A hybrid approach involving data driven forecasting and super twisting control action for low-carbon microgrids

Abstract

This research paper introduces a two-level dense residual neural network-based optimization framework designed to enhance the efficiency of energy management systems in microgrids. The framework addresses the shortcomings of conventional numerical optimization methods for solving the economic dispatch problem, which often prioritize accuracy over real-time performance and fail to maximize power generation from renewable energy sources. The proposed framework’s upper-level control not only solves the economic dispatch problem but also optimizes power output from renewable sources. At the local level, a super-twisting sliding mode control is employed to accurately track EMS-generated references and ensure precise DC bus regulation. The stability of the framework is validated using Lyapunov stability criteria. The framework is tested on a 600 V electric-hydrogen based islanded microgrid system with a 550 kW capacity. Real-time simulations are validated through hardware-in-the-loop experiments using the OPAL-RT OP5707XG.