关键词: 分式聚合, 过程神经网络, 航空发动机, 气路参数偏差值, 时间序列预测

Abstract: To solve the problem that the mutation value in the aero-engine gas path parameter deviation time sequences was difficult to predict,a fractional nonlinear polymerization process neural network prediction model was proposed based on the theory of the rational function with better nonlinear approximation ability.The network structure had a process neuron dual layer in the hidden layer,which achieved incentives and disincentives of signal to neurons through fractional nonlinear space aggregation.According to the discrete characteristics of sampling points,the discrete Walsh transform data inner product was used to substitute integral operator,which simplified the calculation process and eliminated the accuracy loss in data fitting.The Levenberg-Marquardt (LM) method based on discrete Walsh transform was used to network training,the trained model was applied in the gas parameter deviation time series prediction,and the conclusion suggested that the model proposed had higher efficiency and sensitivity for the mutation value series forecasting with innovative and practical value in the approximation of the mutation time series.

Key words: fractional aggregation, process neural network, aeroengine, gas path parameter deviation, time series prediction