Identification and control of induction motors using artificial neural networks with random weight change training

This paper highlights some of the latest developments in ongoing work on the practical implementation of a continually online trained neural network induction motor controller. In particular, it considers the potential use of a proposed analogue VLSI sigmoidal feedforward neural network, with on-chip random weight change training, for rapid control of induction motor stator currents. The random weight change training algorithm is described and its convergence mechanism explained by comparison with that of backpropagation training. Simulation results also demonstrate the potential induction motor current control performance achievable with random weight change in comparison with that of backpropagation training.