< Back to previous page

Project

In-situ data-driven estimation of dynamic solar gains in buildings: Grey- and black- box approaches

On-site measured data are increasingly combined with statistical techniques to create reduced-order models, which are used for evaluating a building's actual energy performance, model predictive control (MPC), defect detection, and energy grid optimization. Solar gains, the part of the buildings’ indoor energy gain contributed by the sun, play a vital role in the indoor heat balance and thermal dynamics. A precise estimation of dynamic solar gains can enhance the quality of the reduced-order models. Gauging solar gains and their time dependency in practice is, however, challenging. Therefore, most reduced-order models estimate solar gains based on assuming the solar gain coefficient (solar aperture, gA-value) as constant. This assumption results in solar gain estimation uncertainties since, in reality, the gA-value is highly dependent on the sun's position during the time of the day and year.

To fill this gap, in this thesis, two statistical modelling approaches are proposed to estimate and predict the time-varying solar gains more precisely. The two techniques were developed, based on in-situ measurement data on a simplified reduced-size building, as parallel approaches for gauging buildings’ solar gain dynamics to satisfy different scenarios (e.g., physical interpretation-demanded or prediction accuracy-orientated). The two approaches are known as B-splines integrated grey-box and ARX models (autoregressive with extra input) and their outcomes are verified by white-box simulations. Additionally, the robustness of B-splines integrated grey-box models is further tested in more sophisticated and realistic building cases, starting with data from a reduced-size building with well-controlled heating inputs (i.e., pseudorandom binary sequence PRBS signal) to full-size buildings with synthetic occupant profiles in use. It revealed that the robustness of the proposed B-splines integrated grey-box modelling is excellent, at least in revealing the primary dynamic trend of solar gains for a specific building case. The two modelling approaches proposed in this study, which are based on limited low-frequency data, can partially fill the gap in absence of an efficient and precise method for estimating solar gain dynamics. The proposed two methods i.e., B-splines integrated grey-box and ARX models show the potential to reduce prediction errors of building energy conservation in practice.

The two data-driven modelling techniques, i.e., B-splines integrated grey-box and ARX modelling are presented in Chapter 2, based on on-site datasets from a one-zone reduced-size building with one level on/off PRBS heating signal. The promising performance of both techniques is demonstrated in this chapter. From Chapter 3 onwards, the key focus of this thesis is on B-splines integrated grey-box models since grey-box models generally have enhanced extrapolation capacity compared to black-box models (e.g., ARX model). Chapter 4 examines the potential impacts of detailed modelling of solar gains on heat loss coefficient (HLC)-determination in grey-box models. The results show elaborating solar gains in grey-box models does not hamper reliable estimations of HLC. Chapter 4 investigates the robustness of B-splines integrated grey-box modelling in the same reduced-size building but with more realistic settings, where two thermal zones are established, and energy gains related to two (synthetic) occupants are imitated by heaters. Chapter 5 further explores the robustness of B-splines integrated grey-box modelling in a more realistic full-size building case. Chapter 6, finally, summarizes the key findings of this dissertation and the limitation of this Ph.D. research and presents opportunities for future works based on this Ph.D. dissertation.

Date:15 Jan 2019 →  9 Jan 2023
Keywords:Overall heat loss coefficient, Dynamic statistical analysis, Building energy efficiency
Disciplines:Building physics
Project type:PhD project