WP1.2 Coordination Meeting February 20, 2020

Meeting Report WP1.2 ‘Modelica library for MPC’

1. MEETING SUBJECT, DATE

Subject: WP1.2

Date: 20-02-2020

Location: Skype for Business

Minutes taken by: Lieve Helsen (KU Leuven)

2. PARTICIPANTS

Company/Organisation - Participant

KU Leuven - Lieve Helsen (WP Leader)

KU Leuven - Filip Jorissen

KU Leuven - Iago Cupeiro

KU Leuven - Javier Arroyo

LBNL - David Blum

LBNL - Michael Wetter

PNNL - Sen Huang

SDU - Toa yang

Politecnico de Milano - Ettore Zanetti

SINTEF - Harald Taxt Walnum

3. AGENDA and REPORT

3.1. Approval/Modification Agenda

Approved

3.2. Status & further discussion where needed

a. Emulators:

4 models selected to start MPC testing

i. BESTEST Air (LBNL/KU Leuven):

Dave made adjustments, final comments made by Filip, will be finalized (action Dave). Test case data are collected as CSV files, where headers have been adjusted to have more information, Dave adds to the review checklist

ii. BESTEST Hydronic (KU Leuven /LBNL):

weather data to be finalized (JModelica issues – old version), Dave sends the file generated in the new JModelica version. Javier has given feedback.

iii. 8-Zone Residential Hydronic (Engie/SDU):

Review has been finalized (Toa) and sent to Valentin. Confirmation needed by Valentin (action Valentin).

iv. Multi-Zone Office Hybrid Simple (KU Leuven/Engie):

Sent to Valentin for review, Valentin has still problems with the Modelica settings. (action Iago and Valentin). No update from Valentine. Lieve contacts Valentin. In the meantime Iago adds signal exchange blocks.

However further progress has been made in other models too!

v. Single-Zone Commercial Building (SDU):

Ready, to be put on the BOPTEST repository (Tao)

vi. Multi-Zone Air-Based Prototype Complex (PNNL):

Model sent by Sen to Iago for review.

Emulators on BOP-TEST repository in the directory Testcases:

b. KPIs - scenarios

Choice between TMYx (1946-2018) and TMY3 (1973-2005) weather files. We have a data reader for TMY3 in IBPSA. Both will be compared in the representative days/weeks selection (action PNNL). No update.

c. BOP-TEST:

i. JModelica is no longer open source

How to continue? PyFMI is still available to simulate FMUs. Compilation of emulator models is an issue. For now we keep on using JModelica (as long as it is compatible), however Python 2.7 is needed for compilation of models. We may run into trouble on the long run, but new solutions may pop up by then. For now it is frozen on docker. Michael comments that there is progress in OpenModelica, more than in the past.

ii. Test case peer review document (Dave):

Multi-zone model: which measurements correspond to which zone? Zone is added to the read block for KPIs associated to a zone, and zone tags are added in the CSV files. Updated code in IBPSA Library, pull request will be sent (Dave). Documentation should describe the building: spreadsheet links to the emulator used.

iii. Google spread sheet to post results of BOPTEST test cases (Javier):

See https://docs.google.com/spreadsheets/d/1E-5wR7nasW8h6kEtrXnUzcRrEA6f7TIxjbtSdB1Cz3w/edit will be later integrated in the website.

To which level of detail should the controller be described? High-level controller description (pre-defined) is needed to enable some statistics. What do the testers want to share? Give the option to non-disclose some information. Give the option to link to documents/papers for more information.

Number of states: differentiate between discrete and continuous states? A good measure for complexity or scale is needed. How to benchmark the complexity of an algorithm - see literature on optimization. We could agree on a list of important measures (optional). The emulator type selected gives already an idea about the complexity of the building case considered.

iv. How to parameterize and implement measurement noise in the Modelica model (Harald):

No further update, how to tackle the increase in simulation time? Michael: sample signal before sending it to the controller and add the noise there should not increase the computation time significantly. However that’s then only on the variable exported, not on the measurement data themselves. Sample noise in the baseline controller case using the same sampling time as in the MPC. Low-level controllers are not that sensitive to measurement noise. Maybe internal load and occupant uncertainty may be more important. Up to now emulators without noise.

v. How to define workflow for training data for grey-black-box models different than testing data (action Dave)?

No update

d. MPC Library (IbpsaMpc):

i. JModelica

JModelica no longer open source! Filip checked the conditions with Hubertus: final version stays open source, we can continue with that academic version won’t be too expensive Python 3 is supported in the OCT, updated to CASADI 3.2 Optimization still in Python 2

ii. Unit testing framework (action Dave):

No progress

e. Dissemination:

Broaden the user space (e.g. to Machine Learning people). How to reach these people? Action Jan: make a first suggestion. Cfr OpenAIGym, no update

5. Initiatives joint papers

a. MPC Review paper: Lieve is performing the final review.

6. Miscellaneous

a. Break-out sessions definition (progress meeting of March 19)

WP Leaders coordination meeting on April 27 (Lieve)

i. Spreadsheet for internal reporting of BOPTEST test cases

ii. Public reporting (limited information, end-user agreement)

iii. MPC competition: how to set up?

iv. How to deal with uncertainties: forecast of weather, loads, occupants?

b. Will join the Montreal meeting:

Harald, Dave, Dongun Kim, Jan Drgona, Lieve, Filip, Javier, Iago, Michael