Artificial Neural Network based controller design for SMPS

This paper presents a novel approach to design a controller for a half-bridge converter based Switched Mode Power Supply of 150W rating. A half-bridge converter is designed for obtaining a stiffly regulated DC output voltage of 5V irrespective of load and supply voltage variations, with an artificial neural network (ANN) based controller. The duty ratio of the controller is determined by this ANN depending on the input voltage and the output load conditions. Data collection for training the ANN was done using a PID controller-based approach in which the transfer function for the SMPS was derived by performing the large signal and small signal analyses and appropriate parameter values for the PID controller were calculated. For different load and input supply voltage conditions, the output voltage was regulated by making use of the PID controller and appropriate duty ratios were arrived at. These 150 data points were used for the training the proposed ANN with four hidden layers. The trained neural network was further utilized in arriving at the duty ratio value directly by sensing the input voltage and load current and it is found to yield stiffly regulated voltage irrespective of supply voltage and load variations where the error is within 8 %.