Research Axes - Axis 1

Research Axis 1: Model predictive control and optimized responsive building operation

Project 1.1 Development of grey box/data driven models of building zones and main HVAC components

Collaborators: J. Bouchard (HQ), R. Zmeureanu, J. Candanedo; HQP: 2 PhD, 1 MASc, 1 PDF

Task 1: Data driven models are at the core of MPC. The accuracy of their prediction depends on numerous factors: 1. Data history: sensor type and location, duration, simulated or measured, steady state and/or transient; 2. Formulation: Type of model (grey, black etc) and number of inputs and outputs, target (temperature, energy, power); 3. Solution method: global, iterative; 4. Use: prediction horizon, timescale of models – this will depend on grid impact (e.g. peak-shaving, cold load pickup, integration of renewables) and building time constants.

This sub-project will focus on quantifying the impact of these factors on the prediction accuracy and robustness of the model of a system. Systems considered will include different types of zones: massive – with large areas of exposed concrete slabs, medium mass – some limited areas of exposed interior concrete and low mass – with mainly carpeted or wood floors and gypsum board as wall lining. In addition, zones with thermoactive slabs and phase change materials will be considered and major HVAC components including active thermal storage (e.g. ThermElect), floor heating/cooling, chilled beams.

Task 2: This task will develop a methodology for the rapid generation and calibration of grey-box models, targeting control applications in buildings. The development of control-oriented models can benefit significantly from the application of artificial intelligence (AI) techniques for the analysis of vast amounts of data (thermostat measurements, occupancy sensors, power meters, weather reports, and even smartphone data). It is possible today to leverage this information to quickly characterize the dynamic response of a building and thus develop a low-order model. However, although it is possible to develop purely data-driven models (which would be difficult to generalize and interpret), the goal is to use models relying on first principles and effective resistance and capacitance values. Artificial intelligence is a powerful tool to derive a set of simple model archetypes that can be easily adjusted during the operation of the building based on measured real-time data.

Project 1.2 Development of a methodology for design of model-predictive control (MPC) for buildings

Collaborators: A. Daoud and J. Bouchard (HQ), J. Candanedo; HQP: 1 PhD, 2 MASc, 1 PDF

A methodology for real-time MPC based on reduced order and hybrid models (including simpler energy signature models developed at LTE) will be developed. The methodology will incorporate the following: Models for loads due to occupants and actions such as opening/closing shading devices (results from IEA Annex 66 and project 2.1 will be integrated); Models for the effect of weather inputs; bounds on expected building response under extreme conditions (cold cloudy days, cold sunny day, hot sunny day, hot cloudy day); Development of techniques for optimizing prediction accuracy in terms of prediction horizon based on reduced order models; experiments with a test-room in the environmental chamber to test techniques for establishing appropriate prediction horizons; establishment of confidence intervals for the prediction; and, Development of models for thermal comfort penalty associated with zone temperature setpoint profiles. This is essential due to the fact that operation at the margins of the comfort zone needs to be considered, particularly in buildings with high levels of thermal mass during low occupancy periods or at night.

Project 1.3 Implementation and calibration of MPC, real time use of energy flexibility

Collaborators: A. Daoud (HQ), CanmetENERGY J. Candanedo; HQP: 1 PhD

Usually, prediction of loads and response due to weather variables and people will result in some error, causing the zone temperature, energy or power to deviate from the expected upcoming value in the next control time interval. This project will develop algorithms and techniques for recovery from such situations, as well as those due to diagnosed faults. The project will involve the following: Simulations with different scenarios/changes of loads from the expected values/profiles will be studied and rules/actions will be developed for recovery of indoor temperature in the comfort range within an acceptable time period; Use of concepts and models of thermal and electrical energy flexibility in implementing MPC; Limiting/worst case scenarios will be studied to develop robust MPC algorithms that recover from human behavior (such as leaving shades open on a weekend, resulting in high cooling loads on the following Monday). It can be studied at building level, based on data coming from smart meters (15 minutes meter reading) and/or at sub-system level with sub-metering data; Based on models developed, identification of the difference between weather impact, user behavior and faults due to sensors and other equipment; and, Use of statistical theory for the expected upcoming value error quantification to determine a reliability index on prediction.

Project 1.4 Optimal operation of existing buildings based on calibrated strategies and retrofits

Collaborators: H. Nouanègue, K. Lavigne (HQ); HQP: 1 PhD, 2 MASc, 1 PDF

To be efficient, predictive models representing the behavior of building dynamics and control of systems must be embedded in the global building control in order to plan appropriate commands of the HVAC systems and the integration of the building into the smart grid. The challenge lies in integration and the online calibration of those models. This project will establish the procedure to integrate artificial intelligence techniques (AI, e.g. neural networks) onto the control system of buildings to calibrate and operate the appropriate predictive model to achieve optimized interaction with the grid.

Links to research axes: Axis 1, Axis 2, and Axis 3.