| The mechanical rotating equipment of wind turbine has a complex surrounding environment and the wind resources with strong randomness,large volatility and poor stability,which makes the fault warning of wind turbine become extremely important.Wind turbine fault warning can effectively change the post-maintenance into maintenance prevention,improve the utilization rate of parts and equipment maintenance efficiency at the same time,but also make the incidence of major failures greatly reduce,protect the safety of wind field manpower,material resources,financial resources.The traditional fault warning model of wind turbine is mostly in a single machine environment,which cannot meet the generation of massive data.The emergence of big data processing technology solves this problem.In this project,the Spark platform was established,SCADA data was used as experimental data for research,and the early warning analysis and parallel processing research of fault data of wind turbines based on Spark platform was realized.The parallel processing of wind turbine fault warning model is realized based on Spark platform.Introduces the kernel extreme learning machine algorithm,constructs the kernel extreme learning machine early warning algorithm based on the Spark framework parallelization implementation scheme,including the design model and warning test.The kernel function is introduced into extreme learning machine learning algorithm,method,the elasticity of the distributed data sets using Spark platform RDD to block storage of data set,each machine learning model parallel training,synthesis of the early warning model,eventually completed the parallelization process of wind turbine fault early warning,greatly improve the efficiency of the fault data batch.This paper introduces the whale optimization algorithm,and based on the warning model of parallel kernel extreme learning machine,combined with the improved whale optimization algorithm,realizes the fault warning of a LPP-MAWOA-KELM hybrid warning model under the Spark platform,including the establishment of the model and the warning test.Firstly,the guaranteed local projection algorithm is used to extract the characteristics of the unit state parameters,then the improved whale algorithm is used to optimize the kernel extreme learning machine,and the warning model is established.Finally,the parallelization operation of Spark platform is realized.Simulation results show that the proposed algorithm can warn potential faults at least 3 days in advance.Compared with genetic algorithm,particle swarm optimization algorithm and whale algorithm,the fault warning model of kernel extreme learning machine is obviously better than other selected models,and has higher accuracy and stability.Compared with the data processing in the single computer environment and the parallel cluster,the parallelization greatly improves the efficiency of the model.Ant-lion optimization algorithm is introduced,based on the warning model of parallel kernel extreme learning machine,combined with the improved Ant-lion optimization algorithm,a LPP-CALO-WKELM hybrid warning model fault warning is realized on the Spark platform,including the establishment of the model and the warning test.Firstly,feature vectors of SCADA data were extracted by using local preserving projection algorithm.Secondly,CALO algorithm uses Cauchy mutation operator to improve the Ant-lion optimization algorithm and enhance the global optimization ability.Finally,the parameters of the wavelet kernel extreme learning machine are optimized by CALO algorithm to improve the warning accuracy and convergence speed of the algorithm.In order to verify the effectiveness of the proposed diagnosis model,the actual operation data of the fan equipment in a wind farm in Northwest China were used for simulation experiments.The simulation results show that the LPP-CALO-WKELM diagnosis model can effectively identify different faults of the gearbox and meet the requirements of gearbox fault warning. |
PDF Full Text Request