A.I. model forecast of building cooling load demand for the reduction of energy consumption to work towards carbon neutrality

Abstract

The world needs to achieve carbon neutrality or net zero emissions of greenhouse gases (GHG) by 2050. Buildings are major sources of GHG emissions. Applications of the latest innovative technologies of machine learning/A.I. algorithms have opened up new opportunities. The optimal control of cooling plant systems is important to reduce energy consumption and therefore emissions. Knowing the cooling load demand in advance can help facility managers operate cooling plants much more efficiently. This paper presents a real-life application of nine A.I. models for time-series forecasting of the cooling load demand of a commercial building. LSTM neural networks, Facebook Prophet time series model, and DeepAR recurrent neural network models are found to be the most accurate with a Mean Absolute Percentage Error (MAPE) in the range of 15 to 16 with a computing time in the range of 294 to 319 seconds respectively. The LightGBM machine learning model on the other hand proves to be the fastest with a MAPE of 18.96 in just 7 seconds. Thus, different models can be deployed for different requirements. Optimising the operation of cooling systems as per the forecast cooling demand can bring enormous energy savings that are essential for achieving carbon neutrality.