State forecasting for rotary machine based on neural network and genetic algorithm

A state forecasting is a key technology to achieve the advanced predictive maintenance. A Prediction based on neural network is a new approach to realize the state predicting. The present neural networks predicting models are comparatively poor in adaptability to environment and in predicting accuracy, therefore, a new rotary machine online state forecasting method based on the genetic algorithm (GA) and neural network (NN) was presented. GA was used for dynamical optimizing the structure parameters of BP network to obtain the optimal network structure. A training algorithm combining GA with BP was adopted to avoid the local minimum and to heighten the learning precision. The state predicting results for hydraulic pump indicate that the predicting model purposed may dynamically optimize the structure parameters in accordance with different conditions, and gained satisfactory results.