Enhancing Energy Efficiency in Cooling Systems through Advanced Machine Learning and Meta-Heuristic Algorithms for Precise Cooling Load Prediction

Abstract

Estimating the cooling load is essential for cooling systems’ energy efficiency, affecting chiller optimization and air conditioning control. Regression analysis and other conventional techniques are insufficient, but AI and ML have superior capabilities. AI and ML models are able to independently capture complicated patterns, adapt to changing settings, and scale with data efficiently to anticipate cooling loads that are impacted by variables such as weather, building materials, and occupancy. The dynamic, responsive predictions that result from this adaptability improve energy optimization and effective building management. This study delves deep into the intricate landscape of cooling load systems, where energy optimization complexities present challenges requiring thorough investigation and innovative problemsolving approaches. The Least Square Support Vector Regression (LSSVR) model is seamlessly combined with 2 meta-heuristic methods, the Dandelion Optimizer (DO) and the Electric Charged Particles Optimisation (ECPO), for improving precision and correctness. These algorithms use cooling load data that has undergone extensive validation through previous stability tests. 3 different models, LSDO, LSECH, and an independent LSSVR, are introduced in the study, each of which provides insightful information for accurate cooling load prediction. With an impressive R^2 of 0.990 and an incredibly low RMSE of 0.927 , the LSECH stands out as a standout performer. These outstanding outcomes demonstrate the LSECH model’s ability to accurately predict cooling load outcomes, emphasizing its potential for use in actual building management scenarios.

Keywords

• building energy
• cooling load
• least square support vector regression
• dandelion optimizer
• electric charged particles optimization