Automated grey box model implementation using BIM and Modelica

Abstract

A large part of energy usage in buildings occurs during the operational phase, emphasising the need for efficient and improved facility management, operation and control. Model Predictive Control (MPC) or Fault Detection and Diagnosis (FDD) are among the strategies that allow minimising energy use and costs during operation. However, the need for fast and accurate dynamic models (e.g. grey box model), which are time-consuming and challenging to implement, precludes their systematic integration in the built environment. A typical grey-box modelling approach consists of manually implementing several grey-box model structures with an increasing level of complexity before performing a forward selection procedure to identify the optimal configuration. The link between the different grey-box models and the monitored data is also established manually. Such an approach can be both time-consuming and error-prone and involves a significant cost that hampers the broad adoption of strategies such as MPC and FDD. This study proposes a tool-chain that uses BIM to automatically generate several grey-box structures with added complexities stemming from the specific geometry and design of the building. More specifically, an existing rule-based IFC to Modelica interface is extended to automatically create several Modelica-based grey box models that gradually take into account the building's specific information and characteristics. Additional rules are also proposed to automate the connection between the models and the building monitoring system. As a forward selection approach, a multi-objective optimisation using the NSGA-2 algorithm is adopted. The application of the tool-chain on two case studies shows that the integration of BIM to automate the implementation of grey box models, not only reduces the human involvement in the modelling process but can also produce more accurate models. Besides, this study shows that the use of multi-objective optimisation with datasets from two different seasons results in models that are valid for all seasons.

Envisioning Building-as-Energy-Service in the European context. From a literature review to a conceptual framework

“…For example, the concept of grey box (e.g. Andriamamonjy et al, 2019) can be expanded, taking into account a Buildings-as-Energy-Service vision, which can be crucial to the generation of new energy services, urban policies and building codes.…”

“…Additionally, they can be used to characterise the behaviour of technologies in experimental test-facilities (Oliveira Panão, Santos, Mateus, & Carrilho da Graça, 2016). Furthermore, they can be integrated into the design process using Building Information Modelling software (Andriamamonjy, Klein, & Saelens, 2019). Therefore, they can create a certain degree of continuity among different phases of performance analysis during the building life-cycle, from design to operation (Lehmann, Gyalistras, Gwerder, Wirth, & Carl, 2013).…”

Quick energy prediction and comparison of options at the early design stage

“…However, the energy-efficient design is not intuitive, and energy simulation must be an integral part of the design process. Researchers have previously developed methods to integrate architectural modelling and energy simulation tool [1,11,[54][55][56]. Dynamic energy simulation tool such as EnergyPlus [11] or Modelica [54,57] have been integrated into an architectural modelling tool.…”

“…Researchers have previously developed methods to integrate architectural modelling and energy simulation tool [1,11,[54][55][56]. Dynamic energy simulation tool such as EnergyPlus [11] or Modelica [54,57] have been integrated into an architectural modelling tool. This integration was focussed on minimising remodelling efforts by extracting information from the BIM model to develop BEM [58][59][60].…”

Smart Meter Data Analysis of a Building Cluster for Heating Load Profile Quantification and Peak Load Shifting

“…Supervised machine learning mostly relies on the historical measured data, which are used to develop black-box models to predict future energy consumption (profiles), e.g., by using ANN [24] or applying a linear prediction model for MPC [25]. However, it requires a large amount of historical data to be adequately trained, and the results sometimes are absent with physical meaning [26].…”