Bayesian network-based offshore wind turbine support structure safety assessment method

Abstract

The invention discloses an offshore wind turbine support structure safety assessment method based on a Bayesian network, and relates to the technical field of safety assessment of offshore wind turbine support structures, and the method comprises the steps: obtaining real-time monitoring data; building a dynamic Bayesian network according to a preset Bayesian network structure; inputting the real-time monitoring data into the dynamic Bayesian network; updating the posterior probability of the dynamic Bayesian network by using a junction tree algorithm to obtain the real-time safety probability of the offshore wind turbine support structure; according to the dynamic Bayesian network, starting from a safety state abnormity result of the supporting structure of the offshore wind turbine, tracing potential reasons causing abnormity, and determining a risk chain of the offshore wind turbine system; and performing security assessment according to the real-time security probability and the risk chain. According to the invention, based on the Bayesian network updated in real time, the safe state of the support structure of the offshore wind turbine can be evaluated quickly and accurately.

Description

Technical Field

[0001] This invention relates to the technical field of safety assessment of offshore wind turbine support structures, and in particular to a safety assessment method for offshore wind turbine support structures based on Bayesian networks. Background Technology

[0002] With the increasing global demand for clean energy, wind power, as a clean and renewable energy source, has received widespread attention. Offshore wind energy resources are more abundant than onshore wind resources, with higher wind speeds and better stability, enabling offshore wind farms to achieve higher power generation efficiency. In recent years, the scale of offshore wind farms in my country has shown a trend of continuous expansion.

[0003] However, the harsh marine environment presents numerous challenges to offshore wind turbines. Complex marine environmental loads, such as waves, wind, currents, and seawater corrosion, make the safety hazards facing the support structures of offshore wind turbines far more complex than those for onshore wind power. Problems with the support structure can easily lead to safety accidents, causing not only severe economic losses but also potential pollution of the marine environment.

[0004] Currently, safety inspections of offshore wind turbines mainly rely on frequent manual patrols, which has many drawbacks. On the one hand, manual inspections are costly and time-consuming, requiring a significant investment of manpower, resources, and funds. On the other hand, inspection results are often based on the subjective judgment of the inspectors, lacking objectivity and accuracy. Therefore, it is urgent to develop a reliable and efficient safety assessment method for offshore wind turbine support structures.

[0005] Therefore, how to quickly and accurately assess the safety status of offshore wind turbine support structures has become an urgent technical problem to be solved. Summary of the Invention

[0006] The technical problem solved by this invention is the inability to quickly and accurately assess the safety status of offshore wind turbine support structures.

[0007] To address the aforementioned technical problems, this invention provides the following technical solution: a safety assessment method for offshore wind turbine support structures based on Bayesian networks, comprising: acquiring real-time monitoring data; constructing a dynamic Bayesian network according to a preset Bayesian network structure; inputting the real-time monitoring data into the dynamic Bayesian network; updating the posterior probability of the dynamic Bayesian network using a connection tree algorithm to obtain the real-time safety probability of the offshore wind turbine support structure; based on the dynamic Bayesian network, tracing the potential causes of the anomalies from the abnormal safety status results of the offshore wind turbine support structure to determine the risk chain of the offshore wind turbine system; and performing a safety assessment based on the real-time safety probability and the risk chain.

[0008] Preferably, the method for setting the Bayesian network structure includes: combining fault information and wind turbine system decomposition results from a pre-defined integrated database to summarize the failure modes and causes of offshore wind turbines, and establishing a fault tree model of the offshore wind turbine support structure; mapping the events of the fault tree model to Bayesian network nodes, mapping the fault logic of the fault tree model to directed edges of the Bayesian network, and mapping the logic gates of the fault tree model to conditional probability tables, to establish a static Bayesian network model of wind turbine support structure failure; based on the Markov assumption and the time homogeneity assumption, extending the static Bayesian network model into a dynamic Bayesian network model; and determining the Bayesian network structure according to the dynamic Bayesian network model.

[0009] Preferably, determining the Bayesian network structure based on the dynamic Bayesian network model includes: quantifying the risk of fault causes through a parameter learning method to determine the risk probability distribution of the Bayesian network structure, wherein the risk probability distribution is used to characterize the prior probability and conditional probability of faulty nodes in the Bayesian network structure.

[0010] Preferably, the step of quantifying the risk of fault causes and determining the risk probability distribution of the Bayesian network structure through parameter learning includes: performing parameter learning on the comprehensive database using a preset EM algorithm to obtain objective probability parameters; quantifying the risk of fault causes based on the experience of multiple industry experts to obtain subjective probability parameters; and fusing the subjective probability parameters and objective probability parameters to obtain the risk probability distribution of the Bayesian network structure.

[0011] Preferably, the step of learning parameters from the comprehensive database using a preset EM algorithm to obtain objective probability parameters includes: obtaining the initial prior probability of the faulty node in the Bayesian network structure; obtaining historical monitoring data; determining the expected value of missing data in the historical monitoring data based on the initial prior probability; generating a learning dataset based on the historical monitoring data and the expected value of the missing data; and determining the objective prior probability and objective conditional probability of the faulty node in the Bayesian network structure based on the learning dataset to obtain the objective probability parameters.

[0012] Preferably, the step of quantifying the risk of fault causes based on the experience of multiple industry experts to obtain subjective probability parameters includes: obtaining the weights of each expert in the expert group; obtaining the subjective prior probabilities of each expert for each fault cause; determining the occurrence probability of each fault mode; determining the subjective conditional probability of the fault node in the Bayesian network structure based on the occurrence probability of each fault mode; and determining the subjective probability parameters based on the weights of each expert, the subjective prior probabilities, and the subjective conditional probabilities.

[0013] Preferably, determining the probability of occurrence of each failure mode includes: obtaining OSD scores from experts for each failure cause; determining the RPN value of the failure cause based on the OSD scores to obtain a first evaluation vector composed of the risk values ​​of each part under each failure cause; obtaining the occurrence state of the failure cause to obtain multiple occurrence state vectors of failure causes; determining the RPN value corresponding to the occurrence state of the failure cause based on the first evaluation vector and the occurrence state vectors; and determining a second evaluation vector corresponding to the failure mode based on the RPN value corresponding to the occurrence state of the failure cause, wherein the second evaluation vector is used to characterize the probability of occurrence of the failure mode under the occurrence state.

[0014] Preferably, the method of combining fault information and wind turbine system decomposition results from a pre-defined integrated database to summarize the failure modes and causes of offshore wind turbines and establish a fault tree model of the offshore wind turbine support structure includes: acquiring fault information and support structure information of the offshore wind turbine; decomposing the structure of the offshore wind turbine according to the support structure information to obtain decomposition information; determining the failure modes and causes of the offshore wind turbine according to the fault information and the decomposition information; and establishing a fault tree model of the offshore wind turbine according to the failure modes and causes.

[0015] Preferably, the decomposed information includes multiple subsystems, each subsystem including multiple components, and the multiple subsystems include at least one of the following: pile foundation subsystem, jacket foundation subsystem, pile cap subsystem, tower subsystem, and auxiliary subsystem.

[0016] Preferably, before quantifying the risk of fault causes using a parameter learning method and determining the risk probability distribution of the Bayesian network structure, the method further includes: iteratively optimizing the dynamic Bayesian network model using a random forest K-fold cross-validation algorithm.

[0017] The beneficial effects of this invention are as follows: A dynamic Bayesian network is constructed using a preset Bayesian network structure to pre-store the prior probabilities of risk assessment. Real-time monitoring data is input into the dynamic Bayesian network, and the posterior probabilities of the dynamic Bayesian network are updated using a connection tree algorithm to obtain the real-time safety probability of the offshore wind turbine support structure. This allows for accurate assessment of the safety of the offshore wind turbine support structure, and the real-time updating of the posterior probabilities of the dynamic Bayesian network enables real-time assessment of the safety probability of the offshore wind turbine support structure, thereby achieving rapid and accurate assessment of the safety status of the offshore wind turbine support structure. Attached Figure Description

[0018] Figure 1 A schematic diagram of the basic process of a safety assessment method for offshore wind turbine support structures based on Bayesian networks, provided as an embodiment of the present invention;

[0019] Figure 2 This is an exploded view of an offshore wind turbine support structure system provided in one embodiment of the present invention;

[0020] Figure 3 An example diagram of a fault tree model for an offshore wind turbine support structure provided in one embodiment of the present invention;

[0021] Figure 4 A schematic diagram of a static Bayesian network model based on fault tree construction provided in one embodiment of the present invention;

[0022] Figure 5 A schematic diagram of a dynamic Bayesian network structure based on monitoring data updates provided in one embodiment of the present invention;

[0023] Figure 6 A flowchart for determining Bayesian network parameters provided in one embodiment of the present invention;

[0024] Figure 7 A schematic diagram of a partial conditional probability table of a Bayesian network provided in an embodiment of the present invention;

[0025] Figure 8 This is a schematic diagram illustrating the structural security of a Bayesian network forward inference evaluation system, provided as an embodiment of the present invention.

[0026] Figure 9 This is a schematic diagram of a high-risk unit for reverse inference in a Bayesian network, provided as an embodiment of the present invention. Detailed Implementation

[0027] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0028] Example 1, referring to Figure 1 As an embodiment of the present invention, a safety assessment method for offshore wind turbine support structures based on Bayesian networks is provided, including steps S110 to S160:

[0029] S110, acquire real-time monitoring data.

[0030] The real-time monitoring data can be environmental load data, such as wind speed, wind direction, wave height, wave period, and ocean current speed; it can also be structural dynamic response data, such as vibration acceleration of key nodes and microstrain of strain gauges; it can also be foundation state data, such as pile foundation tilt angle and scour depth (sonar monitoring); and it can also be finite element simulation data, etc.

[0031] S120: Construct a dynamic Bayesian network based on a preset Bayesian network structure.

[0032] Bayesian networks are a technology developed based on probability theory (Bayes' theorem) and graph theory (directed acyclic graphs). They can clearly express the causal relationships of structural failures and can express and reason about expert prior knowledge, historical data, and other incomplete and uncertain information in probabilistic form. Based on the solid probabilistic foundation of Bayesian networks, they have unique advantages in handling various uncertain problems, such as safety status assessment, risk assessment, and fault analysis; at the same time, their graph theory foundation allows users to intuitively understand the causal relationships between events. Therefore, applying Bayesian networks to the safety assessment of offshore wind turbine support structures has strong practical significance.

[0033] The preset Bayesian network structure already includes prior probabilities and conditional probabilities, and the prior probabilities and conditional probabilities of the dynamic Bayesian network are consistent with those of the preset Bayesian network structure.

[0034] S130 inputs real-time monitoring data into a dynamic Bayesian network.

[0035] S140 uses the connection tree algorithm to update the posterior probability of the dynamic Bayesian network to obtain the real-time safety probability of the offshore wind turbine support structure.

[0036] Among them, the Junction Tree Algorithm (also known as the Clique Tree algorithm) is a core algorithm in probabilistic graphical models for efficient computation of accurate inference. It is particularly suitable for updating the posterior probability of Bayesian networks (including dynamic Bayesian networks). It is used to inject real-time monitoring data (such as vibration values ​​exceeding the standard) into the corresponding clique nodes, transmit (collect) from leaf node cliques to root node cliques, and distribute (distribute) from root node cliques to leaf node cliques. Local probability fusion is performed within each clique (which must satisfy: joint probability of variables within the clique = conditional probability of the separating set × probability of the sub-clique).

[0037] S150, based on dynamic Bayesian networks, starts from the abnormal results of the safety status of the support structure of offshore wind turbines, traces the potential causes of the abnormalities, and determines the risk chain of the offshore wind turbine system.

[0038] Based on the fault modes in dynamic Bayesian networks, the fault modes corresponding to abnormal results are determined.

[0039] S160 performs security assessments based on real-time security probabilities and risk chains.

[0040] The aforementioned Bayesian mesh was constructed using GeNIe software to achieve dynamic modeling of the entire system. The connection tree algorithm was used as the algorithm for solving the posterior probability of the Bayesian network, and real-time probability updates were performed based on real-time monitoring data. Real-time monitoring data (collected every 10 minutes) was input into the model, and the parameter learning process was repeated to achieve dynamic modeling and real-time probability updates. Forward reasoning was performed based on the dynamic Bayesian network to infer the structural safety status. Backward reasoning was also performed based on the dynamic Bayesian network to trace potential causes from abnormal safety status results. Based on the analysis results, operation and maintenance strategies were formulated, professional divers were assigned to conduct key inspections of the pile foundations, and severely scoured areas were promptly reinforced to ensure the safe and stable operation of the wind turbine.

[0041] The fundamental principle of Bayesian inference is Bayes' theorem:

[0042]

[0043] If A1, A1, ..., A n This constitutes a complete set of events, meaning that they are mutually exclusive and... For any event B, when P(B) > 0, Bayes' theorem can be expressed as:

[0044]

[0045] Preferably, the method for setting the Bayesian network structure includes steps S210 to S250:

[0046] S210: Combine the fault information and wind turbine system decomposition results from the pre-set integrated database, summarize the failure modes and causes of offshore wind turbines, and establish a fault tree model of the offshore wind turbine support structure.

[0047] S210 includes S211 to S214:

[0048] S211, obtain fault information and support structure information of offshore wind turbines.

[0049] S212, the structure of the offshore wind turbine is decomposed based on the supporting structure information to obtain decomposition information.

[0050] S213, determine the failure mode and cause of failure of the offshore wind turbine based on the fault information and decomposition information.

[0051] S214. Establish a fault tree model for offshore wind turbines based on failure modes and causes of failure.

[0052] S220: Map the events of the fault tree model to the nodes of the Bayesian network, map the fault logic of the fault tree model to the directed edges of the Bayesian network, and map the logic gates of the fault tree model to the conditional probability table to establish a static Bayesian network model of the fault of the wind turbine support structure.

[0053] S230 extends the static Bayesian network model into a dynamic Bayesian network model based on the Markov assumption and the time homogeneity assumption.

[0054] S240 uses the random forest K-fold cross-validation algorithm to iteratively optimize the dynamic Bayesian network model.

[0055] S250, determine the Bayesian network structure based on the dynamic Bayesian network model.

[0056] Preferably, step S250 includes: quantifying the risk of fault causes using a parameter learning method to determine the risk probability distribution of the Bayesian network structure. This risk probability distribution characterizes the prior and conditional probabilities of faulty nodes in the Bayesian network structure. Further, it includes steps S251–S253: S251, performing parameter learning on a comprehensive database using a pre-defined EM algorithm to obtain objective probability parameters; S252, quantifying the risk of fault causes based on the experience of multiple industry experts to obtain subjective probability parameters; and S253, fusing the subjective and objective probability parameters to obtain the risk probability distribution of the Bayesian network structure.

[0057] Preferably, step S251 includes steps S251a to S251e: S251a, obtaining the initial prior probability of the faulty node in the Bayesian network structure; S251b, obtaining historical monitoring data; S251c, determining the expected value of the missing data in the historical monitoring data based on the initial prior probability; S251d, generating a learning dataset based on the expected value of the historical monitoring data and the missing data; S251e, determining the objective prior probability and objective conditional probability of the faulty node in the Bayesian network structure based on the learning dataset, and obtaining the objective probability parameter.

[0058] Preferably, S252 includes S252a to S252e: S252a, obtaining the weights of each expert in the expert group; S252b, obtaining the subjective prior probabilities of each expert for each fault cause; S252c, determining the occurrence probability of each fault mode; S252d, determining the subjective conditional probability of the fault node in the Bayesian network structure based on the occurrence probability of each fault mode; S252e, determining the subjective probability parameters based on the weights, subjective prior probabilities, and subjective conditional probabilities of each expert.

[0059] Preferably, S252c includes S252ca to S252ce: S252ca, obtaining the OSD scores of each expert for each fault cause; S252cb, determining the RPN value of the fault cause based on the OSD scores, and obtaining a first evaluation vector composed of the risk values ​​of each part under each fault cause; S252cc, obtaining the occurrence state of the fault cause, and obtaining multiple occurrence state vectors of fault causes; S252cd, determining the RPN value corresponding to the occurrence state of the fault cause based on the first evaluation vector and the occurrence state vector; S252ce, determining a second evaluation vector corresponding to the fault mode based on the RPN value corresponding to the occurrence state of the fault cause, the second evaluation vector being used to characterize the occurrence probability of the fault mode under the occurrence state.

[0060] Specifically, data collection and database establishment: Through extensive research, and by using the websites of organizations such as OREDA, WWEA, and GWEC, we collected 50 offshore wind turbine accident cases, maintenance records of 10 offshore wind farms over the past 5 years, and finite element simulation data from around the world as data reserves. Considering the potential limitations and randomness of the selected samples, we also invited 5 senior industry experts, including those specializing in wind turbine support structures, to supplement the basic component failure probability information with their professional experience, further enhancing the reliability of the data.

[0061] Data cleaning techniques are used to supplement or modify important missing, duplicate, and erroneous values ​​in the data; after standardizing the data format, a comprehensive database for safety analysis of offshore wind turbine support structures is established.

[0062] Structural Decomposition and Information Gathering: Relevant standards such as GB / T 10105-2018 and GB / T 31517.1-2022 were consulted to obtain design information for the wind turbine support structure, including structural and component composition and key safety control indicators. The wind turbine support structure was decomposed into five subsystems: pile foundation subsystem, jacket frame subsystem, pier cap frame system, tower subsystem, and auxiliary subsystem. These were further subdivided into 12 main components, such as... Figure 2 As shown.

[0063] Fault tree model construction: such as Figure 2 As shown, by combining the fault information in the above-mentioned integrated database with the wind turbine system decomposition results, the failure modes and causes of the wind turbine are summarized, 36 types of failure causes are identified, and a fault tree model of the offshore wind turbine support structure is established.

[0064] like Figure 3 As shown, the failure of the wind turbine support structure is taken as the top-level event. According to the causal relationship of logic gates, the failure of subsystems and components is taken as intermediate events, and the basic cause of component failure is taken as the basic event. A fault tree model is constructed through logical relationships such as AND gates and OR gates to clearly present the failure path.

[0065] Bayesian Network Model Construction and Extension: Based on the aforementioned fault tree model, and following the corresponding transformation algorithm, a static Bayesian network model of the wind turbine support structure fault is established. Specifically, according to the fault tree model transformation algorithm, events in the fault tree are mapped to Bayesian network nodes, fault logic to directed edges, and logic gates to conditional probability tables, thus establishing a static Bayesian network model. Based on the Markov principle, the static Bayesian network model is extended into a dynamic Bayesian network model, such as... Figure 4 As shown.

[0066] like Figure 5 As shown, based on the Markov assumption (the probability of a node occurring in each time slice depends only on the probability of a node occurring in the previous time slice) and the homogeneity assumption (the random process is stationary and invariant within a finite time interval, with consistent conditional probabilities), the static model is extended into a dynamic Bayesian network model using the Markov principle to reflect the system's time-varying characteristics. The Markov assumption states that throughout the entire process, the probability of a node occurring in each time slice depends only on the probability of a node occurring in the previous time slice, and is independent of any other time slices prior to it. P(X) t |X1,X2,...X t-1 )=P(X t |X t-1 The time homogeneity assumption states that the stochastic process is stationary and invariant over a finite time period, meaning that the conditional probability remains constant for any given time t.

[0067] The Bayesian network structure is optimized using the random forest K-fold cross-validation algorithm to improve model accuracy and generalization ability. Specifically, the random forest-K-fold cross-validation algorithm (K=5) is used to perform 100 iterations to optimize the Bayesian network structure and obtain the optimal structure. The risk of fault causes is quantified through parameter learning methods to determine the risk probability distribution, providing prior probabilities and conditional probabilities of nodes for subsequent network inference. The parameter learning process is as follows: Figure 6 As shown.

[0068] Considering the potential limitations of only considering wind turbine failure data, we should comprehensively consider both subjective and objective data information on wind turbine failures, and adopt a 1:1 fusion of the two data results to determine the risk probability distribution of each failure cause, so as to provide the prior probability and conditional probability of nodes for network inference.

[0069] (1) The EM algorithm is used to learn parameters from objective data in the database. The steps are as follows:

[0070] (a) Initialize parameters: Set initial values ​​for the parameters in the Bayesian network based on domain knowledge, experience, or simple statistical analysis.

[0071] The prior probabilities of each fault node are initially estimated, such as assuming the initial prior probability of pile foundation failure is 0.1, and the initial values ​​of the conditional probabilities between each node are also estimated.

[0072] (b) E-step (expectation step): Calculate the expectation of the latent variables based on the current parameter estimates. In the case of missing data, latent variables can be understood as the missing data portions.

[0073] For monitoring data of offshore wind turbines, if some pile foundation support reaction force data is missing, the expected value of the missing pile foundation reaction force data can be estimated by probability calculation based on the parameters of the current Bayesian network model, using other existing data (such as tower displacement, wind speed and other related data) and model structure.

[0074] (c) M-step (maximization step): The expected values ​​of the latent variables calculated in the E-step are added to the dataset, the dataset is treated as complete, and then the parameters of the model are updated using maximum likelihood estimation or other suitable methods.

[0075] Taking an offshore wind turbine failure prediction model as an example, a complete dataset with missing expected values ​​is used to calculate the new prior probability and conditional probability for each node. In this way, parameter values ​​that maximize the likelihood function are found, improving the model's fit to the data.

[0076] (d) Convergence Judgment: Check the changes before and after the parameter update. The difference between the new parameter value and the old parameter value can be calculated. For example, calculate the change of each parameter. If the sum of the absolute values ​​of the changes of all parameters is less than a preset threshold (such as 0.001), or the likelihood function value no longer increases significantly after multiple iterations (the increase is less than a certain minimum value), then the algorithm is considered to have converged and the iteration stops; otherwise, return to step E and continue to perform the next round of expectation calculation and parameter update.

[0077] (2) In order to avoid the limitations and biases of the collected objective data, the experience of 5 industry experts was also considered to quantify the risk of failure causes. The specific steps are as follows:

[0078] (a) Determine the weight of the expert group by inviting senior experts in the industry and calculating the weight of the experts using the entropy weight method based on information such as the experts' education level, professional title, service time, and age.

[0079] (b) Experts provide the prior probabilities of each cause of failure based on their experience;

[0080] (c) After experts in relevant fields score the main causes of CNC machine tool failures, the three risk parameters of occurrence (O), severity (S) and detectability (D) are multiplied to calculate the RPN (Risk Priority Number) value, which is used to represent the risk priority coefficient of the failure cause.

[0081] The risk values ​​of the n causes of failure of the i-th part constitute the evaluation vector R:

[0082] R i =[RPN1,RPN2,...,RPN n i = 1, 2, ..., n;

[0083] For a single part, a risk assessment vector is determined based on the relevant failure modes and their causes. An occurrence vector is determined based on the states of p failure causes: L = [L1, L2, ..., L...]. n ], j = 1, 2, ..., n, the value of L is determined according to whether the corresponding fault cause occurs, a value of 0 represents that it does not occur, and a value of 1 represents that it occurs;

[0084] (d) The RPN value after expert evaluation is used to measure the contribution of each failure cause to the component failure mode. The RPN is determined by multiplying the risk assessment vector and the occurrence vector under the corresponding failure cause state.

[0085] The risk assessment vector for each failure mode is determined by summing the RPN values ​​under various failure causes: R FM =[RPN FM1 RPN FM2 ,...,RPN FMK ], j = 1, 2, ..., n.

[0086] (e) Normalize the RPN value of the entire part failure mode to obtain the probability of the corresponding failure mode. Use the result as the conditional probability between the failure cause and the failure component, and use it as the parameter definition of the nodes between the failure cause node and the component in the later dynamic Bayesian network.

[0087]

[0088] (e) Combining the weights of the experts, after weighted averaging and normalization, we obtain the prior probability and conditional probability tables of the reasoning network based on subjective experience.

[0089] (3) Subjective and objective probability parameters are fused with a 1:1 weighting to obtain the sum probability table for Bayesian network inference, which serves as the basis for subsequent inference. A partial conditional probability table is shown below. Figure 7 As shown;

[0090] Dynamic Modeling and Probability Update: The aforementioned Bayesian mesh is constructed using GeNIe software to achieve dynamic modeling of the entire system. The connection tree algorithm is used as the algorithm for solving the posterior probability of the Bayesian network. Real-time probability updates are performed based on real-time monitoring data. Real-time monitoring data (collected every 10 minutes) is input into the model, and the above parameter learning process is repeated to achieve dynamic modeling and real-time probability updates. Risk Analysis and Maintenance Decision: Forward reasoning is performed based on the dynamic Bayesian network to infer the safety status of the structure, such as... Figure 8 As shown; reverse reasoning is performed based on dynamic Bayesian networks to trace potential causes from abnormal security state results, such as... Figure 9 As shown; based on the analysis results, an operation and maintenance strategy was formulated, professional divers were arranged to conduct key inspections of the pile foundation, and severely scoured areas were reinforced in a timely manner to ensure the safe and stable operation of the wind turbine.

[0091] The method provided in this invention represents the first application of Bayesian network technology in offshore wind power reliability assessment. This method integrates multi-source data to establish a comprehensive database, providing a complete and accurate data foundation for safety assessment. It utilizes fault tree and Bayesian network models to clearly reveal the causal relationships of structural failures, improving the accuracy and reliability of the assessment. It optimizes the network structure using random forest K-fold cross-validation and combines parameter learning to determine the probability distribution, enhancing model performance. Based on real-time monitoring data, it updates the safety probability and issues early warnings in real time, promptly identifying potential risks. Based on real-time monitoring data, this invention can update the safety probability in real time and issue timely danger warnings, effectively achieving rapid detection and early warning of potential risks, saving valuable time for maintenance personnel and reducing the possibility of safety accidents. By identifying risk chains and risk units through risk analysis, it provides a scientific basis for targeted maintenance, enabling the rational allocation of maintenance resources, effectively reducing the safety risks of offshore wind turbine support structures, and bringing significant economic and environmental benefits. Compared with traditional manual inspection methods, the Bayesian network-based offshore wind turbine support structure safety assessment device used in this invention has significant advantages. It overcomes the shortcomings of manual inspection, such as high cost, long cycle and low accuracy, and can realize long-term and stable health monitoring of offshore wind power structures, providing a strong guarantee for the sustainable development of the offshore wind power industry.

[0092] This application embodiment constructs a dynamic Bayesian network through a preset Bayesian network structure to pre-store the prior probabilities of risk assessment. Real-time monitoring data is input into the dynamic Bayesian network, and the posterior probabilities of the dynamic Bayesian network are updated using a connection tree algorithm to obtain the real-time safety probability of the offshore wind turbine support structure. This allows for accurate assessment of the safety of the offshore wind turbine support structure, and the real-time updating of the posterior probabilities of the dynamic Bayesian network enables real-time assessment of the safety probability of the offshore wind turbine support structure, thereby achieving rapid and accurate assessment of the safety status of the offshore wind turbine support structure.

[0093] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0094] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A safety assessment method for offshore wind turbine support structures based on Bayesian networks, characterized in that, include: Obtain real-time monitoring data; A dynamic Bayesian network is constructed based on a pre-defined Bayesian network structure. The real-time monitoring data is input into the dynamic Bayesian network; The posterior probability of the dynamic Bayesian network is updated using the connection tree algorithm to obtain the real-time safety probability of the offshore wind turbine support structure. Based on the dynamic Bayesian network, starting from the abnormal safety status of the support structure of the offshore wind turbine, the potential causes of the abnormality are traced back to determine the risk chain of the offshore wind turbine system. A security assessment is performed based on the real-time security probability and the risk chain.

2. The method as described in claim 1, characterized in that, The method for setting the Bayesian network structure includes: By combining the fault information and wind turbine system decomposition results from the pre-set integrated database, the failure modes and causes of offshore wind turbines are summarized, and a fault tree model of the offshore wind turbine support structure is established. The events of the fault tree model are mapped to Bayesian network nodes, the fault logic of the fault tree model is mapped to directed edges of the Bayesian network, and the logic gates of the fault tree model are mapped to conditional probability tables to establish a static Bayesian network model for wind turbine support structure faults. Based on the Markov assumption and the time homogeneity assumption, the static Bayesian network model is extended into a dynamic Bayesian network model. The Bayesian network structure is determined based on the dynamic Bayesian network model.

3. The method as described in claim 2, characterized in that, Determining the Bayesian network structure based on the dynamic Bayesian network model includes: The risk of fault causes is quantified by using parameter learning methods to determine the risk probability distribution of the Bayesian network structure. The risk probability distribution is used to characterize the prior probability and conditional probability of faulty nodes in the Bayesian network structure.

4. The method as described in claim 3, characterized in that, The step of quantifying the risk of fault causes through parameter learning methods and determining the risk probability distribution of the Bayesian network structure includes: Objective probability parameters are obtained by learning parameters from the comprehensive database using a preset EM algorithm. Based on the experience of multiple industry experts, the risk of failure causes was quantified to obtain subjective probability parameters; By integrating the subjective probability parameters and the objective probability parameters, the risk probability distribution of the Bayesian network structure is obtained.

5. The method as described in claim 4, characterized in that, The objective probability parameters are obtained by learning parameters from the comprehensive database using a preset EM algorithm, including: Obtain the initial prior probability of the faulty node in the Bayesian network structure; Obtain historical monitoring data; The expected value of the missing data in the historical monitoring data is determined based on the initial prior probability. A learning dataset is generated based on the historical monitoring data and the expected values ​​of the missing data. Based on the learning dataset, the objective prior probability and objective conditional probability of the faulty node in the Bayesian network structure are determined, and the objective probability parameters are obtained.

6. The method as described in claim 5, characterized in that, The risk quantification of failure causes based on the experience of multiple industry experts yields subjective probability parameters, including: Obtain the weight of each expert in the expert panel; Obtain the subjective prior probabilities of each expert regarding the cause of each failure; Determine the probability of occurrence of each failure mode; The subjective conditional probability of the fault node in the Bayesian network structure is determined based on the occurrence probability of each fault mode. The subjective probability parameter is determined based on the weights of each expert, the subjective prior probability, and the subjective conditional probability.

7. The method as described in claim 6, characterized in that, Determining the probability of occurrence of each failure mode includes: Obtain OSD scores from various experts for each cause of the fault; Based on the OSD score, the RPN value of the cause of failure is determined, and the first evaluation vector is obtained by the risk value of each part under each cause of failure. Obtain the occurrence status of the fault causes to obtain multiple fault cause occurrence status vectors; The RPN value corresponding to the occurrence state of the fault cause is determined based on the first evaluation vector and the occurrence state vector. A second evaluation vector corresponding to the fault mode is determined based on the RPN value corresponding to the occurrence state of the fault cause. The second evaluation vector is used to characterize the probability of occurrence of the fault mode in the occurrence state.

8. The method as described in claim 7, characterized in that, The fault information and wind turbine system decomposition results from the pre-defined integrated database are used to summarize the failure modes and causes of offshore wind turbines, and a fault tree model of the offshore wind turbine support structure is established, including: Obtain fault information and support structure information of offshore wind turbines; The structure of the offshore wind turbine is decomposed based on the supporting structure information to obtain decomposition information; The failure mode and cause of failure of the offshore wind turbine are determined based on the fault information and the decomposition information. A fault tree model for offshore wind turbines is established based on the failure modes and failure causes.

9. The method as described in claim 8, characterized in that, The decomposed information includes multiple subsystems, each subsystem comprising multiple components. The multiple subsystems include at least one of the following: pile foundation subsystem, jacket support subsystem, pile cap support subsystem, tower subsystem, and auxiliary subsystem.

10. The method as described in claim 9, characterized in that, Before determining the risk probability distribution of the Bayesian network structure by quantifying the risk of fault causes through parameter learning methods, the method further includes: The dynamic Bayesian network model was iteratively optimized using the random forest K-fold cross-validation algorithm.

Images