Energy efficiency in building: Entropy-based Grey Wolf Optimization for improved MLP performance

Energy efficiency in building HVAC systems is essential for sustainable development, requiring accurate predictions of heating and cooling loads. Traditional hyperparameter tuning methods often lead to suboptimal performance in artificial neural network models due to local optima and premature convergence issues. This study proposes an Entropy-Based Grey Wolf Optimization to enhance the hyperparameter tuning of Multi-Layer Perceptron models for improved load prediction. Unlike traditional Grey Wolf Optimization, which relies on a fixed linear decay for search balancing, the Entropy-Based Grey Wolf Optimization dynamically adjusts the search strategy based on entropy levels. This prevents premature convergence, enhances global search capabilities, and improves fine-tuning of Multi-Layer Perceptron hyperparameters for HVAC load prediction. The proposed optimization approach is compared with ten different optimization techniques, and results indicate that Entropy-Based Grey Wolf Optimization achieves superior accuracy with lower mean squared error and faster convergence. The study contributes to both the improvement of energy efficiency in HVAC systems and the development of a generalizable entropy-based optimization framework that can be applied to various machine learning tasks.