Research On Aeroengine Life Prediction Method Based On Long And Short Term Memory Neural Network And Multi Objective Optimization

The industry is growing fast,aviation equipment is evolving,engines are becoming more complex and smarter,and data needs to be collected on a larger scale.Engine failure will cause serious consequences,a large number of economic losses,as well as people’s safety and ecological environment damage.Therefore,RUL prediction of engine is of great significance to solve the above problems.However,in the traditional RUL prediction method,to judge whether the data is effective,its features need to be extracted manually,and the superparameters of the model need to be manually adjusted.Therefore,its prediction effect is low,but the rapid development of computer technology.The upper short-term memory neural network is applied to the residual life prediction of aeroengine.In this dissertation,a residual life prediction method based on the long and short term memory neural network and multi-objective optimization is proposed.In this dissertation,aiming at the problems of too many features and difficult to extract in RUL prediction,as well as the problems of super parameters requiring certain experience,this dissertation carried out a multi-objective optimization research on RUL prediction based on the neural network of long and short term memory.Features were extracted from data mainly through LSTM and its variant network,and the optimization algorithm was used to automatically optimize the super parameters.To improve the accuracy of prediction results.The research in this dissertation is as follows:(1)Research on residual life prediction based on BiLSTM.Analyze the original data,analyze its trend of change,determine whether the variable is valid,construct a time series of filtered features into the model,and predict the remaining life of the engine.At the same time,conduct sliding window comparison experiments to analyze the impact of window size on the predicted results under different operating conditions.The experiment shows that BiLSTM’s inaccurate prediction in the early stage and the size of the time window have an impact on the experiment.(2)Research on residual life based on GM-BiLSTM fusion model.A comprehensive prediction method based on CNN feature extraction,GM(1,n),and BiLSTM was proposed and implemented.The time series constructed from the output of CNN was used as the input of the model.The results of the GM model were smoother and more accurate in the early stage.The predicted values of the engine were obtained by weighting the gray prediction model and BiLSTM prediction results,and the weight coefficients were changed by the number of cycles.This method combines the advantages of the grey prediction model with BiLSTM to improve the prediction accuracy of the model.(3)Research on residual life prediction of GRUS model based on multi-objective optimization.Firstly,KPCA algorithm feature extraction is used to fully characterize the degradation process of the engine data set.The two-layer GRU structure is adopted.Compared with single-layer and multi-layer models,it has better regression ability and is not prone to overfitting and other problems.particle swarm optimization algorithm is used to optimize the model hyperparameter.By changing its weight update formula,the model can search more widely at the beginning and converge more quickly before reaching the stop condition,Reduce model training time.The experimental results indicate that the prediction method of this model is effective and reliable.