Work Plan

The approved proposal can be found here: PDF

Summaries of the Project and tasks are described below.

Overview (from proposal)

Needs for advanced and improved control strategies (CS) in building and district energy systems are growing due to requirements for reducing energy use, greenhouse gas emissions, and operating costs, providing flexibility to the electrical grid, as well as ensuring performance of novel hybrid and collective system architectures. Examples of such CS are advanced rule-based control, Model Predictive Control (MPC) [Drg20], and Reinforcement Learning [Wan20]. However, while these and other CS show promise, three challenges slow their widespread adoption:

1. The performance of each CS is typically demonstrated on individualized case studies and quantified using different metrics, making it difficult to benchmark and compare their performance, identify the most promising approaches, and identify needed development.

2. Demonstrations in real buildings and district energy systems pose large operational risks and difficult environments for controlled experiments.

3. Development of realistic simulation models for CS testing and evaluation requires significant building science and modeling expertise not necessarily held by experts from fields which could contribute to new CS development, such as process control, optimization, and data science.

The building simulation (BS) community can address these challenges by providing suites of publicly available, high-fidelity simulation models, called emulators, to be used for benchmarking CS. Furthermore, providing a comprehensive framework to deploy, interact with, and generate key performance indicators (KPI) from these emulators would ensure their benchmarking capability and make them readily available to related control and data science fields outside of the BS community. There exists precedent for such an approach within the BS field with the development of the BESTEST [Jud95] and subsequent ASHRAE Standard 140 [ASH11] as well as the optimization fields (e.g. Decision Tree for Optimization Software [Mit22]) and data science (e.g. OpenAI Gym [Ope22]).

Task 1: Outreach and Community Building

Task Leader: Javier Arroyo, KU Leuven, Belgium

This task will focus on activities that encourage, facilitate, and disseminate BOPTEST usage, adoption, and feedback to development.

Task 2: Methods and Infrastructure

Task Leader: David Blum, Lawrence Berkeley National Laboratory, USA

This task will focus on development and maintenance of core framework software and closely related extensions in response to needs identified by Project participants and community feedback. Related components include test case and framework architecture specification, FMU simulation and data management, KPI calculation, forecast delivery, API, online results sharing dashboard, web-service, and interface extensions.

Task 3: Test Cases

Task Leader: Ettore Zanetti, Lawrence Berkeley National Laboratory, USA

This task will focus on development and maintenance of benchmark emulators, so-called “test cases.” Emulator development will continue to utilize the Modelica language and Functional Mockup Interface (FMI) standard, particularly open-source libraries that extend from the Modelica IBPSA Library developed through IBPSA Project 1 WP1.1 and continuation Modelica Working Group.

Task 4: Controller Testing

Task Co-Leaders: Esther Borkowski, ETH Zurich, Switzerland and Zhe Wang, Hong Kong University of Science and Technology, Hong Kong

This task will focus on testing and benchmarking CS developed by both participants in this Project, and also comparison with CS developers external to the Project, using the framework developed in Task 2 and test cases developed in Task 3.

References