Research on Fault Early Warning of Wind Turbine Based on IPSO-DBN

Aiming at the problem of wind turbine generator fault early warning, a warning method based on nonlinear decreasing inertia weight and exponential change learning factor particle swarm optimization is proposed to optimize deep belief network (DBN). With data farm supervisory control acquisition (SCADA) as input, weights biases are pre-trained layer by layer. Then BP neural used fine-tune parameters whole network. The improved algorithm (IPSO) determine number neurons in hidden model, pre-training rate, reverse fine-tuning times training other parameters, DBN predictive regression model established. experimental results show that has better performance accuracy, time fitting ability than PSO-DBN model.