LAUSR

More accurate prediction does not necessarily lead to better decision. Distinguished from existing accuracy based “predict then optimize” methods, this work offers optimal distributed energy management strategies in modern power system operation utilizing a decision regret oriented “smart predict and optimize” approach as the first trial. A two-stage day-ahead scheduling and real-time redispatching problem in a local energy community (LEC) is specifically modeled, where peer-to-peer (P2P) energy sharing is considered. The asymmetric and nonmonotonic relationship between prediction error and decision regret in this problem is revealed. A centralized energy and auxiliary service joint market and a decentralized P2P energy market are introduced to ensure efficient local energy consumption and tackle with prediction errors, which are almost inevitable. Two computational efficient prediction models, linear regression, and extreme learning machine, are employed as market predictors. Considering the potential risk of communication failures among smart agents in practical applications, and to avoid being trapped into local optimum points, a novel censored communication robust distributed algorithm is proposed. Simulation for a highly distributed energy sources penetrated LEC statistically prove the economic benefits of the proposed strategies.