Extended virtual in-situ calibration method in building systems using Bayesian inference

Measurements from sensors and knowledge of key parameters are of great importance in the operation of modern building systems. Accurate and reliable information as these serves as the base for ensuring the desired performance of control algorithms, fault detection and diagnostics rules, analytical optimization strategies. They are also crucial for developing trust-worthy building models. However, unlike mass produced industrial devices, building systems are generally one of a kind and sparsely instrumented. Despite the indispensable need, dense deployment of sensors or a periodic manual calibration for ensuring the quality of thousands variables in building systems is not practical. To address the challenge, we extend our virtual in-situ calibration method by marrying it with Bayesian inference, which has a better capability in handling uncertainties. Strategies, including local, global, and combined calibration, are evaluated in a case with various sensor errors and uncertain parameters. The detailed procedure and results are presented.