Abstract
This research utilizes time series models to forecast electricity generation from renewable energy sources and electricity consumption. The configuration of optimal parameters for these models typically requires optimization algorithms, but conventional algorithms may struggle with fixed search patterns and limited robustness. To address this, we propose an auto-evolution hyper-heuristic algorithm named AE-GAPB. AE-GAPB integrates a genetic algorithm (GA) at the high-level component and employs particle swarm optimization (PSO) and the bat algorithm (BA) at the low-level component. The GA continuously finds the best hyperparameters for PSO and BA based on prediction accuracy, which significantly accelerates the optimization and improves the accuracy. Additionally, the crossover and mutation rates of GA evolve over iteration time and fitness value space, further enhancing its adaptability. We validated AE-GAPB on six time series forecasting models and compared it with five well-known optimization algorithms as well as GAPB without auto-evolution at the high-level component. As a result, AE-GAPB achieved excellent results on renewable energy generation and electricity consumption datasets from the Hokkaido, Kyushu, and Tohoku regions of Japan.