Data-driven robust optimization scheduling for microgrid day-ahead to intra-day operations based on renewable energy interval prediction

The uncertainty in renewable energy forecasting significantly impacts microgrid scheduling, and traditional scheduling schemes are often overly conservative and limited by a single time scale, resulting in unreasonable strategies that fail to balance reliability and cost-effectiveness. Based on this, a comprehensive two-stage day-ahead and intra-day microgrid scheduling framework is proposed integrating forecasting, regulation, and decision-making. Based on historical power data of renewable energy, a multi-kernel covariance function is applied to improve Gaussian process regression (GPR) for adaptively generating renewable energy power prediction intervals at various confidence levels and verifies the construction of robust optimization uncertainty scenario sets. In the day-ahead scheduling phase, a two-stage adaptive robust optimization model based on interval probability uncertainty sets is established to ensure minimal scheduling costs under the worst-case scenario. Meanwhile, a modified deep Q network (MDQN) algorithm based on a k-priority sampling strategy is proposed to transform the two-stage iterative process into a discrete Markov decision process for solution. In the intra-day scheduling phase, the day-ahead scheduling scheme is followed, and the intra-day scheduling scheme is optimized through multi-time-scale rolling optimization to reduce the impact of renewable energy power fluctuations. Case studies validate that the proposed scheduling method ensures robustness against uncertainties in renewable energy output while also maintaining the economic efficiency of system operations. • Propose an improved GPR adaptively generates renewable energy prediction intervals. • Propose a two-stage robust optimization scheduling method. • Propose a MDQN algorithm to solve the robust model.