Abstract:

A BP(back propagation)neural network is used as the controller of a two spool turbo jet engine. Because the neural network adjusts its weights by using gradient algorithm, the partial derivatives of the outputs to the inputs of the plant are needed. However, the partial derivatives of a complicated thermal system are very difficult to be presolved precisely , and thus another neural network is added to the control system and acts as a mapper of the nonlinear plant composed of the engine and its fuel auxiliary. The simulation results show that the control structure does not need any model parameters of the plant and is of stronger robustness.

Key words: BP neural network, robustness, nonlinear