Search

CN-122022776-A - Multi-source heterogeneous data fusion and state evaluation method, device and medium for offshore wind turbine

CN122022776ACN 122022776 ACN122022776 ACN 122022776ACN-122022776-A

Abstract

The invention discloses a method, a device and a medium for multi-source heterogeneous data fusion and state assessment of an offshore wind turbine, wherein the method comprises the steps of constructing an offshore wind turbine multi-source heterogeneous database, preprocessing data, mapping the multi-source heterogeneous data preprocessed by the data into a unified embedded representation space by utilizing a physical modeling and deep learning technology, constructing a characteristic interaction operator based on a multi-head cross attention mechanism in the unified embedded representation space, executing characteristic decoupling, extracting a panoramic service state vector, introducing the panoramic service state vector into a variational Bayesian calibration framework of an integrated physical information neural network as a global observation item, realizing dynamic calibration of digital twin physical parameters, and generating an offshore wind turbine service state assessment report with uncertainty quantitative support. The invention effectively solves the problems of data island, evaluation conservation, decision delay and the like, and generates the service state evaluation report with physical interpretability and uncertainty support.

Inventors

  • ZHANG DUO
  • LI XUEYOU
  • MENG ZHENZHU

Assignees

  • 中山大学

Dates

Publication Date
20260512
Application Date
20260409

Claims (10)

  1. 1. The multi-source heterogeneous data fusion and state evaluation method for the offshore wind turbine is characterized by comprising the following steps of: acquiring offshore wind turbine running state data, structural power response data, external environment meteorological data, key part visual morphology and full life cycle operation and maintenance records, and constructing an offshore wind turbine multi-source heterogeneous database; carrying out data preprocessing on the multi-source heterogeneous database of the offshore wind turbine, wherein the data preprocessing comprises data cleaning and space-time alignment; mapping the multi-source heterogeneous data subjected to data preprocessing to a unified embedded representation space by utilizing a physical modeling and deep learning technology; Constructing a characteristic interaction operator based on a multi-head cross attention mechanism in the unified embedded representation space, performing characteristic decoupling, and extracting a fusion characteristic vector reflecting the panoramic running state of the offshore wind turbine as a panoramic service state vector; and introducing the panoramic service state vector serving as a global observation item into a variational Bayesian calibration framework of the integrated physical information neural network, realizing dynamic calibration of the digital twin physical parameters, and generating an offshore wind turbine service state evaluation report with uncertainty quantitative support.
  2. 2. The method for multi-source heterogeneous data fusion and state assessment of an offshore wind turbine according to claim 1, wherein the obtaining of offshore wind turbine operational state data, structural dynamic response data, external environmental weather data, critical-part visual morphology and full life cycle operation and maintenance records comprises: The system comprises an offshore wind turbine running state data, structural power response data, external environment weather data, critical part visual morphology and full life cycle operation and maintenance records, and is used for constructing an offshore wind turbine multi-source heterogeneous database integrating low-frequency time sequences, high-frequency waveforms, continuous video streams, discrete high-definition images and semi-structured texts.
  3. 3. The method for multi-source heterogeneous data fusion and state assessment of an offshore wind turbine according to claim 2, wherein the step of synchronously collecting the offshore wind turbine running state data, the structural power response data, the external environment weather data, the critical part visual morphology and the full life cycle operation and maintenance records by the offshore wind turbine data collection system comprises the following steps: Acquiring low-frequency time sequence data reflecting the running state of the offshore wind turbine through an offshore wind turbine SCADA system, wherein the low-frequency time sequence data comprises the rotating speed of the blades, the generating capacity, the generating power, the voltage, the current, the pitch angle, the yaw angle and the cabin position; vibration sensors which are deployed at the top of the tower, the engine room and key basic parts of the tower through a CMS vibration monitoring system are used for acquiring high-frequency acceleration waveform signals, tower shaking signals and blade waving and vibrating signals of the generator bearing and the offshore wind turbine tower; Acquiring external environment weather data acting on an offshore wind turbine structure through a cabin wind measuring system and a sea surface weather buoy, wherein the external environment weather data comprises wind speed, wind direction, tide level, effective wave height, wave direction, environment temperature and humidity and atmospheric pressure; The method comprises the steps that through industrial cameras arranged in an offshore wind turbine cabin, at the bottom of the tower and on a key structural surface, continuous visual images of icing on the surface of a monitoring blade, shaking of a tower barrel and rotation of the blade are obtained, and a real-time monitoring video stream database of key parts of the offshore wind turbine is formed; The method comprises the steps of carrying out inspection shooting on the surfaces of the offshore wind turbine blades, the engine room and the tower barrel through an unmanned aerial vehicle in a short distance, carrying out inspection shooting on an offshore wind turbine anticorrosive coating below the sea surface and a submarine cable through an underwater robot, and obtaining an inspection image; Based on the electronic archive file and the paper log scanning file in the operation and maintenance management system, the periodic maintenance records, fault report and repair sheets, spare part replacement histories and text description filled by operation and maintenance personnel are collected, and maintenance history information reflecting the service life cycle of the unit is obtained.
  4. 4. The method for merging multi-source heterogeneous data and evaluating states of offshore wind turbines according to claim 3, wherein the step of preprocessing data of the multi-source heterogeneous database of offshore wind turbines comprises the steps of: Aiming at the running state data, identifying and correcting signal drift through multi-machine transverse comparison of the air blower, removing outlier data in the running state data of the air blower, complementing missing values in the running state data by adopting a Lagrange interpolation method, and gradually aligning on a time stamp to serve as a reference time axis for evaluating the state of the air blower; Aiming at structural dynamic response data, wavelet packet denoising and variation modal decomposition are adopted, low-frequency background noise caused by wave force and bandwidth random fluctuation of a mechanical transmission system are filtered, an energy operator is adopted to extract section characteristics of high-frequency acceleration waveform signals, a tower drum shaking degree signal and a blade shaking and swinging signal are reduced to be in dimension with sampling frequency consistent with running state data, and synchronous alignment of dynamic response and running state is realized by resampling and mapping to a reference time axis by taking a signal acquisition trigger time stamp as a standard; correcting the shielding deviation of the anemometer aiming at the external environment meteorological data, and carrying out smooth processing on the tide level and wave height to eliminate the instantaneous interference generated by sea wave crushing; The method comprises the steps of aiming at videos of a monitoring cabin, blades and a tower, running an image enhancement algorithm, removing picture shaking noise caused by fan vibration by using an optical flow method, triggering high-frequency frame extraction according to an alarm time point in an offshore fan SCADA system by adopting a key frame extraction technology, and performing low-frequency monitoring in other time to realize time alignment of visual characteristics and alarm information; Aiming at the inspection images acquired by the unmanned aerial vehicle and the underwater robot, removing invalid images with overexposure, sea surface reflection, and over-darkness of seabed brightness or focusing blurring, and repairing bluish green tone of the images acquired by the underwater robot, and aligning to a state reference time axis of the offshore wind turbine according to image acquisition time; Aiming at the full life cycle operation and maintenance record, converting a paper log scanning file into an electronic archive file by utilizing an optical character recognition technology, correcting non-standard abbreviations and ambiguous professional terms, extracting key elements, and performing time sequence correlation matching on maintenance time points of the text record and state characteristic evolution inflection points in original operation data of an offshore wind turbine SCADA system and a CMS monitoring system.
  5. 5. The method for merging multi-source heterogeneous data and evaluating state of offshore wind turbines according to claim 3, wherein said mapping the multi-source heterogeneous data preprocessed by the data to the unified embedded representation space by using the physical modeling and deep learning technique comprises: mapping the power generation power, voltage, current and blade rotation speed generating capacity of the offshore wind turbine to the working strength characteristic of the offshore wind turbine; mapping a generator bearing and a high-frequency acceleration waveform signal of a tower frame of the offshore wind turbine and a tower drum shaking degree signal and a blade waving and swinging signal into structural health indexes of the offshore wind turbine; semantically converting the external environment weather data into external excitation source characteristics; The pixel information in the monitoring video and the inspection image is converted into the rotation speed of the blade, the shaking amplitude of the tower, the width of the crack, the peeling area of the coating and the exposed length of the submarine cable, as a visual structuring index; Semantically converting the maintenance and replacement actions in the full life cycle operation and maintenance records into operation and maintenance knowledge weight vectors of the part wear degree recovery time points and maintenance experience; performing minimum-maximum standardization processing on the working strength characteristics of the offshore wind turbine, the attitude control characteristics of the offshore wind turbine, the structural health indexes of the offshore wind turbine and the external excitation source characteristics, and mapping physical characteristics of different dimensions to a unified high-dimensional linear vector space by adopting a fully-connected embedded layer to construct a physical perception characteristic tensor with time sequence continuity; Performing feature fusion on the visual structural index and the operation and maintenance knowledge weight vector to form visual and text semantic feature tensors capable of reflecting discrete events and morphological features; Mapping the physical perception feature tensor and the visual and text semantic feature tensor to a unified embedded representation space, and introducing sine/cosine position codes for all feature vectors in the unified embedded representation space.
  6. 6. The method for multi-source heterogeneous data fusion and state assessment of an offshore wind turbine according to any one of claims 1-5, wherein constructing a feature interaction operator based on a multi-head cross attention mechanism, performing feature decoupling, extracting a fusion feature vector reflecting a panoramic operation state of the offshore wind turbine, comprises: constructing a feature interaction operator based on a multi-head cross attention mechanism, mapping physical perception features, visual and text semantic features uniformly embedded in a representation space into query tensors, key tensors and value tensors through a linear transformation matrix respectively, and forming a multi-source heterogeneous feature tensor group with aligned dimensions; Performing scaling dot product operation by utilizing the multi-source heterogeneous feature tensor group, capturing implicit association strength between physical response and unstructured modes, and calculating a normalized cross-mode attention scoring matrix; Performing weighted mapping on the value tensor by using a cross-modal attention scoring matrix, introducing residual links, and constructing an enhanced physical feature vector fusing environmental context information; mapping the enhanced physical feature vector to a decoupling subspace through a linear projection matrix, and separating out independent feature dimensions, wherein the feature dimensions comprise structural service rigidity, external load environment and short-term operation fluctuation; and performing layer normalization processing on the structural service rigidity to generate a normalized panoramic service state vector.
  7. 7. The method for multi-source heterogeneous data fusion and state assessment of an offshore wind turbine according to any one of claims 1-5, wherein the introducing the panoramic service state vector as a global observation term into a variational bayesian calibration framework of an integrated physical information neural network realizes dynamic calibration of digital twin physical parameters, and generates an offshore wind turbine service state assessment report with uncertainty quantization support, comprising: Based on design parameters and historical service data of the offshore wind turbine, obtaining initial priori distribution of physical parameters to be calibrated, introducing hierarchical super parameters to represent commonality rules and evolution uncertainty of structural properties in different service stages and variable working conditions, and outputting a physical parameter space model to be optimized; Constructing a proxy model of the integrated physical information neural network, embedding a structural dynamic balance equation as a penalty term into a loss function, and outputting a likelihood function mapping operator from a physical parameter to multi-modal response; Introducing parameterized variational distribution to approach real posterior distribution, constructing a evidence lower bound loss function containing a physical constraint term by using a likelihood function mapping operator and a panoramic service state vector, and outputting a minimized divergence objective function to be solved; Sampling the variation distribution by utilizing a re-parameterization skill, optimizing an objective function by adopting a random gradient descent algorithm, updating variation parameters, and outputting a converged physical parameter posterior probability density function; Extracting a mean value from the posterior probability density function as an optimal calibration value of a physical parameter, extracting a variance as an uncertainty index of evaluation confidence of the quantized digital twin service state, and outputting a parameter calibration set with a confidence interval; Feeding back the parameter calibration values in the parameter calibration set to a finite element model for modal analysis, comparing actual measurement residual errors, executing dynamics monotonicity criteria, eliminating solutions against physical logic, and outputting final calibration parameters checked by physical criteria; And injecting the final calibration parameters into a digital twin model, calculating the basic stiffness degradation rate and the fatigue cumulative damage probability, and generating an offshore wind turbine service state evaluation report with uncertainty quantitative support.
  8. 8. An offshore wind turbine multisource heterogeneous data fusion and state assessment device, comprising: The acquisition module is used for acquiring the running state data, the structural power response data, the external environment meteorological data, the visual morphology of the key part and the full life cycle operation and maintenance record of the offshore wind turbine, and constructing an offshore wind turbine multi-source heterogeneous database; the preprocessing module is used for preprocessing data of the multi-source heterogeneous database of the offshore wind turbine, and the data preprocessing comprises data cleaning and space-time alignment; the mapping module is used for mapping the multi-source heterogeneous data subjected to data preprocessing to a unified embedded representation space by utilizing a physical modeling and deep learning technology; The fusion module is used for constructing a characteristic interaction operator based on a multi-head cross attention mechanism in the unified embedded representation space, performing characteristic decoupling, extracting a fusion characteristic vector reflecting the panoramic running state of the offshore wind turbine and taking the fusion characteristic vector as a panoramic service state vector; the evaluation module is used for introducing the panoramic service state vector into a variational Bayesian calibration framework of the integrated physical information neural network as a global observation item, realizing dynamic calibration of digital twin physical parameters and generating an offshore wind turbine service state evaluation report with uncertainty quantitative support.
  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 offshore wind turbine multisource heterogeneous data fusion and state assessment method of any one of claims 1-7.
  10. 10. A storage medium storing a program, wherein the program, when executed by a processor, implements the offshore wind turbine multisource heterogeneous data fusion and state assessment method of any one of claims 1 to 7.

Description

Multi-source heterogeneous data fusion and state evaluation method, device and medium for offshore wind turbine Technical Field The invention relates to a multi-source heterogeneous data fusion and state evaluation method, device and medium for an offshore wind turbine, and belongs to the field of offshore wind power big data processing and structural health monitoring. Background With the development of offshore wind power to deep sea and large scale, the fan is in service in extreme environments such as strong wind, billow, salt fog, complex seabed flushing and the like for a long time, and the structural safety of the fan is critical to the electric field benefit. At present, a marine wind power plant accumulates massive multi-element heterogeneous data, including visual information such as high-frequency waveforms generated by a CMS vibration sensor, low-frequency environment scalar generated by a weather station, monitoring video, inspection pictures and the like, and unstructured texts such as operation and maintenance logs, fault reports and the like. However, the existing monitoring system faces serious challenges that firstly, the effective semantic alignment and space-time correlation mechanism is lacking among various data, the existing monitoring system is in a serious 'information island' state, and deep coupling of multidimensional features is difficult to realize. Secondly, the traditional evaluation model is mostly aimed at single-mode data, expert experience and macroscopic morphology features contained in pictures and texts cannot be automatically extracted and fused, and the data utilization depth is insufficient. Finally, because accurate mapping can not be established between the real-time monitored heterogeneous information and the real bearing capacity of the structure, operation and maintenance decisions still depend on delayed regular inspection, the prejudgement based on multi-mode risk perception is lacking, and balance between power generation synergy and scientific operation and maintenance is difficult to realize in a complex environment. Disclosure of Invention In view of the above, the invention provides a multi-source heterogeneous data fusion and state evaluation method, device computer equipment and storage medium for an offshore wind turbine, which effectively solve the problems of data island, evaluation conservation, decision delay and the like and generate a service state evaluation report with physical interpretability and uncertainty support. The invention provides a multi-source heterogeneous data fusion and state assessment method for an offshore wind turbine. The invention provides a multi-source heterogeneous data fusion and state assessment device for an 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: an offshore wind turbine multisource heterogeneous data fusion and state assessment method, comprising the following steps: acquiring offshore wind turbine running state data, structural power response data, external environment meteorological data, key part visual morphology and full life cycle operation and maintenance records, and constructing an offshore wind turbine multi-source heterogeneous database; carrying out data preprocessing on the multi-source heterogeneous database of the offshore wind turbine, wherein the data preprocessing comprises data cleaning and space-time alignment; mapping the multi-source heterogeneous data subjected to data preprocessing to a unified embedded representation space by utilizing a physical modeling and deep learning technology; Constructing a characteristic interaction operator based on a multi-head cross attention mechanism in the unified embedded representation space, performing characteristic decoupling, and extracting a fusion characteristic vector reflecting the panoramic running state of the offshore wind turbine as a panoramic service state vector; and introducing the panoramic service state vector serving as a global observation item into a variational Bayesian calibration framework of the integrated physical information neural network, realizing dynamic calibration of the digital twin physical parameters, and generating an offshore wind turbine service state evaluation report with uncertainty quantitative support. Further, the acquiring operation state data, structural power response data, external environment weather data, visual morphology of key parts and full life cycle operation and maintenance records of the offshore wind turbine comprises: The system comprises an offshore wind turbine running state data, structural power response data, external environment weather data, critical part visual morphology and full life cycle operation and maintenance records, and is used for constructing an off