Research On Remaining Useful Life Prediction Method Based On LSTM

To meet the actual industrial production needs,the complexity of modern industrial system is constantly increasing.More and more attention has been paid to the security and reliability of the system.Prognostic and health management system is an important method to ensure the stable and efficient operation of the system.Remaining useful life(RUL)prediction is the core technology of prognostic and health management system to realize its function.RUL of the system is the length from the current time to the time when the system is no longer performing as expected.RUL prediction methods can be categorized into model-based methods and datedriven methods.With the continuous development of sensor technology and artificial intelligence technology,data-driven prediction methods have been widely used in complex industrial systems and deep learning algorithm has become an important research content in the data-driven RUL prediction methods.RUL prediction can provide scientific and reasonable suggestions for system maintenance.This paper mainly studies the data-driven RUL prediction method.The target system is turbofan engine,and the data set used is C-MAPSS data set.Firstly,the data samples in the C-MAPSS dataset are analyzed.Variable selection,data normalization and piecewise linearization are carried out on the data samples.Convolutional neural network(CNN)is used to extract system performance degradation information hidden in data samples.Then,long short-term memory network(LSTM)for time series prediction is introduced,and the basic prediction model based on CNN-LSTM is established.Gated recurrent unit(GRU)is adopted to build the RUL prediction model,and the second basic prediction model based on CNN-GRU is established.Finally,ensemble learning method is applied to improve the prediction performance of RUL basic prediction models,and RUL hybrid prediction model is established.Batch normalization and dropout techniques are used to optimize the training process of the prediction model and shorten the training time.The prediction performance of the proposed RUL prediction model is verified and evaluated on C-MAPSS data set.Firstly,the prediction performance of the RUL hybrid prediction model and the two RUL basic prediction models are compared.The comparative results show that the hybrid prediction model has better comprehensive prediction performance,which proves that ensemble learning method can effectively improve the prediction performance of the basic prediction model.Finally,the prediction performance of the proposed RUL prediction model and existing RUL prediction methods are compared and the results show that the proposed RUL prediction model has the best comprehensive prediction performance,which verifies the effectiveness and advancement of the proposed RUL prediction model.