Chiller energy prediction in commercial building: A metaheuristic-Enhanced deep learning approach

Chiller systems hold a critical role in upholding comfort and energy efficiency within commercial buildings. Precise prediction of chiller energy consumption is imperative for operational optimization and the reduction of energy expenditures. This paper introduces an innovative methodology that integrates deep learning (DL), specifically Fixed Forward Neural Networks (FFNN), with Teaching-Learning-Based Optimization (TLBO) to enhance the accuracy of chiller energy consumption forecasts. Drawing on a diverse dataset from a commercial building, encompassing vital input parameters such as Chilled Water Rate, Building Load, Cooling Water Temperature, Humidity, and Dew Point, the study conducts a comprehensive comparison of metaheuristic algorithms (Genetic Algorithm (GA), Particle Swarm Optimization (PSO), Barnacles Mating Optimizer (BMO), Harmony Search Algorithm (HSA), Differential Evolution (DE), Ant Colony Optimization (ACO), and the latest RIME algorithm). TLBO's adept navigation of the intricate parameter space of DL yields highly precise predictions for chiller energy consumption. The study's outcomes underscore TLBO's potential, along with other metaheuristics, in optimizing DL and refining energy management practices in commercial buildings. This research significantly contributes to the evolving discourse on the symbiosis between DL, particularly FFNNs, and metaheuristic optimization, offering a robust framework for chiller energy consumption prediction, thereby advancing sustainability and cost-effectiveness in building operations.