Machine-Learning Based Optimal Energy Management for Building Level Integrated Energy Systems

This paper proposes a data-driven realtime economic regulation method for household energy. This method utilizes historical operating data to construct a training dataset through a model-driven approach, thereby training ANNs to construct a data-driven scheduling decision model. Electrical and ESS constraints are applied to ensure that equipment operation does not exceed constraint limits. The advantage of this method is that it takes realtime operating data as input, without the need to predict uncontrollable load power, realtime electricity price, and PV power generation data. Compared to the model-driven MPC scheduling method, it avoids the impact of prediction errors on the scheduling results. The simulation results show that the method proposed in this paper can optimize the operation of electrical appliances and ESS based on changes in realtime electricity prices, to reduce operating costs. The proposed solution is evaluated through a range of simulation experiments and the numerical results have confirmed that the proposed solution can performance well in building-level energy systems.