Research On Multi-fault Diagnosis Method Of BBPSO-PF Based On CUDA In Wind Turbine Pitch System

As a kind of clean renewable energy,wind energy has been paid more and more attention.However,most wind farms are concentrated in the harsh environment,which leads to frequent failures of wind turbines,and brings huge economic loss es and safety risks.Therefore,real-time fault diagnosis of wind turbines is of great significance to ensure the safe and stable operation of wind turbines.Particle filter algorithm plays an important role in fault diagnosis of nonlinear and non-Gaussian systems.However,after several iterations of the traditional particle filter algorithm,particle depletion will occur and it is easy to fall into local optimization,which reduces the accuracy of the algorithm,the general solution is to increase the num ber of particles,but this method will reduce the real-time algorithm.To solve this problem,this paper proposes a GPU-based parallel backbone particle swarm optimization particle filter algorithm,which can improve the real-time performance of the algorithm on the premise of ensuring the accuracy of the algorithm,and is applied to the fault diagnosis of the variable pitch system of wind turbines to improve the real-time performance of fault diagnosis.The work contents of this paper are as follows:1.In the resampling process of particle filter algorithm,due to the need for interaction between particles to complete the position update,the real-time performance of the algorithm is poor.In order to improve the real-time performance of the algorithm,the backbone particle swarm optimization(BBPSO)has the characteristics of parallel structure to optimize the resampling link of the particle filter algorithm,reduce the time complexity of the algorithm,and improve the parallelism and real-time performance of the algorithm.2.The particle filter algorithm based on backbone particle swarm optimization is implemented in parallel on GPU.Firstly,CUDA’s multi-threaded architecture is used to implement the algorithm in parallel.The threads correspond to the p article swarm one by one.Each thread is responsible for the data processing of a particle swarm,so that the particle swarm can be processed in parallel.Secondly,according to the principle of GPU’s alignment and merging memory access,an efficient parti cle swarm data structure is designed to make multi-threaded data access complete data access with the least amount of memory access transactions,and improve the real-time performance of the algorithm from the aspects of multi-threaded and memory access.3.The particle filter algorithm optimized by the backbone particle swarm optimization is applied to the multi-fault diagnosis of the variable pitch system of wind turbine to improve the real-time performance of multi-fault diagnosis.Firstly,a parallel algorithm program is designed using a single GPU device to perform multi-models fault diagnosis in parallel at the data level.Each particle filter is responsible for the diagnosis of one fault model,and multiple particle filters are queued in the GPU to diagnose all fault models.Secondly,in the multi-GPU environment,CUDA streaming technology is used to design parallel algorithm program based on multi-GPU.Each GPU is responsible for the diagnosis of one fault model,and the diagnosis of multiple fault models is completed in parallel at the task level.Finally,the method of combining Open MP and CUDA architecture is used to parallelize the threads on the CPU side by using Open MP.Each CPU thread is responsible for scheduling a GPU,removing dependencies between GPUs,and solving pseudo-parallel problems between multiple GPUs,thus,the real-time performance of multi-fault diagnosis in the variable pitch system of wind turbine is further improved.