The Bridge - Issue 3, 2020 - 19

THE FUTURE OF RENEWABLE ENERGY CONSUMPTION: Grid-Interactive Efficient Buildings

reset during seasonal transitions. Rule-based systems are
intuitive but do not necessarily lead to optimal operation
due to their inability to anticipate or predict how the
system will react. The ability to anticipate operational
requirements is critical to the success of the GEB vision.
Model predictive control (MPC) has emerged as the
most promising control algorithm to operate and plan
building thermal operations and demand flexibility over
a time horizon. It uses optimization over a sliding finite
time horizon to find control strategies that optimize for
selected criteria. It incorporates many inputs, including
forecasts and measured data. MPC is attractive due to
its conceptual simplicity and its ability to: (1) effectively
control systems with complex time-varying system
dynamics and multiple control inputs and constraints;
(2) handle multiple objectives; (3) handle disturbances
and uncertainty in predicted variables; (4) deal with
inaccurate equipment models and load forecasts;
and (5) optimize control decisions when faults are
present in the system, provided that models capture
their effects.
Some early experiments at building sites have shown
that MPC can yield significant energy savings (up to
17%-65%) compared to the performance of installed
control sequences [7, 8, 9]. Further, stochastic MPC
algorithms can formally deal with inaccurate equipment
models and load forecasts to find optimized control
decisions when the predicted variables have uncertainty
[10]. Distributed formulations of some MPC algorithms
provide increased implementation scalability [11, 12].
The majority of MPC research has concentrated on
implementation in large commercial buildings. The broad
adoption of MPC in these buildings faces challenges
like scalability, computational complexity, interpretability,
and the need for expensive custom models, which are
open areas of research. Development, training, and
calibration of models that are sufficiently accurate and
robust is another significant challenge, as is a lack of
acceptance by building operators. These challenges
lead to increased costs and long estimated periods for
return-on-investment despite the demonstrated energy
efficiency potential.
GEBs have taken advantage of the recent revolution in
data science. Recent research efforts explore machine
learning methods, such as reinforcement learning, to

Feature

learn energy efficiency control strategies. The methods
learn the relationship between control variables (i.e.,
zone temperature setpoints and airflow rates), other
variables (i.e., outdoor temperature, time of day, and
day of week), and energy cost to represent them in
structures such as neural networks [13]. Machinelearning-based building control may be competitive,
especially when scalable and accurate control-oriented
models are challenging to develop. Machine learning
for energy consumption prediction [14] and MPC
approximations for easier deployments [15] show
immense potential to solve many of the drawbacks of
traditional MPC systems.

Barriers to Technology Adoption
Despite its significant role in the GEB vision, advanced
building control systems face significant barriers to the
maximum adoption of algorithmic innovation. Setting
up building automation infrastructure is an expensive
process, heavily relying on multiple contractors with
varying expertise, time-consuming installation procedures
due to a lack of standardized data taxonomy, and
tailored modeling and control design for each building
system. Moreover, cost-benefit trade-offs for advanced
control strategies are difficult to assess due to existing
technical challenges, uncertainty in guaranteed savings
stemming from implementation and verification errors,
and uncertainty in model or training data accuracy
requirements and corresponding computational efforts
compared to projected cost savings from performance
improvements. Another significant obstacle for the
penetration of autonomous building controls is the
lack of standardized and interoperable hardware and
software that can interconnect across multiple vendors,
equipment types, and buildings.
Further, advanced control algorithms like MPC and
machine learning may provide nonintuitive solutions,
making it difficult for operators to interpret, tune, and
adjust according to their needs. Lack of customer and
operator education, interest, and awareness in new
product development and implementation is a significant
deployment barrier for new control technology, especially
in small and medium commercial buildings. Additionally,
comparison of performance features across products is
difficult without an established baseline, especially for
risk-averse owners and operators.

HKN.ORG

https://hkn.ieee.org/ https://hkn.ieee.org/

The Bridge - Issue 3, 2020

Table of Contents for the Digital Edition of The Bridge - Issue 3, 2020