A Sliding-Window Modeling Approach for Neural Network
Xiao Laisheng · International Journal of Control and Automation · 2014
A sliding-window modeling approach of neural network (SWMANN) was presented.The basic idea is that training data for neural network should be reconstructed by means of a slide-window way to build input and target samples.It means that the output is determined not only by the current input, but also by the past input, which could better follow the dynamic change and development of real systems.In this paper, SWMANN was introduced.A detailed theoretical derivation was described for its implementation by taking wavelet neural network as a modeling tool.Moreover, SWMANN was compared with classical modeling approach of neural network (CMANN).Theoretical results indicated that CMANN is a special case of SWMANN when sliding-window parameters are selected as 1.Therefore, compared with CMANN, SWMANN presented in this paper is a more general form.