Search

CN-121980715-A - Dynamic evaluation method, device, equipment and medium for time-varying reliability of offshore wind turbine

CN121980715ACN 121980715 ACN121980715 ACN 121980715ACN-121980715-A

Abstract

The invention discloses a dynamic evaluation method, device, equipment and medium for the time-varying reliability of an offshore wind turbine, which comprise the steps of obtaining multi-source heterogeneous monitoring data of the offshore wind turbine, carrying out edge pretreatment, establishing a depth operator network proxy model according to the multi-source heterogeneous monitoring data of the offshore wind turbine, carrying out offline training to obtain initial model parameters, calculating the output residual of the depth operator network proxy model according to predicted tower top displacement, basic dip angle time course and measured values, constructing an adaptive asynchronous weighted likelihood function frame, establishing a Bayesian asynchronous sequential update mechanism based on the arrival time difference of the monitoring data of the multi-source asynchronous monitoring system of the offshore wind turbine and combining the adaptive asynchronous weighted likelihood function frame, and carrying out failure probability threshold judgment and dynamic evaluation for the time-varying reliability based on the model parameters updated by the Bayesian asynchronous sequential update mechanism. The invention solves the problems of real-time early warning and intelligent operation and maintenance decision of the safety state of the fan structure in a complex marine environment.

Inventors

  • ZHANG DUO
  • LI XUEYOU
  • MENG ZHENZHU

Assignees

  • 中山大学

Dates

Publication Date
20260505
Application Date
20260409

Claims (10)

  1. 1. The method for dynamically evaluating the time-varying reliability of the offshore wind turbine is characterized by comprising the following steps of: Acquiring multi-source heterogeneous monitoring data of the offshore wind turbine through an offshore wind turbine multi-source asynchronous monitoring system, and performing edge preprocessing, wherein the multi-source heterogeneous monitoring data comprises offshore wind turbine vibration data, strain data, SCADA data and marine environment data; according to multi-source heterogeneous monitoring data of the offshore wind turbine, establishing a depth operator network proxy model with physical information embedded therein, and performing offline training to obtain initial model parameters; Calculating a depth operator network agent model output residual error according to the tower top displacement and the basic inclination angle time interval predicted by the depth operator network agent model and the actual measurement value transmitted by the offshore wind turbine multisource asynchronous monitoring system, and constructing a self-adaptive asynchronous weighted likelihood function frame; based on the difference of arrival time of monitoring data of the offshore wind turbine multisource asynchronous monitoring system, a Bayesian asynchronous sequential updating mechanism is established by combining an adaptive asynchronous weighted likelihood function framework; And executing failure probability threshold judgment and time-varying reliability dynamic evaluation based on the model parameters updated by the Bayesian asynchronous sequential updating mechanism.
  2. 2. The method for dynamically evaluating the reliability of the offshore wind turbine according to claim 1, wherein the steps of obtaining the multi-source heterogeneous monitoring data of the offshore wind turbine and performing the edge preprocessing comprise: Monitoring incoming wind conditions through a laser radar arranged at the top of a cabin of an offshore wind turbine, capturing early faults through vibration sensors arranged on a gear box and a generator bearing seat, monitoring fatigue accumulation through fiber bragg grating strain sensors stuck to the root of a blade and a key section of a tower barrel, and acquiring seabed flushing depth and wave load in real time through an acoustic Doppler wave velocity profiler deployed near a single pile foundation; Aiming at the heterogeneous sampling frequencies of the offshore wind turbine vibration data, the strain data, the SCADA data and the marine environment data, synchronizing the time references of the sensors, independently triggering the acquisition task at the edge computing unit, and marking the data freshness label to represent asynchronous delay; in an edge computing unit close to the sensor, performing Butterworth low-pass filtering on vibration data, performing wavelet threshold denoising on corresponding data, and marking abnormal values as missing; The laser radar data, the vibration data and the strain data are used as fast-changing data, the seabed scouring depth and the wave load are used as slow-changing data, the fast-changing data and the slow-changing data are aligned to a unified time axis, and a delay tag is generated for the asynchronously-arriving acoustic Doppler wave velocity profiler data; and extracting gear meshing frequency, bearing characteristic frequency energy and kurtosis indexes from the vibration data by the edge computing unit, realizing data dimension reduction, and transmitting the data to a land centralized control center.
  3. 3. The method for dynamically evaluating the reliability of the marine fan according to claim 2, wherein the steps of establishing a depth operator network proxy model embedded with physical information according to the multi-source heterogeneous monitoring data of the marine fan and obtaining initial model parameters through offline training include: mapping four types of heterogeneous data including laser radar wind speed time course, wave time course, seabed scouring depth time sequence and SCADA data into a high-dimensional function space; Designing Trunk network as fully-connected neural network, encoding space-time coordinates y= (t, z, d_scour) by an input layer, wherein t is a time variable, z is a height coordinate along a tower barrel, d_scour is a time-varying scouring depth, and generating a value vector of a p-dimensional depth space-time characteristic basic function at a given coordinate position by an output layer; Receiving discrete sampling points of an input function through a Branch network, automatically extracting wind speed spectrum frequency characteristics, wave load time sequence modes and soil parameter spatial distribution through a convolution layer, and generating a p-dimensional depth characteristic vector; Carrying out Monte Carlo sampling by using OpenSees finite element model, generating a plurality of groups of structural dynamic response samples under the combined action of wind and waves in four-dimensional parameter space of wind speed, effective wave height, soil non-drainage shear strength and scouring depth, wherein each group of structural dynamic response samples comprises tower top displacement, foundation dip angle and pile body bending moment as a data set for training a depth operator network proxy model; defining a physical constraint type composite loss function of a depth operator network proxy model; Dividing the data set into a training set and a verification set, performing offline training on the depth operator network proxy model for multiple times by using the training set in a land centralized control center, and freezing model parameters when the relative error of the verification set is smaller than a preset percentage to serve as initial model parameters.
  4. 4. The method for dynamically evaluating the reliability of a marine fan according to claim 2, wherein the calculating the depth operator network agent model output residual error according to the tower top displacement predicted by the depth operator network agent model, the basic inclination angle time course and the actual measurement value transmitted by the marine fan multi-source asynchronous monitoring system, and constructing the adaptive asynchronous weighted likelihood function framework comprises: The method comprises the steps of obtaining residual vectors by carrying out point-by-point difference on tower top displacement and basic inclination angle time interval predicted by a depth operator network proxy model and actual measurement values transmitted by an offshore wind turbine multisource asynchronous monitoring system, and decomposing the residual into inherent approximation errors of the depth operator network proxy model, measurement noise introduced by salt spray corrosion of an offshore sensor and environmental interference caused by extreme working conditions; the method comprises the steps of introducing a data freshness label, carrying out exponential decay weighting on residuals at different moments along with time step increase, respectively giving weights to high-frequency vibration residuals of early faults of offshore fans and wave height residuals of acoustic Doppler wave velocity profilers, and realizing multi-source heterogeneous data differential utilization; when the turbulence intensity monitored by the laser radar is larger than the preset intensity, the structural response residual error weight is increased to strengthen the safety precaution; And constructing an asynchronous weighted likelihood function frame according to each weight, and introducing an exponential forgetting factor to carry out recursive attenuation on the historical residual covariance matrix to form a self-adaptive asynchronous weighted likelihood function frame.
  5. 5. The method for dynamically evaluating the time-varying reliability of the offshore wind turbine according to claim 2, wherein the establishing a bayesian asynchronous sequential update mechanism based on the difference of arrival time of monitoring data of the offshore wind turbine multisource asynchronous monitoring system in combination with the adaptive asynchronous weighted likelihood function framework comprises: The method comprises the steps that for three types of asynchronous flow registration independent triggers of vibration, wind speed and wave, an event queue is ordered according to the data quality weight multiplied by the reciprocal of freshness; the prescribed sequential update sequence is vibration, wind speed, wave and scouring; when the turbulence intensity of the laser radar is larger than the preset intensity, the data period interpolation extrapolation compression of the acoustic Doppler wave velocity profiler is carried out, and the non-critical parameters are frozen for updating; Adopting variation inference, calculating a new posterior by taking the posterior at the last moment as a priori and combining an adaptive asynchronous weighted likelihood function framework when an event arrives, and asynchronously skipping over a missing data source; and when the scouring data at the last moment conflicts with the current vibration data on pile foundation bending moment estimation, updating the recent vibration weight and the long-term scouring weight.
  6. 6. The method for dynamically evaluating the time-varying reliability of an offshore wind turbine according to claim 1, wherein the performing the failure probability threshold judgment and the dynamic evaluation of the time-varying reliability based on the model parameters updated by the bayesian asynchronous sequential update mechanism comprises: Based on model parameters updated by a Bayes asynchronous sequential updating mechanism, respectively calculating the ultimate strength reliability structure of the tower, the foundation of pile foundation scouring instability reliability and the reliability fatigue of the blade fatigue crack; reliability integration is carried out on three series failure modes, and dynamic evaluation is realized by utilizing the coupling failure probability of pneumatic-hydrodynamic-structural-basic full links; continuously calculating a time-varying gradient of failure probability, and when the reliability reduction rate is monitored to be larger than a preset rate, judging that the structure is in an accelerated degradation stage, and triggering active early warning; Performing threshold grading early warning on the failure probability; Accessing marine weather forecast grid data, and driving a depth operator network agent model to conduct reliability evolution prediction of a future period so as to identify typhoon path deviation or wave attenuation windows; And when the failure probability of the predicted future period is smaller than the preset probability and the duration time is longer than the preset time, automatically generating a working window suggestion that the operation and maintenance ship can go out of sea.
  7. 7. The method for dynamically assessing the reliability of a time-varying offshore wind turbine of any of claims 1-6, further comprising: And generating fan structure health degree early warning and optimal operation and maintenance window decision suggestions according to reliability indexes of each component dynamically evaluated by the time-varying reliability.
  8. 8. A dynamic evaluation device for the time-varying reliability of an offshore wind turbine, the device comprising: The acquisition and edge preprocessing module is used for acquiring multi-source heterogeneous monitoring data of the offshore wind turbine through the multi-source asynchronous monitoring system of the offshore wind turbine and carrying out edge preprocessing, wherein the multi-source heterogeneous monitoring data comprise offshore wind turbine vibration data, strain data, SCADA data and marine environment data; the model building module is used for building a depth operator network proxy model with embedded physical information according to multi-source heterogeneous monitoring data of the offshore wind turbine and obtaining initial model parameters through offline training; The likelihood function framework building module is used for calculating the depth operator network agent model output residual error according to the tower top displacement and basic inclination angle time interval predicted by the depth operator network agent model and the actual measurement value transmitted by the offshore wind turbine multisource asynchronous monitoring system, and building a self-adaptive asynchronous weighted likelihood function framework; The updating mechanism establishing module is used for establishing a Bayesian asynchronous sequential updating mechanism based on the difference of arrival time of monitoring data of the offshore wind turbine multisource asynchronous monitoring system; and the time-varying reliability evaluation module is used for executing failure probability threshold judgment and time-varying reliability dynamic evaluation based on the model parameters updated by the Bayesian asynchronous sequential update mechanism.
  9. 9. A computer device comprising a processor and a memory for storing a program executable by the processor, wherein the processor, when executing the program stored in the memory, implements the method for dynamically assessing the time-varying reliability of an offshore wind turbine according to any one of claims 1-7.
  10. 10. A storage medium storing a program, wherein the program, when executed by a processor, implements the method for dynamically evaluating the time-varying reliability of an offshore wind turbine according to any one of claims 1-7.

Description

Dynamic evaluation method, device, equipment and medium for time-varying reliability of offshore wind turbine Technical Field The invention relates to a dynamic evaluation method, device, equipment and medium for the time-varying reliability of an offshore wind turbine, and belongs to the technical field of health monitoring and reliability evaluation of offshore wind power structures. Background At present, the complex marine environment (typhoon, rough sea, ocean current scouring and salt spray corrosion) enables the offshore wind turbine structure to bear wind-wave-current coupling power load and continuous environment corrosion for a long time, so that accidents such as fatigue crack of a tower barrel, scouring instability of pile foundation, blade fracture and the like are frequent. The prior offshore wind turbine structure health monitoring mainly relies on discrete monitoring means such as SCADA (Supervisory Control And Data Acquisition, data acquisition and monitoring control) systems, vibration sensors, strain gauges, wave radars and the like. The three aspects of the scheme are limited in that 1) sampling frequency isomerism is obvious, namely 10kHz of vibration data, 100Hz of strain data, 1Hz of SCADA data and 10 minutes of marine environment data are obtained, a traditional batch synchronous processing mode is required to wait for the arrival of the slowest data source, so that data timeliness is lost under extreme working conditions such as typhoons and the like and second-level early warning cannot be supported, 2) monitoring data quality difference is large, an offshore sensor is susceptible to salt spray corrosion to drift, strong electromagnetic interference is introduced into noise, ADCP wave data is distorted due to ocean current disturbance, the existing method mostly adopts equal weight fusion, old or low-quality data pollution model update and poor reliability assessment result robustness, 3) an agent model and a physical mechanism are disjointed, a pure data driving neural network can fit nonlinear mapping, but the structural mechanics conservation law is violated in a monitored data sparse area, so that extrapolation is unreliable, and the calculation time consumption of the traditional physical model cannot meet real-time requirements. In the aspect of reliability evaluation, the prior art mostly adopts static reliability indexes or offline Monte Carlo simulation, the update period is as long as hours or even days, and the rapid time-varying characteristics of typhoons, passing through the environment, burst flushing and the like cannot be captured. Some researches introduce a Bayesian updating framework, but still assume that the monitored data arrive synchronously, and the arrival sequence, delay difference and quality layering of the multi-source asynchronous data are not considered, so that the updating efficiency is low. In addition, the coupling association of the existing method for multiple failure modes (tower strength, pile foundation flushing and blade fatigue) is not enough, the weight is fixed under extreme working conditions, the influence of key safety parameters is difficult to adaptively strengthen, and early warning lag or false alarm rate is high. In summary, the prior art is difficult to simultaneously meet four-in-one requirements of multi-source heterogeneous monitoring data, extreme time-varying working conditions, real-time accurate evaluation and active decision look-ahead of offshore wind turbines, and a new method for dynamic reliability evaluation integrating depth operator networks (DeepONet), asynchronous sequential updating and self-adaptive weighting is needed, so that the problems of real-time sensing of structural safety states and intelligent operation and maintenance decision in complex marine environments are solved. Disclosure of Invention In view of the above, the invention provides a dynamic evaluation method, a device, computer equipment and a storage medium for the time-varying reliability of an offshore wind turbine, which solve the problems of real-time early warning and intelligent operation and maintenance decision of the safety state of the wind turbine structure in a complex marine environment through a physical information embedding and residual error self-adaptive weighting mechanism. The first aim of the invention is to provide a dynamic evaluation method for the time-varying reliability of the offshore wind turbine. The second object of the invention is to provide a dynamic evaluation device for the time-varying reliability of the offshore wind turbine. A third object of the present invention is to provide a computer device. A fourth object of the present invention is to provide a storage medium. The first object of the present invention can be achieved by adopting the following technical scheme: a method for dynamically evaluating the time-varying reliability of an offshore wind turbine, the method comprising: Acquiring multi-sour