Efficient and Robust Training Methodology for Inverse Building Modeling and Its Application to a Multi-zone Case Study

This paper presents an efficient and robust parameter training methodology, based on a previous approach for inverse building modeling that utilizes a simplified state-space approach. One new element of this training methodology is that some seasonal effects, such as variation of window transmittance at different times of the year, are taken into consideration and captured during the training process. In addition, a mixed-mode training approach is developed that allows the use of a combination of data obtained when cooling or heating is occurring with the zone temperature under control at setpoint and when the zone temperature is floating during periods of no load. To obtain a “nearly” global optimal model, a multi-start search method was found to be robust and provide good computational efficiency and accurate results. The training methodology is implemented to model three zones of Building 101 at the Navy Ship Yard in Philadelphia, Pennsylvania.