Enhanced Electricity Demand Forecasting through Metaheuristic Optimization of Model Parameters

Accurate prediction of the fluctuating nature of electricity demand remains a persistent challenge, primarily due to the complexity of distribution systems. This paper provides metaheuristic optimization to enhance state-of-the-art prediction methods. We conducted a comparative study between SARIMAX, which proved to be effective for trends and seasonality as well as the impact of exogenous variables, and the GRU deep learning model, which captures complicated non-linear dependencies. Both the models were optimized with Genetic Algorithm (GA), a metaheuristic approach for efficient search in solution space. The effect of optimization was also tested by comparing the performance with and without GA. First, the results using the real dataset showed that SARIMAX was better than GRU. In the optimized version, the SARIMAX-GA model's predictive capability and understandable variance improved significantly compared to the GRU-GA model.