Energy Prediction for Smart Buildings Pertaining to the Heating Ventilation and Air Conditioning Plants Using Machine Learning Techniques

Abstract

The world is facing a critical issue of over-consumption of energy and its wastage. Most of the newlineenergy is consumed by Buildings. Further, within buildings, Heating, Ventilation, and Air newlineConditioning (HVAC) plant is the largest energy consumer. Several researchers have worked newlinein this area as to how the problem of energy wastage should be handled. Researchers have tried newlineto explore different aspects of energy wastage in buildings- Occupancy, Occupant behavior, newlinelack of robust methods for energy prediction and analysis, lack of Building management newlinesystems etc. Machine Learning has tremendous power that can be used to find solutions to newlinevarious real-life problems prevalent in society. An initial and important step to counter newlineenergy wastage is to accurately predict the amount of energy consumed in buildings and by newlineHVAC. newlineThis research addresses the problem of accurate energy prediction by proposing an approach newlineusing Machine Learning, Ensemble Learning, and Deep Learning techniques. The research was newlinecarried out using two datasets- One of them is a standard dataset, related to Energy newlineconsumption, that was collected from a public UCI repository. It consists of eight building newlineparameters as input variables and Heating load and Cooling load as two output variables. newlineAnother dataset related to Energy consumption was collected from a commercial building in newlineIndia. The dataset was raw, so it was pre-processed. Datasets were partitioned into 70%-30% newlineratio and 80%-20% ratio for Training and testing purposes respectively. Feature selection was newlineperformed on datasets by using Filter methods and Wrapper methods. Several Machine newlineLearning models were explored, namely MLR, SVR, KNN, CART. Ensemble Learning newlinemethods that were explored are- Bagging, RF, GBM and XGBoost. LSTM and GRU models newlinewere explored in Deep Learning. Hyper-parameter tuning was done to obtain the best possible newlineresults. Root Mean Square Error, Mean Square Error, Mean Absolute Error and R Squared newlinewere used as performance metrics for model evaluation.