Description

Natural ventilation (NV) with automatic windows is promising to improve indoor air quality while reducing building energy consumption. However, the multi-objective optimal control of the automatic windows is challenging with conventional rule-based heuristic control (RBC) methods, as multiple factors need to be considered simultaneously. Model predictive control (MPC) provides a promising solution to overcome the limitations of RBC, but complex nonlinear dynamics, for example, NV, imposes difficulties in system modeling via widely used linear models in MPC. Machine-learning-based approaches have been demonstrated with great potential in nonlinear dynamic system identification. However, forecasting for a long prediction horizon may become unreliable when the model is recursively called, and the prediction error is accumulated during the process. To address these challenges, this study proposes a deep-learning-based MPC for multi-objective optimal control of NV with automated windows with an ensembled Long-Short-Term Memory (ensembled-LSTMs) model involved in system identification. We found out that the proposed ensembled-LSTMs model offers better prediction with maintained accuracy for long-prediction horizon forecast. Over 10 simulation days, the MPC maintained 93.7% of the simulated time with CO2 concentration of under 800 ppm and 2.6 unmet hours with CO2 concentration of above 1000 ppm. In contrast, in the baseline model with RBC, the CO2 concentration was maintained less than 800 ppm for 70% of the simulated time and there were 8.6 unmet hours. Regarding thermal comfort of the MPC case, 99.9% of occupied hours were maintained with the indoor air temperature of above 20°C (68°F) and none of the occupied hours was lower than 19°C (66.2°F). In addition, the MPC case consumes a similar level of energy consumption as the baseline with RBC does. To conclude, the proposed deep-learning-based MPC significantly improved indoor air quality without compromising occupants’ thermal comfort and heating energy consumption.

Product Details

Published:

2023

Number of Pages:

9

Units of Measure:

Dual

File Size:

1 file , 3.8 MB

Product Code(s):

D-AT-23-C070

Note:

This product is unavailable in Russia, Belarus