Improving cooling load prediction in residential buildings with multi-layer perceptron models

Today, building energy efficiency is prioritised since it affects operational costs. Buildings take a lot of energy to maintain pleasant temperatures. Combining this research's cooling load (CL) forecasting method may optimise building energy use. MLPs forecast household cooling demands. MLP models and regressions generally have linear input-output relationships. This research uses two innovative optimisers, cheetah optimiser (CHO) and adaptive opposition slime mould algorithm, to improve MLP model performance. The data used to train the approaches describes each sample's unique traits. These methods will be tested on a simulated dataset using CLs as neural network output variables and building technical attributes as input factors. During the process testing phase, the MLCO (2) (MLP+CHO in layer 2) gets the lowest RMSE value of 0.672 and the greatest R2 value of 0.995. The results demonstrate that the proposed hybrid models - MLCO and MLAO - significantly outperform the standalone MLP and conventional optimisation methods, achieving a minimum error rate. These findings confirm the proposed models' superior predictive accuracy and reliability, underscoring their potential for practical application in enhancing energy efficiency in residential buildings.