NEURAL NETWORK CONTROL OF SWITCH MODE SYSTEMS: OFF LINE TRAINING BY AN "IDEAL CONTROLLER" DATA SET
Natalia Kodner, D. Adar, Sam Ben‐Yaakov · International Conference on Performance Engineering · 1995
A novel method is proposed for designing Neural Network Controllers (NNC) for switch mode systems. method applies a new concept: The Ideal Controller which is run 'off-line' to generate a record of 'perfect' control signals in response to input and output perturbations. record is then used as an 'off-line' training set for a Neural Network controller. advantages of the proposed method are three fold: (a) the training set is the best possible, (b) training is done 'off line' by simulation and (c) there is no need to derive or guess the control law. present study demonstrates by simulation the potential excellent performance of a NNC for DC-DC Switch-Mode converters when trained by the proposed methodology. proposed methodology can be readily expanded to other Switch-Mode systems such as inverters and power factor conditioners.