System-Type Neural Network Architectures for Power Systems

Abstract

Neural networks have been applied in various new ways to the manifold problems in power systems. The great majority of neural network designs attempt to model a dynamic mapping with one neural network. Recently, attempts have been made at using system-type neural networks for distributed parameter systems, where the system dynamics is distributed over a spatial-temporal domain. In this paper, system-type neural networks is illustrated, which are designed using semigroup theory. The objective will be either to achieve extrapolation of functional patterns along one axis, or to achieve a forecasting of functional patterns in multiple axes.