Data-Driven Optimal Air-Balancing Control for Multizone Ventilation Systems with Design-to-Operation Adaptation

Shanrui Shi, Shohei Miyata, Yasunori Akashi
EIA Nordic 2025, Lecture Notes in Computer Science, vol 16096, Nov. 2025
https://doi.org/10.1007/978-3-032-03098-6_11

Abstract
Accurate air-balancing control is important for maintaining indoor air quality and reducing fan energy consumption in multizone ventilation systems. While data-driven approaches can outperform conventional control methods, their performance depends on their access to large volumes of high-quality operational data. To address this limitation, this study proposes an ANN-based optimal air-balancing strategy that bridges the design and operational phases. A physics-based duct network model was developed using Modelica to generate synthetic operational data. An artificial neural network (ANN) surrogate was then trained to capture the nonlinear actuator–airflow relationship. A hybrid optimization algorithm combining improved particle swarm optimization and Adam was developed to compute the optimal combinations of fan frequency and damper positions while satisfying the operational and energy-saving constraints. Simulation results benchmarked against the ASHRAE Guideline 36 trim-and-respond sequence show that the proposed strategy delivers more accurate air balancing with improved energy efficiency. To assess the robustness to parameter drift across project phases, both uniform and random perturbations are applied to the physical model. Fine-tuning the ANN with as little as 3% of new operational data restored, and often exceeded, the accuracy of the model retrained from scratch. These results demonstrate that the proposed strategy enables high-precision, phase-resilient air-balancing control with minimal real-world measurement effort.