Multi-objective optimization of artificial neural networks using Fast NSGA-II for electricity demand forecasting

Accurate short-term electricity demand forecasting is a critical requirement for modern power systems, as forecast errors directly affect generation scheduling, market prices, and operational costs, particularly under dynamic pricing environments and increasing demand volatility. This study proposes an integrated forecasting framework combining Artificial Neural Networks (ANNs) with Multi-Objective Optimization (MOO) to jointly improve predictive accuracy and model parsimony. A statistical feature selection process is first employed to identify the most relevant input variables. The initial weight configuration of the ANN is then optimized using the Fast NSGA-II algorithm by simultaneously minimizing the forecasting error and the L1-norm of the network weights. From the resulting Pareto-optimal solutions, the TOPSIS decision-making method is applied to select the most balanced trade-off model. Experimental results on real French electricity-consumption data show that the optimized ANN consistently outperforms standard ANN, LSTM, and SARIMAXbaselines, delivering lower forecasting errors and more stable performance across the prediction horizon. Moreover, the proposed approach enhances cost awareness, reducing the cost gap to actual expenditures to 1.29–1.39% depending on the consumption period, and better captures demand elasticity, with forecasting accuracy improvements of up to 25.9%, underscoring its practical relevance for real-world power-system operation and market-oriented decision-making.