Key points are not available for this paper at this time.
In the future, residential energy users can seize the full potential of demand response schemes by using an automated home energy management system (HEMS) to schedule their distributed energy resources. In order to generate high quality schedules, a HEMS needs to consider the stochastic nature of the PV generation and energy consumption as well as its inter-daily variations over several days. However, extending the decision horizon of proposed optimisation techniques is computationally difficult and moreover, these approaches are only computationally feasible with a limited number of storage devices and a low-resolution decision horizon. Given these existing shortcomings, this paper presents an approximate dynamic programming (ADP) approach with temporal difference learning for implementing a computationally efficient HEMS. In ADP, we obtain policies from value function approximations by stepping forward in time, compared to the value functions obtained by backward induction in DP. We use empirical data collected during the Smart Grid Smart City project in NSW, Australia, to estimate the parameters of a Markov chain model of PV output and electrical demand, which are then used in all simulations. To evaluate the quality of the solutions generated by ADP, we compare the ADP method to stochastic mixed-integer linear programming (MILP) and dynamic programming (DP). Our results show that ADP computes a solution much quicker than both DP and stochastic MILP, while providing better quality solutions than stochastic MILP and only a slight reduction in quality compared to the DP solution. Moreover, unlike the computationally-intensive DP, the ADP approach is able to consider a decision horizon beyond one day while also considering multiple storage devices, which results in a HEMS that can capture additional financial benefits
Keerthisinghe et al. (Wed,) studied this question.