Iterative learning for optimal residential load scheduling in smart grid

In this paper, as a fundamental problem in smart grid, the residential load scheduling is studied in a comprehensive way. The main contributions lie in threefold. First, three indices, i.e., the power consumption expense, the robustness of schedule subject to uncertain electricity price and the satisfaction of customer, are taken into full consideration. We propose to optimize simultaneously the three indices via convex optimization. Second, iterative learning is utilized to setting parameters in the objective function, and keeps a proper tradeoff between the consumption expense and the satisfaction index. Third, in order to fully characterize the operation states of appliances, both binary and continuous variables are used, which results in a hybrid mixed-integer quadratic optimization problem. The relaxation technique is utilized to tackle the hybrid optimization problem. Theoretical analysis of the performance gap based on the proposed approach is provided as well. The performance of the proposed approach is illustrated by simulations. The parameter settings reflect actual preference and consumption manner of the consumer. In addition, both peak-to-average ratio of power load and variation of power load are reduced.