Turbojet modeling in wind milling based on neural network incorporating priori knowledge

Abstract

Neural network is an effective method for turbojet modeling in wind milling, but its deficiency in generalization ability has restricted its application in engineering. Nonlinear PCA (principal component analysis), although is very effective in decreasing the dimensions of input variable and subsequently improving neural network's generalization ability, it has difficulty in finding an appropriate nonlinear transform in engineering application. A method, which can be applied in turbojet modeling in wind milling based on neural network, is proposed in this paper. By incorporating priori knowledge of dynamic and static state of rotor, similar parameters and the relationship between residual power and acceleration, this method not only decreases the neural network's dimensions reasonably and improves its generation ability greatly, but overcomes difficulties of nonlinear PCA. The simulation results prove the method to be simple and effective.