-
Notifications
You must be signed in to change notification settings - Fork 39
WP1.2 Coordination Meeting November 12, 2019
Meeting Report WP1.2 ‘Modelica library for MPC’
Subject: WP1.2
Date: 12-11-2019
Location: Skype for Business
Minutes taken by: Lieve Helsen (KU Leuven)
Company/Organisation Participant
KU Leuven Lieve Helsen (WP Leader)
KU Leuven Filip Jorissen
KU Leuven Iago Cupeiro (now @ Université Politechnique de Montréal)
LBNL David Blum
IK4 Tekniker Jesus Febres
SINTEF Harald Taxt Walnum
PNNL Chen Yan
PNNL Jan Drgona
SDU Konstantin
SDU Toa yang
Politecnico de Milano Ettore Zanetti
Approved
4 models selected to start MPC testing
Model reformulated and CSV data for BOP-TESTing, documentation needs to be finalized and to be sent to Filip (action Dave).
Iterated on documentation table (revised checklist will be sent around – action Dave), some more documenting needed (action Filip).
Review has been finalized (Toa).
Meeting organized with Valentin, documentation ready, waiting for review by Valentin.
BOP-TEST ready, review by Valentin is finalized.
Internal testing ready, documentation finished, when ready a meeting will be scheduled with Iago for review.
New directory will be created by Dave, maybe later a new repository for the emulators.
Representative days/weeks (PNNL): further improved by looking into the energy end use. Not all is based on weather only, also on type of the day.
PCA, clustering and sampling to generate regression.
Process differently, in progress.
Development needs identified in Rome Meeting
-
Compilation of test cases that don’t have signal or exchange blocks should receive a warning: pull request has been sent to review (Dave).
-
Docker on Windows generated some issues: documentation needed to get the issues fixed: user guide to be made (issue on BOP-TEST repo till we have a comprehensive user guide)
-
Allow a browser to access and send KPI on docker
-
Start simulation from a custom time
-
…
They have a highly measured and highly controllable test building which allows calibration and maybe experimental validation of BOP-TEST.
To be clarified by Dave what is expected – meeting Dave & Jesus takes place after this meeting.
Initial test of implementing noise (truncated normal noise) in the reader block done but this increases the simulation time a lot (from 1 to 50 minutes). Base-line controller should be designed such that noise is expected, e.g. by including hysteresis. Is it related to the block itself or due to the controller design? Harald checks. Better to implement the noise in the BOP-TEST framework, related to the measurement data?
Consider using historic data instead of TMY, or determine how to create new TMY. EnergyPlus has a tool (ETW) to create TMY data with a subset of measured data, however these should be detailed, difficult to obtain on a broad scale. Dave checks whether Michael has ideas. No update yet. What is the minimum set of data needed in our Modelica models, radiation solar gains? TMYx data is a project that generates TMY data for more locations (e.g. all meteo measurement locations in Norway). Dave looks deeper into the paper describing TMYx:
Climate.OneBuilding.org and paper
3 types of merges were defined
- direct copy: OK
- direct copy with parameter changes: OK
- more advanced changes
Pull request to merge, two comments from Dave, types 1 and 2 were merged, request to do for the 3rd type too (Action Filip).
First small models (Dave). Specific examples (unit testing) needed for models that have been changed substantially: unit testing framework in progress, different issue with different pull request.
Internal testing (of whatever controller) on 4 (or even 6) emulators by Canada meeting. Call to start internal testing!
What needs to be reported (KPIs, controller characterization) - spreadsheet. First draft of spreadsheet to be made in Canada Meeting based on first controller tests.
Limited information.
Permission to share the information publicly.
Dave just finished the review, Jan is revising, Lieve will do the final review.
b. Special issue on Model Predictive Control in energy conversion systems in IEEE Transactions on Energy Conversion:
Call for papers here. Prior to preparing a full paper, an extended abstract of 500-1000 words should be emailed in PDF to the Guest Editor-in-Chief, Tomislav Dragičević ([email protected]). The abstract should concisely describe the main idea of the paper and make a clear case regarding the novelty and technical contribution of the work. The team of Guest Editors will evaluate the submitted abstracts and invite authors to submit full papers, which will then undergo a peer review process. The full manuscripts will be submitted in electronic format through the Manuscript Central web site, by 1st March 2020.
LBNL will be participating (data management, data for fault detection, data-driven models for MPC). More focus on identification techniques for grey- and black-box controller models.
Opportunity to link with the ML community, they need high-fidelity models for testing their reinforcement learning methods (Jan). Dave has been contacted multiple times (IBO Boulder, BS2019 Rome, TAG meeting US) in this context. Dave mentions a challenge announced for urban load control. The paper referenced on the website (www.citylearn.net) looks to use openAIGym for this