State-space models for building control: how deep should you go?

Abstract

Power consumption in buildings show non-linear behaviors that linear models cannot capture whereas recurrent neural networks (RNNs) can. This ability makes RNNs attractive alternatives for the modelpredictive control (MPC) of buildings. However RNN models lack mathematical regularity which makes their use challenging in optimization problems. This work therefore systematically investigates whether using RNNs for building control provides net gains in an MPC framework. It compares the representation power and control performance of two architectures: a fully non-linear RNN architecture and a linear statespace model with non-linear regressor. The comparison covers five instances of each architecture over two months of simulated operation in identical conditions. The error on the one-hour forecast of temperature is 69% lower with the RNN model than with the linear one. In control the linear state-space model outperforms by 10% on the objective function, shows 2.8 times higher average temperature violations, and needs a third of the computation time the RNN model requires. This work therefore demonstrates that in their current form RNNs do improve accuracy but on balance well-designed linear state-space models with non-linear regressors are best in most cases of MPC.

Optimal ventilation strategies for balancing indoor air quality, comfort and energy use in educational buildings. the IAQ4EDU project.

“…Nevertheless, previous studies reported successful implementations of stochastic approaches based on reduced order models and measured data for different purposes such as electrical heating load shifting (Robillart et al, 2019), to model the indoor CO2 concentration (Macarulla et al, 2017), and to estimate buildings occupancy using CO2 concentration measurements (Wolf et al, 2019) or ventilation rates (Macarulla et al, 2018). Reduced order models are seen as a good alternative to complex building models where small accuracy improvements require much more detailed information and increased computational efforts and, with that, its functional applicability gets reduced for businessoriented applications (Schubnel et al, 2020). The IAQ4EDU project aims to take advantage of relatively simple and solid reduced models to identify those ventilations rates that guarantee proper indoor air quality levels in educational centres, being able to address a high number of scenarios with reduced computational resources.…”

A Multilevel Control Approach to Exploit Local Flexibility in Districts Evaluated under Real Conditions

“…The MPC's main function is to maintain indoor thermal comfort while providing a middle control layer that translates the flexibility requests from the DA into the appropriate control actions (i.e., temperature setpoints) for the building. Its objective is to minimise the power exchange between the grid and the building, while respecting the rooms' temperature constraints [22]. For the facilitation of the MPC objectives, the main two actions include using the building envelope to store thermal energy via overheating or overcooling by using heat pumps, chillers, etc., or deploying batteries, e.g., in-house batteries, to store electricity directly.…”

Development of control-oriented models for a building under regular heating, ventilation and air-conditioning operation - a comparative simulation study and an experimental validation

“…The traditional ON/OFF and proportional-integral-derivative (PID) controllers are widely used in many HVAC systems due to their ease of design and implementation. However, they are not as energy-efficient as predictive control methods such as model predictive control (MPC) (Dadiala et al, 2020;Schubnel et al, 2020). MPC has numerous advantages including the ability to perform anticipatory control rather than corrective control, to include a disturbance model for disturbance rejection, to handle constraints and uncertainties, and to deal with time-varying system dynamics as well as various operating conditions (Afram and Janabi-Sharifi, 2014).…”