Learning Automata based Shiftable Domestic Load Scheduling in Smart Grid: Accuracy and Fairness

Abstract

In this thesis, investigation is carried out on scheduling of shiftable loads which involves partly selection of loads within the power budget of operator. Domestic shiftable loads are scheduled along multiple timeslots with the considerations of the accuracy of scheduling in terms of optimization of capacity and of the fairness between appliances in terms of frequency of usage in smart grids. Since the scheduled load can not be over the capacity, the global optimal point is a combination of loads which are most close or equal to but not over the capacity. This optimization problem is shown to be NP hard, and has been formulated as a potential game. To solve this problem in a distributed manner, Learning Automata (LA) based methods are proposed. Although the LA based methods do not favour any participants of scheduling which can serve as a fair selection in the long run, the fairness among the loads in finite time is still worth studying. To make the scheduling process fair in short time, virtual coin game is employed into the scheduling. Simulations have been performed by implementing two LA methods, namely BLA and LR−I, under different number of timeslots, with and without consideration of coin game to evaluate and compare the results. Simulation results show that the accuracy in terms of the closeness of the converged result to the global optimal point achieved by both LA based scheduling methods is high and the fairness of the system is increased by applying the virtual coin game.