Parallel Feedforward Process Neural Network with Time-Varying Input and Output Functions

Abstract

In reality, the inputs and outputs of many complicated systems are time-varied functions. However, conventional artificial neural networks are not suitable to solving such problems. In order to overcome this limitation, parallel feedforward process neural network (PFPNN) with time-varied input and output functions is proposed. A corresponding learning algorithm is developed. To simplify the learning algorithm, appropriate orthogonal basis functions are selected to expand the input functions, weight functions and output functions. The efficiency of PFPNN and the learning algorithm is proved by the exhaust gas temperature prediction in aircraft engine condition monitoring. The simulation results also indicate that not only the convergence speed is much faster than multilayer feedforward process neural network (MFPNN), but also the accuracy of PFPNN is higher.