Research On Aero-engine Exhaust Gas Temperature Prediction

Aero-engine exhaust gas temperature(EGT)is one of the main performance parameters of aero-engines.The state changes of EGT will directly affect the safety and reliability of the aircraft.Therefore,the accurate prediction of EGT not only helps to judge the performance state of the engine,but also provides sufficient decision-making time for troubleshooting and maintenance plan formulation.However,due to the harsh working environment of aero-engines and complicated nonlinear systems,the acquired engine performance parameters are characterized by random fluctuation,instability and nonlinearity,which make it difficult to establish an accurate physical prediction model.Considering the relation of aero-engine EGT with other performance parameters,so the multi-performance parameter based on artificial intelligence method is established as the input and the engine EGT is the output prediction model,and the aero-engine example is used to verify the reliability of the prediction model.The article will conduct in-depth research from the following aspects:(1)Firstly,the prediction methods of the aero-engines EGT at home and abroad were introduced and the advantages and disadvantages of PLS(partial least squares),MLR(multivariate linear regression),SVM(support vector machine)prediction,ANN and combined forecasting methods were analyzed comprehensively.In order to further study the aero-engine EGT prediction laid the foundation.(2)In order to improve the prediction accuracy of the aero-engine EGT.Firstly,the change characteristics of aero-engine-related data were analyzed.Statistical analysis was used to identify the anomaly points in the data and the exponential smoothing method was used to smooth the initial data;Secondly,the data is normalized to eliminate the dimension of the input data;Finally,the mean impact value(MIV)was used to analyze the correlation of EGT and related parameters.(3)Based on the relationship between EGT and related parameters,the single prediction models such as PLS,MLR,Elman neural network and SVM with multi-parameter as input and single parameter as output was established.In order to improve prediction accuracy,the initial parameters of SVM and Elman network are optimized by ADQPSO(adaptive disturbance quantum-behaved particle swarm optimization),and the examples were used to verify the methods.The analysis results show that,compared with traditional SVM and Elman neural networks,the prediction errors of the optimized Elman neural network and the SVM prediction model are relatively smaller.(4)By analyzing the prediction results of the single model,several single prediction models with the better prediction results were selected to establish a combined prediction model based on ADQPSO.Under the condition of minimum error,ADQPSO algorithm was used to assign reasonable weights to each single model and the noise immunity,adaptability and training time of the combination model were evaluated.