FIRE-CNN-LSTM: A Fuzzy Rough Set-Evolved Hybrid Deep Learning Model for Short-Term Load Forecasting Using Computational Intelligence

Short-term load forecasting (STLF) plays a pivotal role in power grid stability and economic dispatch, but conventional models often fail to address the dual challenges of data noise and complex spatiotemporal load dynamics. To bridge this gap, this paper presents FIRE-CNN-LSTM, an innovative hybrid computational intelligence model for short-term load forecasting that synergistically integrates fuzzy rough sets for uncertainty-aware data refinement, adaptive fuzzy membership functions for robust feature representation, and a Differential Evolution-optimized CNN-LSTM architecture for multi-scale temporal pattern learning. The proposed framework addresses critical challenges in power load forecasting by combining fuzzy logic's ability to handle data imprecision with deep learning's capacity for complex pattern recognition, further enhanced by evolutionary optimization of hyperparameters. Evaluated on real-world hourly load data from Malaysia, our model demonstrates superior performance with 60% RMSE reduction compared to conventional approaches, R2 > 0.999 prediction accuracy, and 22% improved generalization over non-fuzzy deep learning benchmarks. The work contributes to computational intelligence applications in energy systems by introducing a novel fuzzy-rough data preprocessing layer for noise resilience, developing an evolutionary-optimized hybrid neural architecture, and validating significant practical improvements in forecasting reliability that translate to 3-5% operational cost savings in grid management scenarios.