Research On Condition Monitoring And Analysis Of Operating Characteristic Of Large Wind Turbines

Development of wind power can effectively relieve global energy crisis, reduce environmental pollution, ensure energy security of society development. Real-time monitoring of running state of wind turbines, identifying abnormal state, can reduce costs of operation, ensure safe and efficient operation of unit. In this paper, model of power characteristic and model of temperature trend forecasting of the gearbox bearing are established based on thorough analysis data of SCADA system, researching on analysis of running state.Five models of power characteristics are established with the method of least squares support vector machines(LSSVM), based on the analysis of the power of volatility with the existing research results. Mutation of genetic algorithm is introduced into simulated annealing particle swarm algorithm(SAPSO), improving its ability of global optimization, and parameters of models for power characteristics are optimized with it. Verify the rationality of the models by comparing their fitting accuracy; at the same time, inspect the ability of generalization and feasibility in condition monitoring of models with actual operation data; providing a new way of thinking for power characteristics.Operation process of gear box is complicated, and cost of maintenance and repair is large, but model of power characteristic can't analyze its mild abnormal state. For the situation that vibration signal acquisition is difficult and quality of signal is poor, model of temperature trend forecasting for gearbox bearing is established with the method of non-linear state estimation(NSET), based on SCADA system data. Grey relational analysis(GRA) is used to examine rationality of variable selection of model; the method of similarity analysis is used to contract process memory matrix, eliminating redundant data and enhancing timeliness of model; the method of sliding window statistical analysis is applied to analyze predict residual of model, eliminating influence of random disturbance of running process. The simulation results show that; running time of NSET prediction model based on similarity analysis is reduced greatly; improving the efficiency of model; feasibility of model in condition monitoring is verified.