Interpretable deep learning for hourly solar radiation prediction: A real measured data case study in Tokyo

Yuan Gao, Shohei Miyata, Yasunori Akashi
Journal of Building Engineering, Volume 79, 2023, 107814
https://doi.org/10.1016/j.jobe.2023.107814

Abstract
Model-based optimal demand-controlled ventilation (DCV) for multizone variable air volume (VAV) systems has significant potential for reducing energy consumption and enhancing occupancy comfort. However, the complexity of ventilation duct networks, building thermal dynamics, and the high computational demand for optimization pose challenges for widespread deployment in real buildings. To address these issues, we propose an event-driven model-based optimal DCV control for multizone VAV systems. The ventilation duct network is represented by an artificial neural network model, and the building thermal dynamics are captured by a multizone thermal network model, both of which are integrated into the control scheme. Unlike conventional approaches, the proposed strategy features an event-driven mechanism that triggers optimization only when necessary, thereby reducing the overall computational load. The controller determines the optimal fan frequencies and damper openings, minimizing energy consumption while maintaining a satisfactory indoor environmental quality (IEQ). Simulation comparisons and case studies validate the proposed strategy against different control methods. Compared to the time-driven method, the proposed strategy achieves similar performance while reducing the optimization runs by 70.83% with a small threshold throughout the occupied period. Additionally, it reduces the total IEQ cost by over 90% compared to well-tuned proportional-integral algorithm-based control and by 70% compared to setpoint optimization. Furthermore, flexible tradeoffs can be made based on the priority of reducing the computational load or maintaining the IEQ.