Application of neural networks to the optimal control of three-phase voltage controlled power inverters

Application of neural networks to the optimal control of three-phase voltage-controlled inverters is described. Precomputed switching angles for the elimination of low-order harmonics of the output voltage are used to train a software-emulated neural network. The trained network generates appropriate switching angles in response to the required value of the modulation index applied to its input. A regular network and a sparse network are analyzed and compared. The model inverter is controlled using the 80170NX ETANN chip coupled with the MC68332 Integrated Microcontroller System. Results of computer simulations and experimental investigation, confirming the feasibility of the proposed technique are presented