Deep learning-based barrier-function super-twisting sliding mode control for integrating renewables in smart grid

Abstract

This study presents a two-step controller design for integrating wind and photovoltaic energy systems into the smart grid. In the first stage, an enhanced deep neural network (DNN), optimised by particle swarm optimisation and a genetic algorithm, is employed to generate maximum power point (MPP) targets for both wind and solar systems. The DNN incorporates advanced features like hyperparameter tuning, softmax attention, dropout, and early stopping to improve prediction accuracy and prevent overfitting. In the second stage, a barrier-function-based adaptive super-twisting sliding mode controller is developed to track the MPP and maintain stable DC bus voltage. This controller effectively reduces chattering and operates without requiring detailed knowledge of disturbance limits. The proposed design aims to maximise power extraction, ensure DC bus voltage stability, and enable smooth power delivery to the grid. MATLAB simulations and real-time hardware testing validate the controller's performance, demonstrating significant improvements over traditional methods.