Accurate cooling load estimation using multi-layer perceptron machine learning models

The article emphasizes the critical significance of accurately anticipating cooling demand, spanning diverse sectors like commercial structures, housing developments, and industrial facilities. This precision is paramount for achieving sustainable and energy-efficient building control. Powerful tools for enhancing the accuracy and adaptability of cooling load forecasts have been introduced through the emergence of Machine Learning (ML) approaches. Methods for forecasting cooling demand are extensively examined in this scholarly document, with a particular emphasis placed on implementing Multi-Layer Perceptron (MLP) models within the field of ML. The advantages of MLP models in addressing non-linear relationships, managing multidimensional datasets, and accommodating fluctuating environmental conditions are exemplified. A predictive analysis of building energy consumption was undertaken to enhance operational efficiency, employing two distinct optimization algorithms, specifically, the Improved Grey Wolf Optimizer and the Honey Badger Algorithm. In contrast to the MLIG model, characterized by high error values during training, an average reduction of prediction errors by more than double was achieved by MLHB. Furthermore, a peak value of R2 = 0.995, which is 3.4% higher than that of MLIG, was attained for cooling load estimation by MLHB in the second layer.