Abstract: In order to improve the fitting precision of engine baseline, an Elman neural network model optimized by chaotic particle swarm is proposed. The position formula of PSO is improved by chaos algorithm to solve the local optimization problem. A nonlinear decreasing function is proposed to improve the PSO particle velocity formula to solve the problem of low convergence accuracy. The model is applied in baseline fitting and the fitting errors are compared with those of traditional BP network, Elman network and SVM. Results show that the fitting accuracy of chaotic PSO_Elman model is higher than that of other traditional models with the same training and testing data and the same training times. When the training samples get fewer, the fitting accuracy is still higher than the traditional models, proving the stronger learning ability of the model.

Key words: aero-engine, baseline mining, chaos, particle swarm optimization, Elman neural network