Bio-inspired puma optimization algorithm for time-series energy consumption forecasting

Abstract

Accurate energy consumption forecasting represents a critical challenge in smart home environments, where consumption patterns are influenced by complex interactions between internal appliance usage and external environmental factors. Traditional forecasting approaches often struggle to capture the non-linear temporal dependencies inherent in energy consumption data, necessitating advanced methodologies that can effectively model these intricate relationships for enhanced energy management systems. This study presents a novel hybrid forecasting framework integrating the Bio-Inspired Puma Optimization (PO) algorithm with Long Short-Term Memory (LSTM) networks for time-series energy consumption prediction. The methodology employs a comprehensive dataset containing real-time energy consumption data from household appliances coupled with meteorological variables including temperature, humidity, and wind speed. The framework incorporates systematic hyperparameter optimization through the PO algorithm and implements binary Particle Swarm Optimization-guided Whale Optimization Algorithm (bPSO-WOA) for intelligent feature selection. The proposed PO-optimized LSTM model demonstrates superior performance across multiple evaluation metrics, achieving a Root Mean Squared Error (RMSE) of 0.00582 and coefficient of determination (R²) of 0.98060. Comparative analysis reveals significant performance improvements over traditional machine learning approaches, including Gated Recurrent Units, Recurrent Neural Networks, and Artificial Neural Networks. The integration of feature selection techniques resulted in substantial computational complexity reduction while simultaneously enhancing model robustness and interpretability. • Hybrid PO-LSTM model for smart home energy time-series forecasting. • bPSO-Guided WOA feature selection reduces complexity and boosts accuracy. • PO-optimized LSTM achieves RMSE 0.00582 and R² 0.98060 on residential data. • Outperforms GRU, RNN, ANN and other metaheuristic-tuned LSTM models. • Supports demand-side management and smart grid residential applications.