the corresponding recommended solution based on the performance bottleneck information to obtain the corresponding optimization solution includes: Using the SHAP interpreter, optimization strategies for each bottleneck metric are generated based on the visualized diagnostic report; Based on the preset priority of each core indicator, the optimization strategies corresponding to each bottleneck indicator are sorted, and optimization suggestions for the corresponding recommended solutions are generated.

6. The method according to any one of claims 1 to 5, characterized in that, The core indicators corresponding to the power generation performance dimension include total output power, platform motion stability, and motion synergy effect; the core indicators corresponding to the safety performance dimension include maximum mooring tension, platform extreme deviation, and fault robustness index.

7. The method according to claim 6, characterized in that, The fault robustness index is determined by quantifying the degree of performance degradation after failure; The process of determining the fault robustness index includes: Determine the motion deterioration ratio and the mooring deterioration ratio; Based on the motion degradation ratio and the mooring degradation ratio, the degree of system performance degradation under failure conditions is determined, and the fault robustness index is obtained.

8. The method according to claim 2, characterized in that, The process of using Bayesian optimization for intelligent parameter testing to select recommended solutions includes: Define the objective function and parameter space; Within the parameter space, an initial sample set containing n test points is generated by Latin hypercube sampling, and the comprehensive performance score is calculated to construct the initial GPR proxy model. Select the next test point based on the acquisition function; Perform the simulation and update the agent model; The algorithm iteratively selects the next test point based on the updated proxy model until it converges and finds the optimal solution.

9. The method according to claim 2, characterized in that, The process of using LightGBM to extract model quantification results, optimizing the weight allocation of core indicators, and evaluating the comprehensive score of the recommended solution includes: Based on the proportion of the benefits of each recommended solution in the cost expenditure, the corresponding benefit-cost ratio is determined, and the benefit-cost ratio is used as the ultimate evaluation index of the corresponding recommended solution. The normalized scores of the six core metrics in each recommendation scheme are used as input features, and the final evaluation metrics of the corresponding recommendation scheme are used as output labels. The gradient boosting decision tree model is trained under supervision to generate the LightGBM extraction model. Based on the feature importance of each core indicator in the LightGBM extraction model, the weight allocation of the core indicators is optimized. The overall score of the recommended solution is evaluated based on the optimized weight allocation of the core indicators.

10. A performance testing and optimization device for a wind-wave hybrid energy harvesting system, characterized in that, The device includes: The first processing unit is used to establish a parameterized fully coupled virtual test model to form a design parameter database; substitute the test object to form a performance index database; and determine the six-dimensional comprehensive performance test matrix and the candidate scheme scoring method; the six-dimensional comprehensive performance test matrix includes six core indicators in two dimensions: power generation performance and safety performance. The testing and evaluation unit is used to perform intelligent parameter testing using Bayesian optimization to select recommended solutions; and to use LightGBM to extract the model quantification comprehensive performance evaluation results, optimize the weight allocation of core indicators, and evaluate the comprehensive score of the recommended solutions. The second processing unit is used to make optimization decisions based on visualization and intelligent diagnosis, and to determine whether the solution needs to be optimized. If not, the target design solution is determined based on the comprehensive score of the recommended solution. If it is needed, the design is improved to form a new candidate solution, and the intelligent parameter test is carried out again.

Images