Entropy-Based Choice of a Neural Network Drive Model

The design of a neural network requires, among other things, a proper choice of input variables, avoiding over fitting and an unnecessarily complex input vector. This may be achieved by trying to reduce the arbitrariness in the choice of the input layer. This paper discusses the relation between the memory range of a particular controlled dynamical system (induction drive) and the dimension of the neural network input vector. Mathematical techniques of process-reconstruction of the underlying process, using coding and block entropies to characterize the measure and memory range were applied. These modeling techniques provide a precise knowledge of the drive dynamics, a fundamental requirement in modern control approaches