Optimization of Electricity Consumption Forecasting Models via Hyper-Heuristic Algorithm

Time-series models are widely employed in power forecasting. However, the initial parameter setting of these models poses challenges for deployment. To address this issue, we propose a novel hyper-heuristic algorithm to optimize the parameters of the time-series model. The proposed hyper-heuristic algorithm integrates the genetic algorithm (GA) as the high-level component while the particle swarm optimization (PSO) and bat algorithm (BA) as the low-level heuristics (LLHs). We name our proposal GA-PB. Through iteratively searching for optimal parameters by LLHs, GA ensures the adaptive and intelligent adjustment for the hyper-parameters in PSO and BA, thereby enhancing the robustness and scalability of the proposed hyper-heuristic algorithm. To validate the efficiency of our proposal, we use four distinct time-series models with electricity consumption data from Japan and the USA. In the comparison experiments with differential evolution (DE), PSO, and BA, our proposal has a competitive performance in various optimization instances, demonstrating the effectiveness and efficiency in parameter identification of electricity consumption forecsting models.