Research On Fault Diagnosis Method Based On Particle Filter In Wind Turbine Pitch System

With the development of wind power technology and mechanical equipment manufacturing technology,the reliability of wind power equipment is gradually enhanced,However due to various uncertain factors such as wind speed,rain and snow,lightning,etc.,making the development of wind power technology is not yet being able to cope with all environmental changes,the probability of wind turbine failure is much higher than other electromechanical equipment.Generally,the main systems for wind turbine failure are blades,transmission systems,and pitch systems.The pitch system is a typical nonlinear,strongly coupled system that is primarily responsible for controlling the pitch angle to accommodate wind speed changes.Due to the need for frequent starts and stop,resulting in frequent failure of the pitch system,which makes the maintenance cost of the wind turbine high,it is important to conduct fault diagnosis research on the wind turbine pitch system.As a method with unique advantages in dealing with nonlinear systems,particle filter is not affected by model linearization and noise obeying Gaussian distribution,and has been widely studied and applied in the field of fault diagnosis.However,the particle filter algorithm has the problems of particle depletion and particle degradation,which leads to low accuracy of fault state estimation.Therefore,the improvement of particle filter algorithm is also a hot research topic.In this paper,the shortcomings in particle filter and the application of improved particle filter algorithm in the fault diagnosis of wind turbine pitch system are deeply studied.The main research contents are as follows:(1)For the problem that particle weight degradation and particle depletion of particle filter algorithm lead to low estimation accuracy of pitch system state,the universal gravitation algorithm is used to optimize particle filter.Firstly,using the strong optimization feature of the universal gravitational algorithm,the gravitational algorithm is optimized to replace the re-sampling process of particle filtering.Each particle is regarded as a mass point,and the particles are guided by gravitation to guide the particle to the mass-sized particle.Secondly,a dynamic inertia weight is introduced to optimize the particle velocity update mechanism to improve the particle degradation problem and improve the accuracy of state estimation.Introducing a perceptual matrix,improve the loss of particle diversity.(2)For the actual working environment of the pitch system,there are random noise,coupling interference,etc.,which leads to the high false alarm rate and time complexity of the traditional fault detection method.The idea of the confidence interval in statistics is introduced to an adaptive threshold design changing with the residual.The residual at time k is greater than the threshold,indicating that the system has failed,and the trained fixed value is introduced as the bandwidth coefficient to optimize the mean and variance of the adaptive threshold,so that the calculation of the mean and variance depends only on the previous one.Thus,the real-time performance is improved while ensuring the accuracy of fault detection.(3)The traditional fault isolation method is affected by signal interference and noise,and can not fully utilize the fault characteristics,resulting inaccurate fault separation.The fault isolation method combining J divergence and multi-model is used to fully extract the non-Gaussian fault characteristics from the particle filter estimation,and accurately isolae faults in the case of inaccurate system modeling.The effectiveness of the proposed algorithm is verified by fault diagnosis of the wind turbine pitch system.