A New Algorithm for Raising the Training Speed of Process Neural Network

Process neural network (PNN) is a new type of artificial neural network studied in recent year. PNN is an extent of traditional neural network, in which the inputs and outputs may be time-variation. Some modified algorithms for raising the training speed of PNN were investigated emphatically. These algorithms were based on function orthogonal basis expansion which exist low-speed convergence in network training. After introducing the improved algorithm which increased function momentum adjustment item and training rate automatic adjustment method for network weight function, and the normalizing rule on original algorithm, the sensitivity of error side particular was reduced, and the convergence of training was accelerated. The fact shows that the stability and training precision are improved with the learning rate automatic adjustment method, and it can also restrain the network falls into local least by introducing momentum adjustment item. A good result in application is represented by simulation