A data-driven model predictive control strategy with integrated rule mining to optimize HVAC operations under multiple occupants’ activities

Abstract

A data-driven model predictive control (MPC) strategy embedded with rule mining was proposed to discover optimal relationships between Indoor Air Quality (IAQ) events and operations of Heating, Ventilation and Air Conditioning (HVAC) systems to optimize IAQ, building thermal comfort, and energy consumption. In this strategy, a rule mining method was used to discover the relationships between occurrences of IAQ events and optimal HVAC operations, including occurrence time rules, co-occurrence rules and sequential occurrence rules. An encoder-decoder Long Short-Term Memory (LSTM) model was used to predict future building performance, and an event detection method was developed to identify the occurrence of pollutants’ events based on the prediction and real-time observations. With the detected occurrences of events, the rules derived from the rule mining method were used to provide preconditioned fuzzy optimal HVAC operations, which were then used to improve the Firefly algorithm (FA) to generate control settings. Simulation tests based on a house with a cooling system showed that, by using the MPC strategy, the pollutants’ peak concentrations of CO₂, NO₂ and PM2.5 were reduced by 25.4 %, 22.8 % and 35.3 %, respectively, compared with those using the baseline strategy. The exposure times of high concentrations of CO₂, NO₂ and PM2.5 were reduced by 400 min, 50 min and 55 min and 8.8 % energy savings were achieved. The HVAC energy consumption using MPC with rules was 5.1 % lower, and the pollutants’ peak concentrations of CO₂, NO₂ and PM2.5 were 13.2 %, 23.4 % and 22.3 % lower, respectively, in comparison with using MPC without rules.