Offshore wind turbine generator reliability distribution modeling and maintenance decision-making method under longevity management

Abstract

The invention discloses an offshore wind turbine generator reliability distribution modeling and maintenance decision-making method under long-life management, and the method employs a multi-source heterogeneous data fusion and meta-transfer learning technology to improve the prediction accuracy and adaptability of the health condition and remaining service life of key parts of an offshore wind turbine generator. Based on the RCM theory and the system analysis component structure and influence factors, uncertainty in fuzzy theory processing evaluation is introduced, predicted health and degradation information is combined, an objective and quantitative reliability index distribution model of the key components is established, and the reasonable reliability index distribution weight of each key component is determined; according to the method, the maximum average reliability of an offshore wind turbine generator system and the minimum full-life-cycle operation and maintenance cost serve as targets, a long-life-period multi-target optimization model considering part degradation, the maintenance effect and multiple constraints is constructed, an intelligent algorithm is adopted for solving, and an optimal preventive maintenance plan is obtained.

Description

Technical Field

[0001] The present invention relates to the technical field of offshore wind turbine operation and maintenance, and in particular to a reliability allocation modeling and maintenance decision-making method for offshore wind turbines under long-life management. Background Art

[0002] As the global demand for renewable energy continues to increase, offshore wind power, as an important clean energy source, has received widespread attention for its development and utilization. However, offshore wind turbines currently have the following shortcomings during operation and maintenance:

[0003] 1. Inaccurate and delayed health prediction: Traditional models have difficulty accurately predicting early and complex faults and adapting to changing working conditions. In addition, high-precision sensors are expensive, which affects maintenance decisions.

[0004] 2. Subjective and rough reliability allocation: Existing allocation methods mostly rely on experience, ignoring complex factors and uncertainties such as the actual working conditions of components, fault characteristics, and predicted health trends, resulting in inaccurate allocation.

[0005] 3. Short-sighted and single-minded maintenance decisions: Existing decisions often ignore long-term benefits and the balance of multiple objectives (such as reliability and cost), and fail to effectively incorporate predictive information for forward-looking planning, resulting in insufficient optimization capabilities.

[0006] 4. Insufficient data value mining: Insufficient integration and utilization of massive heterogeneous operation and maintenance data, failure to effectively explore potential correlations and trends, and restricting the accuracy of prediction and decision-making models. Summary of the Invention

[0007] The purpose of the present invention is to overcome the deficiencies of the prior art and provide a reliability allocation modeling and maintenance decision-making method for offshore wind turbines under long-life management.

[0008] To achieve the above objectives, the technical solutions provided by the present invention are:

[0009] Reliability allocation modeling and maintenance decision-making methods for offshore wind turbines under long-life management, including:

[0010] S1. Predict the health status of target components of offshore wind turbines based on multi-source heterogeneous data fusion and meta-transfer learning;

[0011] S2. Build a reliability index allocation model based on RCM and fuzzy theory to identify key component failure modes and optimize the reliability index allocation of each component, and then allocate the reliability index to each key component of the offshore wind turbine;

[0012] S3. Based on the reliability index allocation result obtained in step S2, a system reliability model based on the minimum path set and the structure function is established;

[0013] S4, combining the health status of the target components of the offshore wind turbine predicted in step S1 with the system reliability model established in step S3, and performing multi-objective maintenance optimization decision-making for longevity;

[0014] S5. Perform multi-objective maintenance optimization based on the multi-objective maintenance optimization decision results, and perform subsequent closed-loop optimization.

[0015] Furthermore, step S1 includes:

[0016] S1-1. Data collection and preprocessing:

[0017] Collect multi-source heterogeneous data during the operation of offshore wind turbines, including SCADA data, CMS vibration signal data, gearbox oil monitoring data, blade condition monitoring data, pitch / yaw system data, meteorological data, power grid data, and historical maintenance / fault records;

[0018] The collected data is cleaned, normalized, and time-aligned;

[0019] S1-2, extracting features related to component health status from the preprocessed data;

[0020] S1-3. Build and train a multi-source heterogeneous data fusion model based on CNN-LSTM, and then obtain the health index HI of each component of the offshore wind turbine through the trained multi-source heterogeneous data fusion model based on CNN-LSTM;

[0021] S1-4. Construct a fault diagnosis and prediction model based on meta-transfer learning, and train the fault diagnosis and prediction model based on meta-transfer learning using the health index HI of each key component of the offshore wind turbine obtained in step S1-3. Then, the health status of the target component, i.e., the predicted value of the remaining service life and its confidence interval, is obtained through the trained fault diagnosis and prediction model based on meta-transfer learning.

[0022] Furthermore, the multi-source heterogeneous data fusion model based on CNN-LSTM constructed in step S1-3 utilizes parallel convolutional neural network (CNN) branches to process local spatial features or patterns from different sensors respectively; utilizes long short-term memory (LSTM) branches to process the temporal dependency features of data acquisition and supervisory control system (SCADA) data; and performs weighted fusion of the features extracted by the parallel convolutional neural network (CNN) and the features processed by the long short-term memory (LSTM) network in the fusion layer. The weights are obtained through network learning, and the fused features are input into the subsequent fully connected layer. Finally, the health index HI of the key components of the offshore wind turbine is output.

[0023] Furthermore, step S1-4 includes:

[0024] The long short-term memory network (LSTM) is used as the basic model for predicting the remaining useful life, and the input is the health index HI of the key components obtained in step S1-3;

[0025] Meta-training: The lifespan data of each offshore wind turbine or key component of the same type is treated as an independent "task." Within each task, the model parameters are updated using a small number of gradient steps. Between tasks, the model's initial parameters are optimized so that the model, starting from this initial state, can quickly achieve good performance on any new task using a small amount of data.

[0026] Meta-testing phase: For the target components of the target offshore wind turbine, a small amount of early operating data or labeled data is used to perform a small amount of gradient updates based on the "good initial parameters" obtained through meta-training to quickly adapt the model to the degradation pattern of the target components.

[0027] Output the predicted value of the remaining useful life of the target component and its confidence interval.

[0028] Furthermore, step S2 includes:

[0029] S2-1. Use the RCM Failure Mode, Effects, and Criticality (FMEC) analysis method, combined with historical failure data and expert experience, to identify the functions, potential failure modes, and impacts of each component of the offshore wind turbine on system performance, safety, and the environment. Based on the risk priority number from the FMEC analysis results, select the critical components that have the greatest impact on the overall reliability, safety, and maintenance costs of the offshore wind turbine.

[0030] S2-2, determine the reliability index allocation influencing factor set, identify the factors that affect the reliability index allocation of key components, and form the influencing factor set U = {u1, u2, ..., u i},u1,u2,...,u i All are influencing factors;

[0031] S2-3, perform fuzzy evaluation of reliability indicators and obtain the fuzzy evaluation matrix of each key component k;

[0032] S2-4, determine the reliability index allocation weight;

[0033] S2-5. Allocate reliability indicators to key components of offshore wind turbines based on reliability indicator weights.

[0034] Furthermore, step S2-3 includes:

[0035] S2-3-1. Establish an evaluation level set: define a fuzzy evaluation level set for the evaluation results of each influencing factor;

[0036] S2-3-2. Construct membership function: For each influencing factor, establish its membership function relative to the evaluation level v according to its quantitative data range or expert experience. j The membership function μ ui (v j );

[0037] S2-3-3, Single factor fuzzy evaluation: For each key component k, collect or calculate its index value under each influencing factor; use the membership function to determine the key component under factor u i Next, for each evaluation level v j The membership degree forms a single factor evaluation vector R ki :

[0038] R ki =[μ ui (v1) k ,μ ui (v2) k ,...,μ ui (v n ) k ]

[0039] S2-3-4. Construct a fuzzy evaluation matrix: Combine the single-factor evaluation vectors of all key components k with respect to all influencing factors to form a fuzzy evaluation matrix for each key component k [R k1 ; R k2 ;...;R km ].

[0040] Furthermore, step S2-4 includes:

[0041] S2-4-1. Determine the weight of factors: Use fuzzy analytic hierarchy process to build a fuzzy judgment matrix and compare the influencing factors u i For the relative degree of importance of reliability index allocation, the fuzzy weight vector W = [w1, w2, ..., w m ], where w i Representation factor u i The weight of

[0042] S2-4-2, fuzzy comprehensive evaluation: For each key component k, use its fuzzy evaluation matrix R k The fuzzy weight vector W of the factors is used to perform fuzzy comprehensive evaluation, using the weighted average model M(·,+), and the formula is:

[0043]

[0044] Get the comprehensive fuzzy evaluation vector B of each key component k k =[b k1 ,bk2 ,...,b kn ], where b kj Indicates that the comprehensive performance of key component k belongs to evaluation level v j The degree of membership;

[0045] S2-4-3, based on comprehensive fuzzy evaluation vector B k Determine the reliability index allocation weights, the process is as follows:

[0046] Defuzzification is used to obtain the score of each component:

[0047] Among them, S k represents the clarity score of key component k;

[0048] Calculate the sum of all key component scores:

[0049]

[0050] Among them, S total represents the sum of all key component scores;

[0051] Calculate the reliability index allocation weight of each key component:

[0052]

[0053] Among them, W k represents the reliability weight assigned to key component k, R total It is the overall reliability target of the system.

[0054] Furthermore, step S3 includes:

[0055] S3-1. Based on the reliability structure diagram of key components of offshore wind turbines, the adjacency matrix method is combined with the depth-first search algorithm to find the minimum path set P = {P1, P2, ..., P q}, where each P i is a set of key components; the reliability structure diagram of key components of offshore wind turbines is obtained by screening out the key component set that has the greatest impact on the overall reliability, safety and maintenance cost of the wind turbine, and analyzing the logical connection relationship of the key component set in the system;

[0056] S3-2, perform the minimum path set disjoint processing: Use the structure function matrix method to convert the minimum path set event {E(P i )} into a set of mutually disjoint events {D1, D2, ..., D r}, so that the event of system success is equivalent to the union of mutually disjoint events;

[0057] S3-3. Establish a system reliability model: Assume that the reliability of the Kth key component at time t is R K (t), then the reliability of the system at time t is R sys (t) By calculating the probability of mutually disjoint events, we can obtain:

[0058] R sys (t)=P(∪D i )=∑P(D i )

[0059] Among them, each P(D i ) represents the mutually disjoint events D i The key components and their connection relationships are included, and the reliability R of the key components at time t is used K (t) The calculated probability.

[0060] Furthermore, step S4 includes:

[0061] S4-1. Constructing a multi-objective optimization function:

[0062] S4-1-1. Objective 1: Maximize the average reliability of offshore wind turbine systems throughout their entire lifespan

[0063]

[0064] Where M is the maintenance strategy vector, which includes the maintenance decisions of each component at different time points; R sys (t,M) is the system reliability after considering the influence of maintenance activity M; T is the time period considered, that is, the longevity period;

[0065] S4-1-2. Objective 2: Minimize the total operation and maintenance costs throughout the entire longevity period T:

[0066] Min f2(M)=C total (T,M)=C pm (M)+C cm (M)+C loss (M)+C op (M)

[0067] Among them, C total is the total operation and maintenance cost, C pm is the planned maintenance cost, C cm is the cost of unplanned failure repair, C loss is the power generation loss cost of shutdown, C op For other operating costs;

[0068] S4-2. Define decision variables and constraints:

[0069] S4-2-1, Decision variables: M = {m k,i}, where m k,i represents the type of maintenance action taken for key component k at the i-th decision time point;

[0070] S4-2-2. Based on the prediction results of step S1, a degradation model for key components is established, specifically using a discrete-time Markov process based on the HI partition state; clarifying the effects of different maintenance actions on the health index HI state transition and the recovery of the remaining service life of the component;

[0071] S4-2-3. Constraints:

[0072] Total maintenance budget upper limit: ∑C(m k,i )≥Budget total , Budget total represents the total maintenance budget;

[0073] Maintenance resource constraints: number of available maintenance teams, availability of maintenance vessels, and spare parts inventory limitations;

[0074] Maintenance window constraints: The annual window for offshore operations, taking into account the influence of weather and sea conditions;

[0075] Operational constraints: minimum acceptable system reliability / availability threshold R sys (t)≥R min ;

[0076] Key component life constraints: mandatory replacement of key components when they reach their design life or their predicted remaining service life is exhausted;

[0077] S4-3, solving multi-objective optimization through improved non-dominated sorting genetic algorithm;

[0078] S4-3-1. Chromosome coding: A three-dimensional binary coding matrix is ​​used to represent the maintenance strategy, where the dimensions are [unit number × time interval × maintenance type];

[0079] S4-3-2, Fitness Assessment: Based on the effectiveness of the maintenance strategy, calculate the fitness of each individual, including the two objectives of system reliability and maintenance cost;

[0080] System reliability:

[0081]

[0082] Among them, λ(τ) is the failure rate function, reflecting the failure probability of the unit;

[0083] Repair costs:

[0084]

[0085] Among them, c pm is the preventive maintenance cost, c cm is the corrective maintenance cost, δ i With ξ j are the execution times of the corresponding maintenance tasks respectively; It is a resource constraint penalty item that handles the constraint violation of maintenance team and resource allocation;

[0086] S4-3-3, genetic manipulation;

[0087] S4-4. Output decision:

[0088] (1) Pareto front generation: After multiple generations of evolution, a set of non-dominated solutions is output, representing different maintenance strategies;

[0089] (2) Multi-dimensional decision support: Provide different types of decision support based on the needs of decision makers:

[0090] Cost-sensitive: choose the maintenance strategy with the lowest cost;

[0091] Risk-averse: choose the most reliable strategy;

[0092] Equilibrium decision point: Based on the entropy weight method, the optimal compromise solution is determined to balance cost and reliability;

[0093] (3) Maintenance plan output: Automatically generate maintenance work orders based on the selected maintenance strategy.

[0094] Furthermore, step S5 includes:

[0095] Decision execution: Convert the selected multi-objective maintenance optimization decision results into specific maintenance work orders, dispatch resources and execute preventive maintenance activities according to the optimization decision results;

[0096] Data update and model iteration: Collect information on actual maintenance activities, maintenance effect data, actual cost data, and subsequent equipment operation data; and feed this latest data back to steps S1 and S4;

[0097] Closed-loop optimization: Regularly rerun the entire process, using updated data and adopted models to dynamically adjust future reliability assessments, allocations, and maintenance decisions, achieving continuous improvement and adaptive optimization of operation and maintenance strategies.

[0098] Compared with the existing technology, the principles and advantages of this technical solution are as follows:

[0099] Utilize multi-source heterogeneous data fusion and meta-transfer learning technology to improve the accuracy and adaptability of predictions of the health status and remaining service life of key components of offshore wind turbines;

[0100] Based on RCM theory, we systematically analyze component structures and influencing factors, introduce fuzzy theory to deal with uncertainty in evaluation, and establish an objective and quantitative reliability index allocation model for key components based on predicted health and degradation information. This model also determines the appropriate reliability index allocation weights for each key component.

[0101] With the goal of maximizing the average reliability of offshore wind turbine systems and minimizing the full life cycle operation and maintenance costs, a long-life multi-objective optimization model is constructed that takes into account component degradation, maintenance effects and multiple constraints. An intelligent algorithm is used to solve the problem and obtain the optimal preventive maintenance plan.

[0102] This technical solution achieves a shift from passive maintenance to proactive preventive maintenance through accurate prediction, scientific allocation, and optimized decision-making, reducing unplanned downtime, lowering total operation and maintenance costs, and improving the reliability, availability, and overall economic benefits of offshore wind turbines. BRIEF DESCRIPTION OF THE DRAWINGS

[0103] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the services required for use in the embodiments or the prior art descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0104] Figure 1 This is a principle flow chart of the offshore wind turbine reliability allocation modeling and maintenance decision-making method under long-life management of the present invention;

[0105] Figure 2 This is the reliability structure diagram of key components of offshore wind turbines;

[0106] Figure 3 This is a principle flow chart of multi-objective maintenance optimization decision-making in the offshore wind turbine reliability allocation modeling and maintenance decision-making method under long-life management of the present invention. DETAILED DESCRIPTION

[0107] The present invention will be further described below in conjunction with specific embodiments:

[0108] like Figure 1 As shown, the reliability allocation modeling and maintenance decision-making method for offshore wind turbines under long-life management described in this embodiment includes the following steps:

[0109] S1. Predict the health status of target components of offshore wind turbines based on multi-source heterogeneous data fusion and meta-transfer learning;

[0110] The specific process of this step is as follows:

[0111] S1-1. Data collection and preprocessing:

[0112] Collect multi-source heterogeneous data during offshore wind turbine operation, including SCADA data (power, speed, temperature, voltage, current, etc.), CMS vibration signal data, gearbox oil monitoring data, blade condition monitoring data, pitch / yaw system data, meteorological data (wind speed, wind direction, temperature and humidity, etc.), power grid data, and historical maintenance / fault records. The collected data is then cleaned (denoising, filling missing values), normalized, and time-aligned.

[0113] S1-2. Extract features related to the health status of key components from preprocessed data; for example, extract time domain, frequency domain, and time-frequency domain features from vibration signals; extract statistical features and working condition-related features from SCADA data; extract wear particles and physical and chemical indicators from gearbox oil monitoring data.

[0114] S1-3. Build and train a multi-source heterogeneous data fusion model based on CNN-LSTM, and then obtain the health index HI of each key component of the offshore wind turbine through the trained multi-source heterogeneous data fusion model based on CNN-LSTM;

[0115] In this step (S1-3), the constructed multi-source heterogeneous data fusion model based on CNN-LSTM utilizes parallel convolutional neural network (CNN) branches to process local spatial features or patterns from different sensors respectively; utilizes long short-term memory (LSTM) branches to process the temporal dependency features of data acquisition and supervisory control system (SCADA) data; and performs weighted fusion of the features extracted by the parallel convolutional neural network (CNN) and the features processed by the long short-term memory (LSTM) network in the fusion layer. The weights are obtained through network learning, and the fused features are input into the subsequent fully connected layer. Finally, the health index HI of the key components of the offshore wind turbine is output.

[0116] S1-4. Construct a fault diagnosis and prediction model based on meta-transfer learning, and train the fault diagnosis and prediction model based on meta-transfer learning using the health index HI of each key component of the offshore wind turbine obtained in step S1-3. Then, the health status of the target component, i.e., the predicted value of the remaining service life and its confidence interval, is obtained through the trained fault diagnosis and prediction model based on meta-transfer learning.

[0117] The specific process of this step (S1-4) includes:

[0118] The long short-term memory network (LSTM) is used as the basic model for predicting the remaining useful life, and the input is the component health index HI obtained in step S1-3;

[0119] Meta-training: The lifespan data of each offshore wind turbine or key component of the same type is treated as an independent "task." Within each task, the model parameters are updated using a small number of gradient steps. Between tasks, the model's initial parameters are optimized so that the model, starting from this initial state, can quickly achieve good performance on any new task using a small amount of data.

[0120] Meta-testing phase: For the target components of the target offshore wind turbine, a small amount of early operating data or labeled data is used to perform a small amount of gradient updates based on the "good initial parameters" obtained through meta-training to quickly adapt the model to the degradation pattern of the target components.

[0121] Output the predicted value of the remaining useful life of the target component and its confidence interval. Use Monte Carlo Dropout technology to sample the Dropout layer multiple times during the neural network inference process to simulate uncertainty. This method allows the solution to randomly discard some neurons in each inference of the model, and after repeated sampling, obtain the mean and variance of the results. In this way, not only can the predicted value of the remaining useful life (RUL) be obtained, but also the corresponding confidence interval can be provided, thereby effectively quantifying the uncertainty of the prediction result. The component's health index HI and the remaining useful life prediction results (including degradation trend information) will serve as important inputs for reliability allocation and maintenance optimization decisions.

[0122] S2. Build a reliability index allocation model based on RCM and fuzzy theory to identify key component failure modes and optimize the reliability index allocation of each key component, and then allocate the reliability index to each key component of the offshore wind turbine;

[0123] The specific process of this step (S2) includes:

[0124] S2-1. Use the RCM Failure Mode, Effects, and Criticality (FMEC) analysis method, combined with historical failure data and expert experience, to identify the functions, potential failure modes, and impacts of each key component of the offshore wind turbine on system performance, safety, and the environment. Based on the risk priority number from the FMEC analysis results, select the set of key components that have the greatest impact on the overall reliability, safety, and maintenance costs of the offshore wind turbine.

[0125] S2-2, determine the reliability index allocation influencing factor set, identify the factors that affect the reliability index allocation of key components, and form the influencing factor set U = {u1, u2, ..., u i},u1,u2,...,u i All are influencing factors;

[0126] S2-3, perform fuzzy evaluation of reliability indicators and obtain the fuzzy evaluation matrix of each key component k;

[0127] The specific process of this step (S2-3) includes:

[0128] S2-3-1. Establish an evaluation level set: define a fuzzy evaluation level set for the evaluation results of each influencing factor;

[0129] S2-3-2. Construct membership function: For each influencing factor, establish its membership function relative to the evaluation level v according to its quantitative data range or expert experience. j The membership function μ ui (v j ); For example, for the failure rate, multiple thresholds can be set, and a trapezoidal membership function can be used to describe the degree to which the failure rate belongs to different levels such as "high", "medium", and "low".

[0130] S2-3-3, Single factor fuzzy evaluation: For each key component k, collect or calculate its index value under each influencing factor; use the membership function to determine the key component under factor u i Next, for each evaluation level v j The membership degree forms a single factor evaluation vector R ki :

[0131] R ki =[μ ui (v1) k ,μ ui (v2) k ,...,μ ui (v n ) k ]

[0132] S2-3-4. Construct a fuzzy evaluation matrix: Combine the single-factor evaluation vectors of all key components k with respect to all influencing factors to form a fuzzy evaluation matrix for each key component k [R k1 ; R k2 ;...;R km ].

[0133] S2-4. Determine the weight of reliability index allocation:

[0134] S2-4-1. Determine the weight of factors: Use fuzzy analytic hierarchy process to build a fuzzy judgment matrix and compare the influencing factors u i For the relative degree of importance of reliability index allocation, the fuzzy weight vector W = [w1, w2, ..., w m ], where w i Representation factor u i The weight of

[0135] S2-4-2, fuzzy comprehensive evaluation: For each key component k, use its fuzzy evaluation matrix R k The fuzzy weight vector W of the factors is used to perform fuzzy comprehensive evaluation, using the weighted average model M(·,+), and the formula is:

[0136]

[0137] Get the comprehensive fuzzy evaluation vector B of each key component k k =[b k1 ,b k2 ,...,b kn ], where b kj Indicates that the comprehensive performance of key component k belongs to evaluation level v j The degree of membership;

[0138] S2-4-3, based on comprehensive fuzzy evaluation vector B k Determine the reliability index allocation weights, the process is as follows:

[0139] Defuzzification is used to obtain the score of each component:

[0140] Among them, S k represents the clarity score of key component k;

[0141] Calculate the sum of all key component scores:

[0142]

[0143] Among them, S total represents the sum of all key component scores;

[0144] Calculate the reliability index allocation weight of each key component:

[0145]

[0146] Among them, W k represents the reliability weight assigned to key component k, R total It is the overall reliability target of the system.

[0147] S2-5. Allocate reliability indicators to key components of offshore wind turbines based on reliability indicator weights.

[0148] S3. Establish a system reliability model based on the minimum path set and structure function. The specific process includes:

[0149] S3-1, based on the reliability structure diagram of key components of offshore wind turbines, such as Figure 2As shown, the adjacency matrix method is combined with the depth-first search algorithm to find all the minimum path sets P={P1,P2,...,P q}, where each P i is a set of key components; the reliability structure diagram of key components of offshore wind turbines is obtained by screening out the key component set that has the greatest impact on the overall reliability, safety and maintenance cost of the wind turbine, and analyzing the logical connection relationship of the key component set in the system;

[0150] S3-2, perform the minimum path set disjoint processing: Use the structure function matrix method to convert the minimum path set event {E(P i )} into a set of mutually disjoint events {D1, D2, ..., D r}, so that the event of system success is equivalent to the union of mutually disjoint events;

[0151] S3-3. Establish a system reliability model: Assume that the reliability of the Kth key component at time t is R K (t), then the reliability of the system at time t is R sys (t) By calculating the probability of mutually disjoint events, we can obtain:

[0152] R sys (t)=P(∪D i )=∑P(D i )

[0153] Among them, each P(D i ) represents the mutually disjoint events D i The key components and their connection relationships are included, and the reliability R of the key components at time t is used K (t0 is the calculated probability.

[0154] S4, combining the health status of the target components of the offshore wind turbine predicted in step S1 to make a multi-objective maintenance optimization decision for longevity;

[0155] like Figure 3 As shown, this step (S4) specifically includes:

[0156] S4-1. Constructing a multi-objective optimization function:

[0157] S4-1-1. Objective 1: Maximize the average reliability of offshore wind turbine systems throughout their entire lifespan

[0158]

[0159] Where M is the maintenance strategy vector, which includes the maintenance decisions of each component at different time points; R sys(t,M) is the system reliability after considering the influence of maintenance activity M; T is the time period considered, that is, the longevity period;

[0160] S4-1-2. Objective 2: Minimize the total operation and maintenance costs throughout the entire longevity period T:

[0161] Min f2(M)=C total (T,M)=C pm (M)+C cm (M)+C loss (M)+C op (M)

[0162] Among them, C total is the total operation and maintenance cost, C pm is the planned maintenance cost, C cm is the cost of unplanned failure repair, C loss is the power generation loss cost of shutdown, C op For other operating costs;

[0163] S4-2. Define decision variables and constraints:

[0164] S4-2-1, Decision variables: M = {m k,i}, where m k,i represents the type of maintenance action taken for key component k at the i-th decision time point;

[0165] S4-2-2. Based on the prediction results of step S1, a degradation model for key components is established, specifically using a discrete-time Markov process based on the HI partition state; clarifying the effects of different maintenance actions on the health index HI state transition and the recovery of the remaining service life of the component;

[0166] S4-2-3. Constraints:

[0167] Total maintenance budget upper limit: ∑C(m k,i )≤Budget total , Budget total represents the total maintenance budget;

[0168] Maintenance resource constraints: number of available maintenance teams, availability of maintenance vessels, and spare parts inventory limitations;

[0169] Maintenance window constraints: The annual window for offshore operations, taking into account the influence of weather and sea conditions;

[0170] Operational constraints: minimum acceptable system reliability threshold R sys (t)≥R min , where R min Indicates the minimum acceptable system reliability;

[0171] Key component life constraints: mandatory replacement of key components when they reach their design life or their predicted remaining service life is exhausted;

[0172] S4-3, solving multi-objective optimization through improved non-dominated sorting genetic algorithm;

[0173] S4-3-1. Chromosome Encoding: A three-dimensional binary encoding matrix is ​​used to represent maintenance strategies, where the dimensions are [unit number × time interval × maintenance type]. For example, for each key component's planned maintenance action in each time interval, the encoding bit string is used to represent the specific preventive maintenance type taken (e.g., preset levels: 0 for unplanned maintenance, 1 for inspection, 2 for minor repair, 3 for major repair, 4 for replacement, etc.). If a failure occurs due to "unplanned maintenance" or poor PM effectiveness, corrective maintenance (CM) is triggered, and its cost and impact are reflected in the fitness function.

[0174] S4-3-2, Fitness Assessment: Based on the effectiveness of the maintenance strategy, calculate the fitness of each individual, including the two objectives of system reliability and maintenance cost;

[0175] System reliability:

[0176]

[0177] Among them, λ(τ) is the failure rate function, reflecting the failure probability of the unit;

[0178] Repair costs:

[0179]

[0180] Among them, c pm is the preventive maintenance cost, c cm is the corrective maintenance cost, δ i With ξ j are the execution times of the corresponding maintenance tasks respectively; It is a resource constraint penalty item that handles the constraint violation of maintenance team and resource allocation;

[0181] S4-3-3, genetic operation:

[0182] Selection operation: Use tournament selection to ensure the delivery of high-quality solutions.

[0183] Crossover operation: Use two-point crossover to generate a new solution.

[0184] Mutation operation: Adaptive mutation to ensure that the algorithm has sufficient exploratory power.

[0185] S4-4. Output decision:

[0186] (1) Pareto front generation: After 200 generations of evolution, a set of non-dominated solutions is output, representing different maintenance strategies;

[0187] (2) Multi-dimensional decision support: Provide different types of decision support based on the needs of decision makers:

[0188] Cost-sensitive: Select the maintenance strategy with the lowest cost (satisfying the reliability R≥0.92);

[0189] Risk-averse: Choose the most reliable strategy (cost increase does not exceed 15%).

[0190] Equilibrium decision point: Based on the entropy weight method, the optimal compromise solution is determined to balance cost and reliability;

[0191] (3) Maintenance plan output: Automatically generate maintenance work orders based on the selected maintenance strategy.

[0192] S5. Perform multi-objective maintenance optimization based on the multi-objective maintenance optimization decision results, and perform subsequent closed-loop optimization;

[0193] This step (S5) specifically includes:

[0194] Decision execution: Convert the selected multi-objective maintenance optimization decision results into specific maintenance work orders, dispatch resources (personnel, spare parts, ships) and execute preventive maintenance activities according to the optimization decision results;

[0195] Data update and model iteration: Collect information on actual maintenance activities, maintenance effect data, actual cost data, and subsequent equipment operation data; and feed this latest data back to steps S1 and S4;

[0196] Closed-loop optimization: Regularly rerun the entire process, using updated data and adopted models to dynamically adjust future reliability assessments, allocations, and maintenance decisions, achieving continuous improvement and adaptive optimization of operation and maintenance strategies.

[0197] The embodiments described above are only preferred embodiments of the present invention and are not intended to limit the scope of implementation of the present invention. Therefore, any changes made based on the shape and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. Reliability allocation modeling and maintenance decision-making method for offshore wind turbines under long-life management, characterized by: include: S1. Predict the health status of target components of offshore wind turbines based on multi-source heterogeneous data fusion and meta-transfer learning; S2. Build a reliability index allocation model based on RCM and fuzzy theory to identify key component failure modes and optimize the reliability index allocation of each key component, and then allocate the reliability index to each key component of the offshore wind turbine; S3. Based on the reliability index allocation result obtained in step S2, a system reliability model based on the minimum path set and the structure function is established; S4, combining the health status of the target components of the offshore wind turbine predicted in step S1 with the system reliability model established in step S3, and performing multi-objective maintenance optimization decision-making for longevity; S5. Perform multi-objective maintenance optimization based on the multi-objective maintenance optimization decision results, and perform subsequent closed-loop optimization.

2. The reliability allocation modeling and maintenance decision-making method for offshore wind turbines under long-life management according to claim 1 is characterized in that: Step S1 includes: S1-1. Data collection and preprocessing: Collect multi-source heterogeneous data during offshore wind turbine operation, including SCADA data, CMS vibration signal data, gearbox oil monitoring data, blade condition monitoring data, pitch / yaw system data, meteorological data, power grid data, and historical maintenance / fault records; and clean, normalize, and time-align the collected data. S1-2, extracting features related to the health status of key components from the preprocessed data; S1-3. Build and train a multi-source heterogeneous data fusion model based on CNN-LSTM, and then obtain the health index HI of each key component of the offshore wind turbine through the trained multi-source heterogeneous data fusion model based on CNN-LSTM; S1-4. Construct a fault diagnosis and prediction model based on meta-transfer learning, and train the fault diagnosis and prediction model based on meta-transfer learning using the health index HI of each key component of the offshore wind turbine obtained in step S1-3. Then, the health status of the target component, i.e., the predicted value of the remaining service life and its confidence interval, is obtained through the trained fault diagnosis and prediction model based on meta-transfer learning.

3. The reliability allocation modeling and maintenance decision-making method for offshore wind turbines under long-life management according to claim 2 is characterized in that: The multi-source heterogeneous data fusion model based on CNN-LSTM constructed in step S1-3 uses parallel convolutional neural network (CNN) branches to process local spatial features or patterns from different sensors respectively; uses long short-term memory (LSTM) branches to process the temporal dependency features of data acquisition and monitoring control system (SCADA) data; and performs weighted fusion of the features extracted by the parallel convolutional neural network (CNN) and the features processed by the long short-term memory (LSTM) network in the fusion layer. The weights are obtained through network learning, and the fused features are input into the subsequent fully connected layer. Finally, the health index HI of the key components of the offshore wind turbine is output.

4. The reliability allocation modeling and maintenance decision-making method for offshore wind turbines under long-life management according to claim 2 is characterized in that: Step S1-4 includes: The long short-term memory network (LSTM) is used as the basic model for predicting the remaining useful life, and the input is the health index HI of the key components obtained in step S1-3; Meta-training: The lifespan data for each offshore wind turbine or key component of the same type is treated as an independent "task." Within each task, the model parameters are updated using a small number of gradient steps. Between tasks, the model's initial parameters are optimized so that, starting from this initial state, the model can quickly achieve good performance on any new task using a small amount of data. Meta-testing phase: For the target components of the target offshore wind turbine, a small amount of early operating data or labeled data is used to perform a small amount of gradient updates based on the initial parameters obtained from meta-training to quickly adapt the model to the degradation pattern of the target components; Output the predicted value of the remaining useful life of the target component and its confidence interval.

5. The reliability allocation modeling and maintenance decision-making method for offshore wind turbines under long-life management according to claim 1 is characterized in that: Step S2 includes: S2-1. Use the RCM Failure Mode, Effects, and Criticality (FMEC) analysis method, combined with historical failure data and expert experience, to identify the functions, potential failure modes, and impacts of each key component of the offshore wind turbine on system performance, safety, and the environment. Based on the risk priority number from the FMEC analysis results, select the set of key components that have the greatest impact on the overall reliability, safety, and maintenance costs of the offshore wind turbine. S2-2, determine the reliability index allocation influencing factor set, identify the factors that affect the reliability index allocation of key components, and form the influencing factor set U = {u1, u2, ..., u i },u1,u2,...,u i All are influencing factors; S2-3, perform fuzzy evaluation of reliability indicators and obtain the fuzzy evaluation matrix of each key component k; S2-4, determine the reliability index allocation weight; S2-5. Allocate reliability indicators to key components of offshore wind turbines based on reliability indicator weights.

6. The offshore wind turbine reliability allocation modeling and maintenance decision-making method under long-life management according to claim 5 is characterized in that: Step S2-3 includes: S2-3-1. Establish an evaluation level set: define a fuzzy evaluation level set for the evaluation results of each influencing factor; S2-3-2. Construct membership function: For each influencing factor, establish its membership function relative to the evaluation level v according to its quantitative data range or expert experience. j The membership function μ ui (v j ); S2-3-3, Single factor fuzzy evaluation: For each key component k, collect or calculate its index value under each influencing factor; use the membership function to determine the key component under factor u i Next, for each evaluation level v j The membership degree forms a single factor evaluation vector R ki : R ki =[μ ui (v1) k ,m ui (v2) k ,...,m ui (v n ) k ] S2-3-4. Construct a fuzzy evaluation matrix: Combine the single-factor evaluation vectors of all key components k with respect to all influencing factors to form a fuzzy evaluation matrix for each key component k [R k1 ; R k2 ;...;R km ].

7. The reliability allocation modeling and maintenance decision-making method for offshore wind turbines under long-life management according to claim 5 is characterized in that: Step S2-4 includes: S2-4-1. Determine the weight of factors: Use fuzzy analytic hierarchy process to build a fuzzy judgment matrix and compare the influencing factors u i For the relative degree of importance of reliability index allocation, the fuzzy weight vector W = [w1, w2, ..., w m ], where w i Representation factor u i The weight of S2-4-2, fuzzy comprehensive evaluation: For each key component k, use its fuzzy evaluation matrix R k The fuzzy weight vector W of the factors is used to perform fuzzy comprehensive evaluation, using the weighted average model M(·,+), and the formula is: Get the comprehensive fuzzy evaluation vector B of each key component k k =[b k1 ,b k2 ,...,b kn ], where b kj Indicates that the comprehensive performance of key component k belongs to evaluation level v j The degree of membership; S2-4-3, based on comprehensive fuzzy evaluation vector B k Determine the reliability index allocation weights, the process is as follows: Defuzzification is used to obtain the score of each component: Among them, S k represents the clarity score of key component k; Calculate the sum of all key component scores: Among them, S total represents the sum of all key component scores; Calculate the reliability index allocation weight of each key component: Among them, W k represents the reliability weight assigned to key component k, R total It is the overall reliability target of the system.

8. The reliability allocation modeling and maintenance decision-making method for offshore wind turbines under long-life management according to claim 1 is characterized in that: Step S3 includes: S3-1. Based on the reliability structure diagram of key components of offshore wind turbines, the adjacency matrix method is combined with the depth-first search algorithm to find the minimum path set P = {P1, P2, ..., P q }, where each P i is a set of key components; the reliability structure diagram of key components of offshore wind turbines is obtained by screening out the key component set that has the greatest impact on the overall reliability, safety and maintenance cost of the wind turbine, and analyzing the logical connection relationship of the key component set in the system; S3-2, perform the minimum path set disjoint processing: Use the structure function matrix method to convert the minimum path set event {E(P i )} into a set of mutually disjoint events {D1, D2, ..., D r }, so that the event of system success is equivalent to the union of mutually disjoint events; S3-3. Establish a system reliability model: Assume that the reliability of the Kth key component at time t is R K (t), then the reliability of the system at time t is R sys (t) By calculating the probability of mutually disjoint events, we can obtain: R sys (t)=P(∪D i )=∑P(D i ) Among them, each P(D i ) represents the mutually disjoint events D i The key components and their connection relationships are included, and the reliability R of the key components at time t is used K (t) The calculated probability.

9. The reliability allocation modeling and maintenance decision-making method for offshore wind turbines under long-life management according to claim 1 is characterized in that: Step S4 includes: S4-1. Constructing a multi-objective optimization function: S4-1-1. Objective 1: Maximize the average reliability of offshore wind turbine systems throughout their entire lifespan Where M is the maintenance strategy vector, which includes the maintenance decisions of each component at different time points; R sys (t,M) is the system reliability after considering the influence of maintenance activity M; T is the time period considered, that is, the longevity period; S4-1-2. Objective 2: Minimize total operation and maintenance costs throughout the entire longevity period T: Min f2(M)=C total (T,M)=C pm (M)+C cm (M)+C loss (M)+C op (M) Among them, C total is the total operation and maintenance cost, C pm is the planned maintenance cost, C cm is the cost of unplanned failure repair, C loss is the power generation loss cost of shutdown, C op For other operating costs; S4-2. Define decision variables and constraints: S4-2-1, Decision variables: M = {m k,i }, where m k,i represents the type of maintenance action taken on key component k at the i-th decision time point; S4-2-2. Based on the prediction results of step S1, a degradation model for key components is established, specifically using a discrete-time Markov process based on the HI partition state; clarifying the effects of different maintenance actions on the health index HI state transition and the recovery of the remaining service life of the component; S4-2-3. Constraints: Total maintenance budget upper limit: ∑C(m k,i )≥Budget total , Budget total represents the total maintenance budget; Maintenance resource constraints: number of available maintenance teams, availability of maintenance vessels, and spare parts inventory limitations; Maintenance window constraints: The annual window for offshore operations, taking into account the influence of weather and sea conditions; Operational constraints: minimum acceptable system reliability threshold R sys (t)≥R min , where R min Indicates the minimum acceptable system reliability; Key component life constraints: mandatory replacement of key components when they reach their design life or their predicted remaining service life is exhausted; S4-3, solving multi-objective optimization through improved non-dominated sorting genetic algorithm; S4-3-1. Chromosome coding: A three-dimensional binary coding matrix is ​​used to represent the maintenance strategy, where the dimensions are [unit number × time interval × maintenance type]; S4-3-2, Fitness Assessment: Based on the effectiveness of the maintenance strategy, calculate the fitness of each individual, including the two objectives of system reliability and maintenance cost; System reliability: Among them, λ(τ) is the failure rate function, reflecting the failure probability of the unit; Repair costs: Among them, c pm is the preventive maintenance cost, c cm is the corrective maintenance cost, δ i With ξ j are the execution times of the corresponding maintenance tasks respectively; It is a resource constraint penalty item that handles the constraint violation of maintenance team and resource allocation; S4-3-3, genetic manipulation; S4-4. Output decision: (1) Pareto front generation: After multiple generations of evolution, a set of non-dominated solutions is output, representing different maintenance strategies; (2) Multi-dimensional decision support: Provide different types of decision support based on the needs of decision makers: Cost-sensitive: choose the maintenance strategy with the lowest cost; Risk-averse: choose the most reliable strategy; Equilibrium decision point: Based on the entropy weight method, the optimal compromise solution is determined to balance cost and reliability; (3) Maintenance plan output: Automatically generate maintenance work orders based on the selected maintenance strategy.

10. The reliability allocation modeling and maintenance decision-making method for offshore wind turbines under long-life management according to claim 1, characterized in that: Step S5 includes: Decision execution: Convert the selected multi-objective maintenance optimization decision results into specific maintenance work orders, dispatch resources and execute preventive maintenance activities according to the optimization decision results; Data update and model iteration: Collect information on actual maintenance activities, maintenance effect data, actual cost data, and subsequent equipment operation data; and feed this latest data back to steps S1 and S4; Closed-loop optimization: Regularly rerun the entire process, using updated data and adopted models to dynamically adjust future reliability assessments, allocations, and maintenance decisions, achieving continuous improvement and adaptive optimization of operation and maintenance strategies.

Images