An offshore wind turbine failure assessment system

Abstract

The application relates to the technical field of offshore wind turbine operation and maintenance, and discloses a system for evaluating the failure of an offshore wind turbine, which comprises a sensor data acquisition module, a data preprocessing module, a failure feature extraction unit, a failure probability calculation module, a maintenance decision module and the like. The system can acquire parameters such as the vibration and temperature of the wind turbine in real time, extract failure features by using a deep convolutional neural network after preprocessing, and calculate the failure probability of components by constructing a dynamic Bayesian network model. An improved ant colony optimization algorithm is used to generate an optimal maintenance path, and a dynamic risk evaluation module, an abnormal signal detection module and a maintenance strategy optimization module are simultaneously provided. The system can accurately evaluate the failure risk of the unit, timely discover abnormalities, optimize maintenance decisions, improve the operation reliability of the offshore wind turbine, reduce operation and maintenance costs, and ensure the stable and efficient operation of offshore wind power generation.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of offshore wind turbine operation and maintenance, and particularly relates to an offshore wind turbine failure evaluation system. BACKGROUND

[0002] With the increasing demand for clean energy worldwide, offshore wind power as an important way of renewable energy utilization has attracted widespread attention. Compared with land-based wind turbines, offshore wind turbines face more complex and harsh operating environments, such as strong winds, sea wave impact, salt spray corrosion, etc., which makes the failure rate of offshore wind turbines higher and the operation and maintenance more difficult.

[0003] Offshore wind turbines are exposed to complex environments for a long time, and key components bear huge mechanical stress and environmental corrosion. For example, blades not only have to withstand the aerodynamic force generated by strong winds, but also have to withstand the impact of sea spray and the erosion of salt spray, which makes the blades prone to cracks, wear and tear and other failures. The lubrication system of the gearbox is prone to failure under the conditions of high load and frequent start-stop, and the influence of high humidity in the sea water environment, which leads to gear wear and tooth surface gluing. The generator also faces the risk of difficult heat dissipation and salt spray damage to electrical insulation. Once these failures occur, not only will the wind turbine be shut down, affecting power supply, but also high maintenance costs will be incurred.

[0004] Currently, the operation and maintenance of offshore wind turbines mostly rely on traditional periodic maintenance and post-failure maintenance modes. Periodic maintenance is often carried out according to fixed time intervals or running time, and cannot accurately adjust the maintenance strategy according to the actual operating state of the unit. This results in some good condition components being over-maintained, wasting a lot of manpower, material and time resources; while some components that have already appeared potential failures are not timely discovered and handled, increasing the risk of sudden failures. Post-failure maintenance is carried out only after the failure occurs, during which the wind turbine is in a shutdown state, causing huge loss of power generation. For example, a serious gearbox failure may cause the wind turbine to shut down for weeks or even months, causing considerable economic losses to the wind farm.

[0005] In terms of failure evaluation, the existing evaluation methods lack accuracy and timeliness. Most methods rely only on a single type of sensor data, which cannot fully reflect the operating state of the unit. Moreover, the data processing and analysis means are relatively simple, and it is difficult to accurately extract failure features from complex monitoring data, resulting in a large error in the evaluation of component failure probability, which cannot provide reliable basis for operation and maintenance decisions. In addition, in terms of maintenance path planning and decision-making, there is a lack of effective optimization algorithms, which cannot comprehensively consider factors such as unit location distribution, failure frequency, environmental factors, etc., and it is difficult to develop an efficient and economic maintenance plan. SUMMARY

[0006] The offshore wind turbine failure evaluation system is provided to solve the problems in the background art.

[0007] To achieve the above object, the present application provides the following technical solution: an offshore wind turbine failure evaluation system, comprising:

[0008] A sensor data acquisition module is configured to acquire vibration, temperature, rotation speed and environmental parameters of the wind turbine in real time, and mark the parameters as original monitoring data.

[0009] A data preprocessing module is configured to perform noise reduction processing on the original monitoring data to generate a standardized data sequence.

[0010] A failure feature extraction unit is configured to extract a multi-dimensional failure feature vector from the standardized data sequence based on a deep convolutional neural network.

[0011] A failure probability calculation module is configured to construct a dynamic Bayesian network model according to the multi-dimensional failure feature vector and output a component failure probability.

[0012] A maintenance decision module is configured to generate an optimal maintenance path based on an improved ant colony optimization algorithm and send a decision instruction to an execution terminal.

[0013] The improved ant colony optimization algorithm has the following specific improvements:

[0014] The unit position distribution and fault frequency are extracted from historical maintenance records to initialize the pheromone concentration matrix.

[0015] A dynamic evaporation factor is introduced to adjust the pheromone evaporation rate according to the environmental wind speed level.

[0016] If the path selection probability exceeds a preset threshold, the path is determined to be an effective maintenance scheme.

[0017] Preferably, the data preprocessing module has the following specific method:

[0018] The original monitoring data is subjected to frequency domain decomposition by a wavelet transform algorithm to separate high-frequency noise and low-frequency effective signals.

[0019] The high-frequency noise components are filtered based on an adaptive threshold to reconstruct the time domain signal after noise reduction.

[0020] The reconstructed signal is normalized by a Z-score standardization method to generate a standardized data sequence.

[0021] Preferably, the failure feature extraction unit comprises:

[0022] A time series segmentation sub-module is configured to divide the standardized data sequence into multiple data segments according to a fixed time window.

[0023] The convolution kernel generation sub-module is configured to optimize the convolution kernel parameters based on a gradient descent algorithm to match the vibration modal characteristics of the unit;

[0024] The feature fusion sub-module is configured to generate a multi-dimensional failure feature vector by fusing the multi-scale convolution outputs through an attention mechanism.

[0025] Preferably, the method for constructing the dynamic Bayesian network model comprises the following steps:

[0026] The prior probability distribution and conditional dependency are extracted from the component failure library to construct a network topology structure;

[0027] The node conditional probability table is iteratively updated based on an expectation maximization algorithm to optimize the network parameters;

[0028] The multi-dimensional failure feature vector is input into the network, and the component failure posterior probability is calculated through a probability propagation algorithm.

[0029] Preferably, the method for adjusting the dynamic volatility factor comprises the following steps:

[0030] Real-time wind speed data are obtained from an environmental monitoring module, and a wind speed fluctuation coefficient is calculated;

[0031] The wind speed fluctuation coefficient is mapped to a volatility factor interval based on a Sigmoid function, and the formula is as follows:

[0032]

[0033] wherein β(t) is the volatility factor at time t, v(t) is the real-time wind speed, and k and v0 are preset adjustment parameters.

[0034] Preferably, the system further comprises:

[0035] The dynamic risk assessment module is configured to generate a real-time risk level according to the unit operating state;

[0036] The method for generating the real-time risk level comprises the following steps:

[0037] The component failure probability is obtained from the failure probability calculation module to construct a fuzzy membership function;

[0038] The membership degrees are weighted and aggregated based on a fuzzy inference rule library to output a risk level quantization value.

[0039] Preferably, the dynamic risk assessment module further comprises:

[0040] The risk mitigation sub-module is configured to trigger a hierarchical early warning signal according to the risk level;

[0041] The method for triggering the hierarchical early warning signal comprises the following steps:

[0042] If the risk level quantization value exceeds the first threshold value, a local shutdown instruction is activated and a backup unit is started;

[0043] If the risk level quantization value exceeds the second threshold value, an emergency repair request is sent to a monitoring center.

[0044] Preferably, the system further comprises:

[0045] An abnormal signal detection module is configured to identify a sudden abnormal signal based on an improved local outlier factor algorithm;

[0046] The specific implementation steps of the improved local outlier factor algorithm are as follows:

[0047] Time domain statistics and frequency domain energy ratios are extracted from the standardized data sequence as detection features;

[0048] A neighborhood radius parameter is dynamically adjusted according to the unit operation stage, and a local outlier factor score is calculated.

[0049] If the score exceeds an adaptive threshold value, the current data segment is marked as an abnormal state.

[0050] Preferably, the adaptive threshold value is determined by the following method:

[0051] The mean and variance of the outlier factor score are calculated from historical normal data.

[0052] An initial threshold value is set based on principles, and the threshold value range is updated according to real-time data distribution.

[0053] Preferably, the system further comprises:

[0054] A maintenance strategy optimization module is configured to iteratively update maintenance decision logic based on a genetic algorithm.

[0055] The iteration method of the genetic algorithm is as follows:

[0056] Historical maintenance paths are encoded as a chromosome population, and path length and risk value are defined as fitness functions.

[0057] High fitness individuals are reserved through a tournament selection strategy, and crossover and mutation operations are performed.

[0058] If the new generation population converges to the Pareto front, an optimal maintenance strategy set is output.

[0059] Compared with the prior art, the present application has the following beneficial effects:

[0060] The offshore wind turbine failure evaluation system based on the application has many significant beneficial effects. In data acquisition and processing, the sensor data acquisition module can acquire rich operation parameters and environmental parameters of the wind turbine in real time, covering vibration, temperature, speed, etc., providing a solid data foundation for comprehensively mastering the unit state. The data preprocessing module effectively removes noise interference and unifies the data scale through wavelet transform algorithm, adaptive threshold filtering and Z-score standardization method, so that the subsequent analysis is more accurate and the data availability is improved.

[0061] The failure feature extraction unit accurately extracts a multi-dimensional failure feature vector from the standardized data sequence with the help of a deep convolutional neural network. The time sequence segmentation submodule reasonably divides the data segments, the convolution kernel generation submodule optimizes the convolution kernel parameters to match the unit vibration modal characteristics, and the feature fusion submodule uses the attention mechanism to weight and fuse the multi-scale convolution outputs. This multi-link cooperative working mode greatly improves the accuracy and comprehensiveness of failure feature extraction, enabling earlier and more accurate detection of potential fault hazards.

[0062] The failure probability calculation module constructs a dynamic Bayesian network model, extracts prior probability distribution and conditional dependence relationship from the component failure library to construct the topological structure, optimizes the network parameters through the expectation maximization algorithm, and inputs the multi-dimensional failure feature vector to calculate the component failure posterior probability. Compared with traditional evaluation methods, it can more accurately evaluate the component failure probability and provide reliable basis for operation and maintenance decision.

[0063] The maintenance decision module initializes the pheromone concentration matrix based on the improved ant colony optimization algorithm, introduces a dynamic evaporation factor to adjust the pheromone evaporation rate according to the environmental wind speed level, determines the effective maintenance scheme by judging the path selection probability, and quickly plans the optimal maintenance path, significantly improving the maintenance efficiency and reducing the maintenance time and cost.

[0064] The dynamic risk assessment module generates real-time risk levels according to the unit operation state, realizes the quantitative evaluation of the unit risk by constructing fuzzy membership functions and weighted aggregation based on fuzzy reasoning rule base. The risk mitigation submodule triggers graded warning signals according to the risk level, activates local shutdown instructions and starts standby units when the risk is low, and sends emergency repair requests to the monitoring center when the risk is high, effectively ensuring the safe and stable operation of the unit and reducing the failure loss.

[0065] The abnormal signal detection module extracts multi-dimensional detection features from the standardized data sequence based on an improved local outlier factor algorithm, dynamically adjusts a neighborhood radius parameter to calculate a local outlier factor score, and judges an abnormal state through an adaptive threshold, so that sudden abnormal signals can be found in time and measures can be taken in advance to prevent faults. BRIEF DESCRIPTION OF DRAWINGS

[0066] Figure 1 A working principle diagram of the offshore wind turbine failure evaluation system is provided.

[0067] Figure 2 A working principle diagram of the failure feature extraction unit is provided.

[0068] Figure 3 A working principle diagram of the dynamic risk assessment and risk mitigation is provided.

[0069] Figure 4 A working principle diagram of the maintenance strategy optimization is provided. DETAILED DESCRIPTION

[0070] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of the present application.

[0071] Please refer to Figures 1-4 The present application provides an offshore wind turbine failure evaluation system, which mainly comprises a sensor data acquisition module, a data preprocessing module, a failure feature extraction unit, a failure probability calculation module and a maintenance decision module. The modules work together to effectively evaluate the failure of offshore wind turbines and make maintenance decisions. The overall implementation scheme is as follows:

[0072] The sensor data acquisition module is responsible for real-time acquisition of vibration, temperature, speed and environmental parameters of the wind turbine, which constitutes the original monitoring data. This module uses high-precision sensors, such as vibration sensors installed on key components of the wind turbine, such as blades, gearboxes, generators, etc., which can accurately capture the vibration of the components during operation; temperature sensors are distributed in heat-generating parts to monitor temperature changes; the speed sensor is installed on the rotating shaft to measure the speed of the wind turbine in real time. At the same time, environmental monitoring equipment is used to collect environmental parameters, including wind speed, wind direction, air temperature, air pressure, etc., to ensure that comprehensive and accurate raw data are obtained.

[0073] The data preprocessing module receives the original monitoring data from the sensor data acquisition module and performs noise reduction processing to generate standardized data sequences. First, the original monitoring data is decomposed in the frequency domain by wavelet transform algorithm, separating high-frequency noise from low-frequency effective signal. Then, based on the adaptive threshold, the high-frequency noise component is filtered to remove noise interference and reconstruct the denoised time domain signal. Finally, the reconstructed signal is normalized by the Z-score standardization method to make different types of data comparable for subsequent analysis.

[0074] The failure feature extraction unit extracts a multi-dimensional failure feature vector from the standardized data sequence based on a deep convolutional neural network. This unit first divides the standardized data sequence into multiple data segments according to a fixed time window to prepare for subsequent convolution operations. Then, the gradient descent algorithm is used to optimize the convolution kernel parameters to better match the vibration modal characteristics of the unit. Finally, the attention mechanism is used to weight and fuse the multi-scale convolution outputs to obtain a multi-dimensional failure feature vector containing rich information.

[0075] The failure probability calculation module constructs a dynamic Bayesian network model based on the multi-dimensional failure feature vector to output the component failure probability. This module extracts the prior probability distribution and conditional dependence from the component fault library to construct the network topology. Then, based on the expectation maximization algorithm, the node conditional probability table is iteratively updated to optimize the network parameters. Finally, the multi-dimensional failure feature vector is input into the network, and the component failure posterior probability is calculated by the probability propagation algorithm.

[0076] The maintenance decision module generates the optimal maintenance path based on the improved ant colony optimization algorithm and sends the decision instructions to the execution terminal. The improved ant colony optimization algorithm extracts the unit location distribution and fault frequency from the historical maintenance records to initialize the pheromone concentration matrix. A dynamic evaporation factor is introduced to adjust the pheromone evaporation rate according to the environmental wind speed level. If the path selection probability exceeds the preset threshold, the path is determined to be an effective maintenance scheme. In this way, the optimal maintenance path can be quickly and accurately planned, improving the maintenance efficiency.

[0077] The implementation of the present application will be further described below in connection with Examples 1 to 6.

[0078] Example 1

[0079] In this embodiment, the specific implementation of the data preprocessing module and the failure feature extraction unit is further described.

[0080] The wavelet transform algorithm plays a key role in processing the original monitoring data of the data preprocessing module. Wavelet transform is a time-frequency analysis method that can analyze signals at different time and frequency resolutions. For the original monitoring data of offshore wind turbines, it contains signals of various frequency components, including low-frequency effective signals reflecting the normal operation state of the unit and high-frequency noise generated by environmental interference, equipment noise, etc.

[0081] By using the wavelet transform algorithm to perform frequency domain decomposition on the original monitoring data, these signals of different frequencies can be separated. For example, during the decomposition process, a suitable wavelet basis function can be selected, such as the db4 wavelet basis, which has good time-frequency localization characteristics and can better adapt to the characteristics of wind turbine data. After decomposition, the coefficients of different frequency subbands are obtained, among which the coefficients of high-frequency subbands mainly contain noise information, and the coefficients of low-frequency subbands retain the effective information of the unit operation.

[0082] When filtering the high-frequency noise components based on the adaptive threshold, the threshold needs to be determined dynamically according to the characteristics of the data. A commonly used method is to determine the threshold based on the statistical characteristics of the noise. For example, the standard deviation of the high-frequency subband coefficients is calculated, and the threshold is determined according to a certain empirical formula. Assuming that the standard deviation of the high-frequency subband coefficients is σ, then the threshold T = kσ, where k is an empirical constant, usually taking a value between 2-3. In this way, high-frequency noise can be effectively removed while retaining the characteristics of low-frequency effective signals.

[0083] Reconstructing the denoised time-domain signal is to perform inverse wavelet transform on the filtered high-frequency subband coefficients and low-frequency subband coefficients. Inverse wavelet transform can convert frequency domain signals back to time domain to obtain denoised signals. During reconstruction, the accuracy and stability of the transformation need to be ensured to ensure that the denoised signals can truly reflect the operating state of the unit.

[0084] The reconstructed signal is normalized using the Z-score standardization method. The Z-score standardization formula is: where X is the original data, μ is the mean of the data, and σ is the standard deviation of the data. Through this standardization process, data of different magnitudes and distributions can be converted to data with a mean of 0 and a standard deviation of 1, making subsequent analysis and processing more convenient.

[0085] The time sequence segmentation submodule in the failure feature extraction unit divides the standardized data sequence into multiple data segments according to a fixed time window. The selection of the time window needs to consider the operating characteristics of the unit and the variation frequency of the data. For example, for the vibration data of a wind turbine, the time window can be selected to be short, such as 1 second, because the variation is relatively frequent; for the temperature data, the time window can be selected to be long, such as 10 minutes, because the variation is relatively slow. By reasonably selecting the time window, the data features can be preserved while improving the efficiency of subsequent processing.

[0086] The convolution kernel generation submodule optimizes the convolution kernel parameters based on the gradient descent algorithm. The gradient descent algorithm is a commonly used optimization algorithm that adjusts the parameters to gradually reduce the value of the objective function. In this module, the objective function can be the error between the predicted value and the true value. The parameters of the convolution kernel determine the extraction ability of the convolution operation on the data features. By continuously optimizing the convolution kernel parameters through the gradient descent algorithm, they can better match the vibration modal characteristics of the unit. For example, during the training process, the data segments are input into the convolutional neural network, the error between the predicted value and the true value is calculated, and then the convolution kernel parameters are adjusted according to the gradient of the error, and the iteration is continuously performed until the error reaches a small value.

[0087] The feature fusion submodule fuses the multi-scale convolution outputs through an attention mechanism. The attention mechanism can automatically learn the importance of different scale convolution outputs and give higher weights to important features. In the failure feature extraction of wind turbines, different scales of convolution can capture different levels of features, such as small-scale convolution that can capture local detailed features and large-scale convolution that can capture overall trend features. By weighting and fusing these different scale convolution outputs through the attention mechanism, a multi-dimensional failure feature vector containing more rich information can be obtained. For example, assuming there are three scale convolution outputs F1, F2, and F3, the attention mechanism calculates their weights w1, w2, and w3, respectively, where w1+w2+w3=1, and the fused feature vector F=w1F1+w2F2+w3F3, which can highlight important features and improve the accuracy of failure feature extraction.

[0088] Embodiment 2:

[0089] This embodiment details the construction method of the dynamic Bayesian network model.

[0090] In constructing the dynamic Bayesian network model, the prior probability distribution and conditional dependence relationship are first extracted from the component failure library to construct the network topology structure. The component failure library is obtained by collecting, organizing and analyzing a large amount of historical failure data of wind turbines. For example, for the gearbox component of a wind turbine, the failure library records the probability of gearbox failure under different operating conditions and the correlation between gearbox failure and failure of other components.

[0091] In extracting the prior probability distribution, it is assumed that the probability of gearbox failure under normal operating conditions is P(G=1), where G represents the failure state of the gearbox, G=1 represents failure, and G=0 represents normal operation. Through statistical analysis of historical data, the specific value of P(G=1) can be obtained. At the same time, the conditional dependence between gearbox failure and other component failures, such as generator failure E, i.e. P(G=1|E=1) and P(G=1|E=0), needs to be extracted, which respectively represent the probability of gearbox failure under the condition of generator failure and normal operation.

[0092] Based on these prior probability distributions and conditional dependencies, the network topology structure is constructed. The nodes in the network represent the components of the wind turbine, and the edges represent the dependencies between the components. For example, if there is a dependency between gearbox failure and generator failure, the nodes representing the gearbox and generator are connected in the network, and the corresponding conditional probability is labeled.

[0093] After the network topology structure is constructed, the node conditional probability table is iteratively updated based on the expectation maximization algorithm to optimize the network parameters. The expectation maximization algorithm consists of two steps: E-step and M-step. In the E-step, the posterior probability of each node is calculated based on the current network parameters. For example, for node G, P(G|O) is calculated given the observed data O. In the M-step, the conditional probability table of the node is updated based on the posterior probability calculated in the E-step. By continuously iterating the E-step and M-step, the network parameters gradually converge to the optimal value.

[0094] The multi-dimensional failure feature vector is input into the network, and the component failure posterior probability is calculated by the probability propagation algorithm. There are various probability propagation algorithms, such as joint tree algorithm, variable elimination method, etc. Taking the joint tree algorithm as an example, first convert the dynamic Bayesian network into a joint tree structure, then perform message passing in the joint tree. The multi-dimensional failure feature vector is input into the joint tree as observed data, and the posterior probability of each component node is calculated through message passing. For example, the posterior probability of gearbox failure P(G=1|O) is calculated, where O is the observed data represented by the multi-dimensional failure feature vector. In this way, the failure probability of the component can be accurately evaluated to provide a basis for subsequent maintenance decisions.

[0095] Example 3:

[0096] This example mainly introduces the adjustment method of dynamic volatility factors.

[0097] Dynamic volatility factors play an important role in the improved ant colony optimization algorithm. They can adjust the pheromone evaporation rate according to the environmental wind speed level, thereby improving the search efficiency and accuracy of the algorithm.

[0098] Real-time wind speed data is obtained from the environmental monitoring module, and the wind speed fluctuation coefficient is calculated. The wind speed fluctuation coefficient can be calculated in various ways, for example, by calculating the ratio of the standard deviation to the mean of the wind speed over a period of time. Assuming that the wind speed data is v(t) in the time interval [t1, t2], the wind speed mean Wind speed standard deviation Wind speed fluctuation coefficient

[0099] The wind speed fluctuation coefficient is mapped to the evaporation factor interval based on the Sigmoid function, and the formula is: where β(t) is the evaporation factor at time t, which determines the evaporation rate of pheromone. v(t) is the real-time wind speed, which is the wind speed value at the current time obtained from the environmental monitoring module. k and v0 are preset adjustment parameters, k controls the slope of the Sigmoid function, affecting the sensitivity of the evaporation factor to changes in wind speed; v0 is a reference wind speed value used to adjust the center position of the Sigmoid function.

[0100] For example, when the wind speed fluctuation is large, i.e., the wind speed fluctuation coefficient ω is large, it indicates that the environmental wind speed is unstable, and the evaporation factor β(t) calculated by the Sigmoid function will increase accordingly. A larger evaporation factor means that the pheromone evaporation speed is faster, which can make the ant colony adapt to environmental changes faster and avoid the algorithm falling into a local optimal solution. Conversely, when the wind speed fluctuation is small, the evaporation factor β(t) will decrease, and the pheromone evaporation speed will slow down, which is beneficial to the algorithm for fine search in the local area to find a better maintenance path.

[0101] In practical applications, the preset adjustment parameters k and v0 need to be adjusted reasonably according to the operating environment and historical data of the wind turbine. Through multiple experiments and data analysis, a set of optimal parameter values can be found, so that the dynamic evaporation factor can better adapt to different wind speed conditions, improve the performance of the improved ant colony optimization algorithm, and thus more accurately generate the optimal maintenance path.

[0102] Example 4:

[0103] This embodiment describes in detail the specific implementation of the dynamic risk assessment module and the risk mitigation sub-module contained therein.

[0104] The dynamic risk assessment module generates real-time risk levels based on the operating state of the unit. First, obtain the component failure probability from the failure probability calculation module, and construct a fuzzy membership function. For different components of the wind turbine, such as blades, gearboxes, generators, etc., the influence of their failure probability on the overall risk of the unit is different. For example, blade failure may cause severe vibration or even shutdown of the unit, which has a greater impact on the risk of the unit; while the failure of some auxiliary components has a relatively small impact on the risk of the unit.

[0105] For each component, a corresponding fuzzy membership function is constructed according to the size of its failure probability. Taking the gearbox as an example, assuming that the gearbox failure probability is P, a triangular fuzzy membership function can be constructed. When P is less than a certain lower threshold P low , the membership is 0, indicating that the risk is very low; when P is between P low and P mid , the membership rises linearly; when P is equal to P mid , the membership is 1; when P is between P mid and P high , the membership decreases linearly; when P is greater than P high , the membership is 0. Here, P low , P mid , and P high are threshold values determined according to historical data and experience.

[0106] The membership degrees are aggregated by weighting based on the fuzzy inference rule base to output the risk level quantization value. The fuzzy inference rule base is established according to expert experience and a large amount of historical data. For example, when the membership degrees of multiple key components are high, the risk level quantization value should be correspondingly increased; when only a few auxiliary components have high membership degrees, the risk level quantization value is relatively small. By weighted aggregation, the fuzzy membership degrees of each component are considered comprehensively to obtain a quantization value that can reflect the overall risk status of the unit.

[0107] The risk mitigation submodule triggers a hierarchical warning signal according to the risk level. If the risk level quantization value exceeds the first threshold value, a local shutdown instruction is activated and a standby unit is started. The first threshold value is determined according to the safety operation standard of the unit and historical failure data. When the risk level quantization value exceeds this threshold value, it indicates that the unit has certain safety hazards and may cause serious damage to the unit. At this time, the local shutdown instruction is activated to stop the operation of the part of the unit where the faulty component is located, avoiding further expansion of the fault. At the same time, the standby unit is started to ensure that the power output of the entire wind farm is not greatly affected.

[0108] If the risk level quantization value exceeds the second threshold value, an emergency repair request is sent to the monitoring center. The second threshold value is usually higher than the first threshold value. When the risk level quantization value exceeds this threshold value, it indicates that the unit is in a serious danger state and needs to be repaired immediately. The emergency repair request is sent to the monitoring center, which can arrange maintenance personnel to go to the scene for maintenance in time, reduce downtime, and reduce economic losses.

[0109] Through the cooperative work of the dynamic risk assessment module and the risk mitigation submodule, potential risks of wind turbines can be discovered in time, and appropriate measures can be taken to deal with them, ensuring the safe and stable operation of wind turbines.

[0110] Example 5:

[0111] This embodiment focuses on the abnormal signal detection module and the determination method of adaptive threshold.

[0112] The abnormal signal detection module identifies the sudden abnormal signal based on the improved local outlier factor algorithm. First, the time domain statistics and frequency energy ratio are extracted from the standardized data sequence as detection features. The time domain statistics can include mean, standard deviation, peak value, etc. For example, the mean μ and standard deviation σ of the standardized data sequence within a certain time window are calculated, and these statistics can reflect the concentration trend and dispersion degree of the data. The frequency energy ratio is calculated by Fourier transform of the data, and the energy of different frequency bands accounts for the proportion of the total energy. For example, the frequency range is divided into multiple sub-frequency bands, and the energy E i of each sub-frequency band is calculated. Then the frequency energy ratio is where n is the number of sub-frequency bands.

[0113] The neighborhood radius parameter is dynamically adjusted according to the operating phase of the unit, and the local outlier factor score is calculated. In different operating phases of the wind turbine, the distribution characteristics of the data will be different. For example, in the starting phase and the stable running phase, the change law and fluctuation range of the data are different. In the starting phase, as the parameters of the unit gradually stabilize, the data fluctuates greatly, and the neighborhood radius parameter can be appropriately increased; in the stable running phase, the data is relatively stable, and the neighborhood radius parameter can be appropriately reduced.

[0114] The calculation of the local outlier factor score is based on the distance and density relationship between the data point and other data points in its neighborhood. Assuming that there are k data points in the neighborhood of data point x, the distance d(x, x i ) between x and these k data points is calculated, i.e. d(x, x LOF ), i = 1, 2, …, k. Then the local reachable density lrd(x) of the data points in the neighborhood is calculated, and the local outlier factor score LOF(x) of x is the average ratio of the local reachable density of x to the local reachable density of other data points in the neighborhood.

[0115] If the score exceeds the adaptive threshold, the current data segment is marked as abnormal. The determination method of the adaptive threshold is as follows: the mean μ LOF and the variance σ LOF of the outlier factor score are calculated from the historical normal data. Based on the 3σ principle, the initial threshold is set, i.e. the initial threshold T0= μ LOF + 3σ LOF . The 3σ principle is a commonly used statistical method. Under normal circumstances, the probability of data points falling within the range of mean plus or minus 3 times the standard deviation is about 99.7%, so the data points exceeding this range are considered abnormal.

[0116] The threshold range is updated according to real-time data distribution. As the wind turbine operates, the distribution of data may change. For example, due to factors such as equipment aging, environmental changes, etc., the fluctuation of data may gradually increase. Therefore, it is necessary to update the threshold range according to real-time data. The mean and variance of the outlier factor score of historical normal data can be recalculated every certain time interval, and then the threshold is updated. Assuming that the current time is t, the last time the threshold was updated is t-Δt, and new normal data is collected in the time interval [t-Δt, t], the mean and variance are recalculated and variance The updated threshold In this way of dynamically updating the threshold, the changes in data can be better adapted to improve the accuracy of abnormal signal detection.

[0117] Embodiment 6:

[0118] This embodiment mainly details the maintenance strategy optimization module and the genetic algorithm based thereon.

[0119] The maintenance strategy optimization module iteratively updates the maintenance decision logic with the help of genetic algorithm, aiming to find a better set of maintenance strategies, thereby improving the maintenance efficiency of offshore wind turbines, reducing operation and maintenance costs, and ensuring stable operation of the units.

[0120] The iteration process of the genetic algorithm first encodes the historical maintenance path into a chromosome population. The historical maintenance path contains a series of operation steps and sequences taken in the past when maintaining the wind turbine. Converting these paths into chromosome form facilitates genetic algorithm operations. For example, using integer encoding, each maintenance operation in the maintenance path corresponds to an integer, which is arranged in sequence according to the operation execution order to form a chromosome. Assuming that the wind turbine has blade inspection, gearbox maintenance, generator repair, etc. maintenance operations, represented by 1, 2, and 3 respectively, then a maintenance path "first check the blade, then repair the generator, and finally maintain the gearbox" can be encoded as [1, 3, 2]. By encoding a large number of historical maintenance paths, an initial chromosome population is formed.

[0121] Define path length and risk value as fitness function. Path length reflects the complexity and time cost of the maintenance process, and shorter path means more efficient maintenance process. The risk value is related to the failure probability of the wind turbine components, considering that if the components with high failure probability are given priority in maintenance, the overall risk of the unit can be reduced. For example, according to the component failure probability calculation module, each component is assigned a risk weight. Assuming that the failure probabilities of the blade, gearbox, and generator are P1, P2, and P3, respectively, and the corresponding risk weights are w1, w2, and w3, the maintenance sequence of each component in the maintenance path is [i1, i2, …, in], then the risk value of the path is nIf RiskValue = 0, then the calculation of the risk value can be The path length is Length = n. The fitness function Fitness can comprehensively consider the path length and the risk value, such as where a is a weight coefficient, and the value is between 0 and 1, used to balance the influence of the path length and the risk value on the fitness, which can be adjusted according to actual requirements, for example, by multiple tests to find the a value that makes the maintenance effect optimal.

[0122] High fitness individuals are reserved by the tournament selection strategy, and crossover and mutation operations are performed. The tournament selection strategy is to randomly select a certain number of individuals, such as k individuals (k is generally taken as 3-5), from the population, and select the individual with the highest fitness in the k individuals to enter the next generation population. Repeat this process until the number of individuals in the new generation population reaches the set value. This selection method can ensure that individuals with higher fitness have a greater chance of being reserved, promoting the evolution of the population to a better direction.

[0123] The crossover operation simulates the gene exchange process in biological genetics. Two chromosomes are randomly selected as parents from the current population, for example, selecting chromosomes A = [1, 2, 3, 4] and B = [5, 6, 7, 8], and randomly determining a crossover point, assuming the crossover point is 2, then the child chromosomes generated after crossover A ′ = [1, 2, 7, 8], B ′ = [5, 6, 3, 4]. Through the crossover operation, the excellent genes between different chromosomes can be combined, which may generate a better maintenance path.

[0124] Mutation operation is to prevent the algorithm from converging too early and maintain the diversity of the population. With a certain mutation probability (such as 0.01-0.1), the genes in the chromosome are randomly changed. For example, for the chromosome [1, 2, 3, 4], if the mutation probability is 0.05, and a gene site is selected for mutation, assuming that the 3rd site is selected, the original gene 3 may be randomly changed to other legal maintenance operation codes. Mutation operation can introduce new gene combinations to the population, avoiding the algorithm falling into a local optimal solution.

[0125] If the new generation converges to the Pareto front, the optimal maintenance strategy set is output. The Pareto front refers to a set of solutions in a multi-objective optimization problem that cannot be further optimized for one objective without sacrificing another. In maintenance strategy optimization, path length and risk value are two conflicting objectives. When the population evolves to a certain extent and cannot find solutions that optimize both objectives, it is considered that the population converges to the Pareto front. At this time, each solution on the Pareto front is the optimal maintenance strategy under different trade-offs. These strategies are combined into a set, and the operation and maintenance personnel can select the most suitable maintenance strategy from the set according to the actual situation, such as maintenance resources, time limits, etc., to achieve more scientific and efficient maintenance of offshore wind turbines.

[0126] It should be noted that, in this document, relational terms such as first and second and the like can be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0127] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, alternatives, and variations can be made in the embodiments without departing from the spirit and scope of the present application as defined by the appended claims and their equivalents.

Claims

1. A failure assessment system based on offshore wind turbines, characterized in that, include: The sensor data acquisition module is used to acquire the vibration, temperature, speed and environmental parameters of the wind turbine in real time, and mark the parameters as raw monitoring data; The data preprocessing module is used to perform noise reduction processing on the raw monitoring data to generate a standardized data sequence; The failure feature extraction unit is used to extract multidimensional failure feature vectors from a standardized data sequence based on a deep convolutional neural network. The failure probability calculation module is used to construct a dynamic Bayesian network model based on the multidimensional failure feature vector and output the component failure probability. The maintenance decision module is used to generate the optimal maintenance path based on the improved ant colony optimization algorithm and send the decision instructions to the execution terminal. The improved ant colony optimization algorithm is specifically improved as follows: Extract the unit location distribution and fault frequency from historical maintenance records, and initialize the pheromone concentration matrix; A dynamic volatile factor is introduced to adjust the pheromone evaporation rate according to the ambient wind speed level; If the probability of path selection exceeds a preset threshold, the path is determined to be a valid maintenance solution. The method for adjusting the dynamic volatile factor is as follows: Real-time wind speed data is obtained from the environmental monitoring module, and the wind speed fluctuation coefficient is calculated. Based on the Sigmoid function mapping of wind speed fluctuation coefficient to the volatile factor range, the formula is as follows: ,in, for Constantly volatile factors, For real-time wind speed, and These are preset adjustment parameters.

2. The failure assessment system for offshore wind turbines according to claim 1, characterized in that, The specific method of the data preprocessing module is as follows: The original monitoring data is decomposed in the frequency domain by wavelet transform algorithm to separate high-frequency noise from low-frequency effective signal; High-frequency noise components are filtered based on adaptive thresholds to reconstruct the denoised time-domain signal. The reconstructed signal is normalized using the Z-score normalization method to generate a normalized data sequence.

3. The failure assessment system based on offshore wind turbines according to claim 2, characterized in that, The failure feature extraction unit includes: The time-series segmentation submodule is used to divide a standardized data sequence into multiple data segments according to a fixed time window; The convolution kernel generation submodule is used to optimize the convolution kernel parameters based on the gradient descent algorithm to match the vibration mode characteristics of the unit. The feature fusion submodule is used to generate a multi-dimensional failure feature vector by weighted fusion of multi-scale convolution outputs through an attention mechanism.

4. The failure assessment system for offshore wind turbines according to claim 3, characterized in that, The method for constructing the dynamic Bayesian network model is as follows: Extract prior probability distributions and conditional dependencies from the component fault database to construct the network topology; The network parameters are optimized by iteratively updating the node conditional probability table based on the expectation-maximization algorithm. The multidimensional failure feature vector is input into the network, and the posterior probability of component failure is calculated through the probability propagation algorithm.

5. The failure assessment system for offshore wind turbines according to claim 1, characterized in that, Also includes: The dynamic risk assessment module is used to generate real-time risk levels based on the unit's operating status; The method for generating the real-time risk level is as follows: Obtain the component failure probability from the failure probability calculation module and construct a fuzzy membership function; The membership degree is weighted and aggregated based on the fuzzy reasoning rule base to output a quantitative value of risk level.

6. The failure assessment system based on offshore wind turbines according to claim 5, characterized in that, The dynamic risk assessment module also includes: The risk mitigation submodule is used to trigger graded early warning signals based on the risk level. The triggering method for the graded early warning signal is as follows: If the risk level quantification value exceeds the first threshold, a partial shutdown command will be activated and a standby unit will be started. If the risk level quantification value exceeds the second threshold, an emergency repair request will be sent to the monitoring center.

7. The failure assessment system for offshore wind turbines according to claim 1, characterized in that, Also includes: An abnormal signal detection module is used to identify sudden abnormal signals based on an improved local outlier factor algorithm. The specific implementation steps of the improved local outlier factor algorithm are as follows: Time-domain statistics and frequency-domain energy ratios are extracted from standardized data sequences as detection features; The neighborhood radius parameter is dynamically adjusted according to the unit operation stage, and the local outlier factor score is calculated. If the score exceeds the adaptive threshold, the current data segment is marked as an abnormal state.

8. The failure assessment system for offshore wind turbines according to claim 7, characterized in that, The method for determining the adaptive threshold is as follows: Calculate the mean and variance of outlier scores from historical normal data; An initial threshold is set based on principles, and the threshold range is updated by sliding according to the real-time data distribution.

9. The failure assessment system for offshore wind turbines according to claim 1, characterized in that, Also includes: The maintenance strategy optimization module is used to iteratively update the maintenance decision logic based on a genetic algorithm. The iterative method of the genetic algorithm is as follows: The historical maintenance path is encoded as a chromosome population, and the path length and risk value are defined as fitness functions. High-fit individuals are retained through a tournament selection strategy, and crossover and mutation operations are performed. If the new generation of population converges to the Pareto front, then the optimal set of maintenance strategies is output.

Images