Sliding mode control of a DC/DC PWM converter with PFC implemented by neural networks

An experimental neural controller implementing a variable structure control (VSC) algorithm is proposed for a power factor preregulator. VSC control laws yield fast response and a robust behavior against large parameters variations. A multilayer perceptron learns through backpropagation to approximate the desired adaptive control functions. The main advantage of the neural network implementation in comparison to the numerical implementation is that decreases complexity and cost of the controller, and increases the switching frequency. A simple analog electronic realization of this neural network using discrete operational amplifiers is proposed. This implementation possesses all good properties of sliding mode while avoiding the unnecessary discontinuities of the control input signals and thus eliminating chattering. Experimental results are summarized confirming the validity of the neural network approach.