Hvac Energy And Occupant Comfort Optimization Using Neural Network-Based Temperature Prediction

Abstract

This thesis presents an approach that utilizes an NN-based model of an HVAC system to optimize energy consumption while keeping the thermal discomfort of occupants under a specified level. The approach provides an accurate prediction of the future indoor room temperature of a building to optimize the current HVAC control decision. We evaluate our approach by comparing it to two baselines that use reactive warming and complete prewarming. We initially test the three schemes assuming perfect occupancy prediction, and then explore the effects of different specified comfort levels and imperfect occupancy prediction. The results show that our approach optimizes future energy consumption based on a set of past system states, when given a custom threshold of the thermal discomfort of occupants. Our approach achieves a better trade-off between energy and discomfort and automatically reacts to the changes of external factors such as weather.