A real-time stepwise supervised learning algorithm for time-series prediction and system identification

This paper presents a new neural network architecture and a real-time stepwise supervised learning algorithm that rapidly updates the weights of the network while importing new observations. The most significant advantage of the stepwise approach is that the weights of the network can be easily updated so that re-training is not necessary when new data or observations are made available later after the neural network is trained. This feature makes the stepwise updating algorithm perfect for time-series prediction and system identification. The network has also been tested on several data sets and the experimental results are compared with some conventional networks in which more complex architectures and more costly training are needed.