BP network based aeroengine identification using modified particle swarm optimization

BP neural network is used to build an identification model for an aeroengine with 4 inputs and 4 outputs under different conditions of flight height and mach number. A modified particle swarm algorithm is presented to optimize the weights of BP network. The proposed algorithm can overcome the disadvantage of that the existing particle swarm optimization gets easily into the local minimum by introducing the average of the individual best position, the acceleration of the swarm coefficients of variation of adaptive and adaptive way to adjust the position of a large amount of deviation and other methods. Simulation results show that the proposed identification model has shorter train time, little prediction error and higher identification precision compared with the other BP network models based on GA or the elementary PSO.