Learning-aided distributionally robust optimization of DC distribution network with buildings to the grid

The large-scale integration of distributed resources in flexible direct current (DC) distribution networks with buildings to the grid presents challenges. These networks can be combined with distributed photovoltaic (PV), energy storage systems (ESS), and DC distribution systems within a single building and realize a flexible energy operation. The distributionally robust optimization (DRO) model, economically efficient and robust, stands out for managing the uncertainty of distributed resources. However, the conventional DRO physical model of DC distribution systems proves inefficient, struggling to meet the demands of stable and economically viable operations of the current DC distribution system. Therefore, we propose a DRO scheduling method for DC distribution systems with buildings to the grid assisted by deep learning. This novel approach replaces the iterative solution process of conventional scenario-based DRO physical models with a deep learning method. By directly predicting the worst probability distribution of typical scenarios, the original DRO model is transformed into a single-level stochastic programming model, significantly enhancing the model's solution efficiency. The effectiveness of our approach is validated through simulations conducted on a 33-node DC distribution network with buildings to the grid, demonstrating improved solving efficiency and calculation accuracy compared with conventional methods.