Paper Conference

Proceedings of eSim 2018: 10th Conference of IBPSA-Canada


Machine Learning Recommendations for Control of Complex Building Systems Using Weather Forecasts

Paul Willem Westermann, Nigel David, Ralph Evins

Abstract: We present a machine learning model used to provide recommendations on chiller operation based on the prediction of cooling demand using a weather forecast. A long short term memory (LSTM) formulation was used, and achieved favourable results compared to a standard approach. The model captured the data to a reasonable extent (R² = 0.70), but was unable to predict very high loads at unexpected times. The model is intended to be used as an aid to a human operator, not as a replacement, and it is likely that many of these unexpected events could be overridden by the operator. Overall, the predictive model reduced the number of occasions in which a chiller was operating unnecessarily by 80.5%, or 469 hours. This demonstrates the power of data-driven predictive control to assist in the efficient operation of complex building systems, saving money, energy and operator time.
Keywords: Energy management system, human-in-the-loop control, Machine Learning
Pages: 9 - 16