One-step optimal space vector PWM current regulation using a neural network

This paper presents the solution for an optimal deadbeat current regulator for use in a three-phase PWM inverter connected to an AC motor. A general solution to the one-step, least-squares current error problem with modulation index constraints is derived using a space vector formulation in conjunction with the modulation index cube. The control produces a deadbeat response when it is possible and the best approximation when it is not. The contours of constant modulation index with respect to the desired current adjustment naturally decompose into a simple, four-neuron-per-phase neural network. It is also shown that the use of this neural network will improve the performance of a standard PI regulator. The neural network is implemented using simple analog circuits