A hybrid non-dominated sorting genetic algorithm for a multi-objective demand-side management problem in a smart building

Abstract

One of the most significant challenges facing optimization models for the demand-side management (DSM) is obtaining feasible solutions in a shorter time. In this paper, the DSM is formulated in a smart building as a linear constrained multi-objective optimization model to schedule both electrical and thermal loads over one day. Two objectives are considered, energy cost and discomfort caused by allowing flexibility of loads within an acceptable comfort range. To solve this problem, an integrative matheuristic is proposed by combining a multi-objective evolutionary algorithm as a master level with an exact solver as a slave level. To cope with the non-triviality of feasible solutions representation and NP-hardness of our optimization model, in this approach discrete decision variables are encoded as partial chromosomes and the continuous decision variables are determined optimally by an exact solver. This matheuristic is relevant for dealing with the constraints of our optimization model. To validate the performance of our approach, a number of simulations are performed and compared with the goal programming under various scenarios of cold and hot weather conditions. It turns out that our approach outperforms the goal programming with respect to some comparison metrics including the hypervolume difference, epsilon indicator, number of the Pareto solutions found, and computational time metrics.