The increasing proliferation of Internet of Things (IoT) devices in energy infrastructure has accelerated the demand for high-resolution forecasting models capable of accurately predicting energy consumption from time series data. In this context, this study addresses the challenge of short-interval energy forecasting by leveraging Bidirectional Long Short-Term Memory networks (BiLSTM) enhanced through advanced hyperparameter tuning. We introduce a novel metaheuristic, the iHow Optimization Algorithm (iHowOA), inspired by human cognitive learning processes, to optimize BiLSTM architecture for improved generalization and accuracy. Our framework is evaluated on a real-world IoT-based HVAC blower energy consumption dataset, recorded at 10–15-minute intervals. Initial baseline modeling using BiLSTM yielded a mean squared error (MSE) of 0. 008487059, while the iHowOA-optimized BiLSTM model achieved a substantially lower MSE of 9. 42 10^-7, reflecting a near order-of-magnitude improvement. These results demonstrate the strength of coupling human-inspired metaheuristics with deep sequence models for energy forecasting tasks. The proposed approach offers a scalable and adaptive solution for intelligent energy systems, enabling real-time prediction, optimization, and integration within smart building management frameworks.
Alharbi et al. (Thu,) studied this question.