Improving a neural networks based HVAC predictive control approach

This paper improves an existing Model Based Predictive Control Approach (MBPC), applied for Heating Ventilation and Air Conditioning (HVAC) control in buildings. The existing approach uses the Predictive Mean Vote (PMV) to assess thermal comfort. It has been found that PMV estimation and forecasts deteriorate when the room is occupied. In order to solve this problem, we propose to incorporate measurements of activity inside the room in the predictive models of the inside air temperature. Another improvement to the existing approach is to use an economic cost function, reflecting the money needed for the HVAC control, instead of a cost function related with the consumption of energy.