The backpropagation neural network being capable of real-time identification

A new structure for backpropagation networks is proposed, which enables the online identification of dynamic systems. Currently, the activation functions used for neural networks are sigmoid-like. Their ability to model any nonlinear dynamic system has been proved. However, the condition of uniformly distributed online training examples is usually not satisfied for identification of dynamic systems. Therefore, the backpropagation network are essentially suitable for off-line training. To solve the problem of real-time systems identification, the sigmoid functions are multiplied with some functions, that can specify the behaviour of each neuron in a limited range of training data. In this way the strong couplings among the network parameters are largely released and the training data for backpropagation networks need not to be chosen randomly from the identified input range, i.e. the network exhibits a real-time memory ability and the online training of backpropagation networks becomes possible.