ASHRAE Journal - September 2024 - 53

TECHNICAL FEATURE
structure, context and materialization- to be able to
use them properly. Namely, control algorithms rely on
specifi c inputs from specifi c devices, representing a
specifi c situation in a specifi c building, and so these
algorithms need to relate to these input data streams
that represent certain features in the building (e.g.,
temperature, humidity).
This is a critical feature that makes the system either
more tailor-made (e.g., tuned to personal comfort)
or more generic (e.g., for overall indoor air quality),
and so this has a considerable impact on the effi ciency
and effectiveness of control systems during design,
engineering and implementation of these systems.
To be able to relate to these input data, control systems
need to rely on metadata- " data that describes other
data in order to help one understand or use it. " 3 These
metadata, which often come in the form of tags and
annotations, represent the context of input data streams
and the building context itself; as such, they allow for
making better control systems for smart buildings.
In many cases, such metadata end up being
hardcoded; in other cases, more code-independent
schemas are being used, often referred to as metadata
schemas. Some more recent metadata schemas used for
smart buildings are the Smart Applications REFerence
ontology (SAREF),4 Haystack5 and Brick.6 Yet, several
other metadata schemas exist, also in very different
formats. In fact, these metadata tend to be described
in a multitude of very different ways and are present in
multiple places in the control systems of a building, also
hardcoded.7
Standardization of such metadata and metadata
schemas would be very helpful and a big advance for this
industry, yet to be able to standardize it is important to
know where all these metadata are located in a building
system and what impacts they have on a system.
Data in Control Systems
Figure 1 shows components typically in place in a system
with multiple devices and sensors as well as a control
system to operate the actuation of these components.
In this case, the example shows a controllable lighting
device, an HVAC system and a dynamic façade system
that function differently depending on outdoor
conditions, utility demand signals and user input-plus
the status of controllable components.
In this example, a device is set up centrally that
FIGURE 1 An example control system that combines multiple input data streams
(based on Konstantoglou and Tsangrassoulis).8
Controllable Lighting
Weather
Sensor
Sensor
CPU
Controllable
Blinds
User
Smart Control
Algorithm
Energy
Supply
organizes the control: the smart control algorithm
in the central processing unit (CPU). Such a control
algorithm is often implemented in a general purpose
programming language (e.g., Python) and is a relatively
straightforward set of commands and functions that
takes in incoming values (real-time data streams),
performs simple or more complex computations
and produces output signals to devices on the edge
that are then actuated and change their status and
behavior accordingly. Control can happen in a few
ways (Figure 2). Besides manual or manual motorized
actuation, automated procedures can be simple rulebased,
integrated rule-based (e.g., decision trees) or
model-based.9
The two rule-based methods (4.1 and 4.2 in Figure 2) are
relatively straightforward and focus mainly on specifi c
physical quantities (e.g., sun-shading control based on
irradiance values), while model-based control is much
more integrated and typically includes optimization
algorithms on one or multiple domains.9 As a result,
model-based control typically aligns better with
performance-based systems,9 as they rely inherently
on a model that evaluates incoming data against a
model (physical, reduced order or black-box). These
models range from simple to more complex models.
The complex models are often increasingly data-driven
and consist of grey-box or black-box models that are
built using historical data and are executed on live data.
Neural network (NN) models, simulation models and
prediction models (grey-box and black-box models) are
built using historical data, and they are applied to newly
incoming data streams in the building. This allows
model-predictive control (MPC) in HVAC systems in a
fi ne-grained level of detail.9
S E PTEM B E R 2 0 2 4 ashrae.org ASHRAE JOURNAL
53
HVAC
System
Sensor
http://www.ashrae.org
ASHRAE Journal - September 2024

Table of Contents for the Digital Edition of ASHRAE Journal - September 2024
