Imitation learning‐based online optimal scheduling for microgrids: An approach integrating input clustering and output classification

Abstract

AbstractStrong uncertainties of distributed renewable generations, loads and real‐time market prices impose higher requirements on the online scheduling for microgrids. Traditional model‐driven methods exist several limitations due to low solving efficiency, difficulty in handling uncertainties, reliance on accurate prediction information, and inability to leverage accumulated historical data. This paper proposes a data‐driven improved imitation learning based approach for online microgrids optimization. First, a mixed integer linear programming model is established to derive offline optimal decisions within the given scenarios, which serve as expert demonstrations to help construct a sample database for imitation learning. Next, a direct imitation learning model based on eXtreme Gradient Boosting (XGBoost) is established to learn the mapping relationship between the system state and the scheduling decision, and the model training is refined by input clustering and output classification. At the input end, the operation scenarios are clustered to form multiple sub‐feature sets to achieve targeted training for different scenarios. At the output end, a binary variable is added to the label set to realize high‐precision decisions of the action of the energy storage system. Numerical case studies demonstrate the performance advantages of the proposed method.