WP1.2 Coordination Meeting December 13, 2021

Meeting Report WP1.2 ‘Modelica library for MPC’

1. MEETING SUBJECT, DATE

Subject: WP1.2

Date: 13-12-2021

Location: Teams

Minutes taken by: Lieve Helsen (KU Leuven)

2. PARTICIPANTS

3. AGENDA

1. BOPTEST trials - controller benchmarking

   a. Extra BOPTEST workshop (October 26)

   b. Follow-up BOPTEST workshop (December 10)

   c. Controller characterization

   d. Integration of uncertainties in BOPTEST - weather forecast uncertainties

2. New developments

   a. Problems with emulators

   b. new KPIs needed

   c. new developments in BOPTEST

3. Outreaching

   a. Initiatives for joint papers

   b. Where have we seen BOPTEST popping up?

4. REPORT

1. BS1 BOPTEST trials - controller benchmarking

a. Extra BOPTEST workshop (October 26): 10 attendees onsite + 7 attendees online same content as BS2021 workshop.

b. Follow-up BOPTEST workshop (December 10): 8 attendees (incl RWTH Aachen, DeltaQ), not many extra results for the test case. Discussion on BOPTEST experience: Few issues with multi-zone emulator (Thomas raised an issue to be fixed) Stochastic scenarios are highly looked forward to! Advanced controller API (combining writing and advancing) might differ from real building operation -> separate writing from advancing (not high priority, rather an implementation issue for deltaQ). DeltaQ uses BOPTEST for teaching purposes to non-technical people, and integrates BOPTEST in their full test setup. How to motivate users to share their results such that benchmarking can be done? NREL dashboard is facilitating this, but this is still work in progress. Dave checks the status with Kyle. Automatic transfer of test case results to the dashboard (with manual (but easy through API and not as a separate application) inputs about the controller) would be an added value. Next meeting: presentation of the Dashboard (Lieve puts on the agenda)

c. Controller characterization: Javier makes a first draft with simplified branches on which we can iterate.

d. Integration of uncertainties in BOPTEST - weather forecast uncertainties (Laura).

    i. Integrate Milano data (Ettore), hourly predictions, time period 2014-2021, 5 locations in the region of Milano, updated only once a day (instead of hourly), 2 models. Results: frequency of update is not high enough: standard error is not increasing as a function of the prediction horizon, which was also the case for Leuven (update each 6 hours).
    
    ii. Can we keep 2 scenarios? High (Berkeley & Norway) and low (Berlin) standard deviation scenarios in the BOPTEST framework to check whether the controllers are robust towards these uncertainties. But there are two parameters to look at: high/low standard deviation and high/low mean error. For the latter, Berlin and Berkeley have the same magnitude (0.4-0.5) but opposite site and Norway has a small value (up to 0.2). No clear pattern to be seen, too few data.

    iii. Integration of uncertainty models in BOPTEST: Plan: Gaussian distribution with mean and standard deviation characterizes the uncertainty, which is superposed on the deterministic forecast using an autoregressive model, for each hour of the year a 24-hours forecast is created, initial error is assumed and distribution is used to create the next errors.

    iv. Solar radiation and temperature are not independent but correlated!

    v. Are these distributions Gaussian?

    vi. Numerical noise might be larger?

    vii. Either we need more data or we go for another approach

    viii. Action Laura: Autoregressive approach using high mean error and high standard deviation for Berkeley to be used in the emulator with TMY data -> to be discussed on next meeting.

2. New developments

a. Update on emulators: see below 

b. New KPIs needed: Dave to be implemented 
    
    i. peak power: maximum 15 minutes averaged power, one for each energy vector (gas, electricity, heat) in kW per square meter floor area (to be consistent with other KPIs).

    ii. energy: now in J or kWh per square meter floor area

c. new developments in BOPTEST

    i. field tests completed for validating a test case (Yeonjin)

3. Outreaching

a. Initiatives for joint papers

    i. Application oriented paper - BOPTEST trials paper. Aim: showing the capabilities of the framework. Which test cases to be included? Now mainly test results from BESTEST hydronic HP case, which is an important limitation. We need to extend to a full matrix: different controllers being tested on different test cases (hydronic and air-based). Too early now. Let’s take the time needed to learn and to improve. Who takes the lead? Javier

b. Where have we seen BOPTEST popping up?

    i. RWTH Aachen: control group wants to use BOPTEST

    ii. DeltaQ is an intensive user

    iii. UGent (Chris Develder) – data-driven control of residential buildings

    iv. BuildSys workshop presentation Dave – interest from RL community

    v. NeurIPS workshop debate Lieve – BOPTEST will be mentioned

4. Emulators

Updates:

• Multi-zone (5z) commercial air-based: stage of review (Filip)
• Multi-zone commercial air-based (Yeonjin): Uploaded to Github and reviewed by Dave, feedback is processed now. Field test is finalized to calibrate and validate the model (Yeonjin)
• Multi-zone prototype air-based (complex): Iago gave advice to speed up the simulation, another solver is tried now.
• Single-zone commercial hydronic: has been merged, ready to be used!
• Ettore is working on a 2-zone (with different occupancy profiles for the 2 zones) apartment case study in Italy using district heating or heat pump coupled to floor heating, on a local branch, no pull request yet.

Plan:

• Multi-zone office hybrid (complex): on hold (Filip)
• Single-zone commercial air-based: 1st version ready by December 2021 (Dave)
• Multi-zone office hybrid (simple): HVAC able to compile but very slow simulation due to large number of events. Trying to reduce the number of events now (Iago & Javier).

Status:

5. Further issues

Action Lieve: doodle for January-March-April progress meetings.

Agenda:

• presentation of the Dashboard (Kyle)
• Controller characterization: Javier makes a first draft with simplified branches on which we can iterate.
• Autoregressive approach using high mean error and high standard deviation for Berkeley to be used in the emulator with TMY data  to be discussed on next meeting (Laura)
• Implementation peak power KPI (Dave)
• Update emulators
• BOPTEST trials (different controllers and different test cases)