2.1.1. IBs of Responsive Architecture

Meyboom et al. ^[40] maintain that Le Corbusier’s design for the 1958 Philips Pavilion, (“Poème Electronique”), which integrated music and visual display with the building, was an early example of an IB in architecture. Michael Webb’s 1960 Magic Carpet presented the IBs of a dynamic fluid and air jet environment supporting a body in space. Jean Nouvel’s responsive screens in the 1988 Institut du Monde Arabe in Paris were also intended to use light-sensitive diaphragm devices. IBs and responsive facades of these types remain a topic of study in recent years ^[40]. For example, interactive kinetic media facades now use Delta robot kinematics (e.g., piston motion, radial motion) and LED effects ^[37]. Adaptive solar facades also use soft pneumatic actuators that bend under applied pressure ^[44]. These mechatronic behaviors are responses to the changing needs of the inhabitants as well as changes in the environment ^[11][40]. Responsively, the behavioral systems transform architectural form, shape, colour or character through actuators ^[38].

Climatically responsive IBs are a common category in the research. They have been widely used in shading systems in contemporary architecture ^{[30][32][33][34][35][36][39][40][42][43][45]}. In façade design, high quality luminous environments using this sort of IB not only satisfy human visual needs, but also develop a sustainable and climatically responsive architecture ^[39]. There are three broad types of solar shading devices mentioned in this research: membranes (awnings, curtains), shutters (folding/sliding panels) and kinetic devices (louvres and other kinetic façades) ^[42]. In addition, climate-responsive behavior supports improved thermal performance and comfort in a building ^[39]. A customized dynamic architectural façade or CABS ^[30][34] can respond to thermal, optical, airflow and electrical behaviors. Micro adaptation of CABS is typically limited to changes in thermophysical, opaque optical properties and the exchange of energy, while its macro adaptation can involve subsystems of the building shell itself as well as movement of the entire façade or the building as a whole ^[34].

In recent years adaptive origami-based modular structures ^[32][33][35] have been proposed as a means of enabling architectural responsiveness to light, thermal and acoustic conditions. The origami-based behavior is typically a sensing and actuation system much like the other climate-responsive building skins ^[33]. In contrast, a kinetic solar skin using a thermo-mechanical SMA actuator can exhibit similar self-organizing behavior to origami folding patterns ^[32][35]. In this case, the responsiveness can be understood as a material-based behavior that enables a lightweight, motorless and silently operable building system ^[35]. Such climate-responsive IBs of adaptive façades have implications for architectural customization, sustainability and aesthetics ^[33] as well as building performance ^[43]. Many of these interactive building skins rely on mechatronic behaviors based on kinematics, while some solar shading devices use smart materials instead (e.g., property-changing smart materials, energy-exchanging smart materials) ^[42].

Research in the dataset for responsive architecture also confirms that the optimization of materials not only impacts on architectural robustness and performance, but also determines its responsive behaviors. Thus, many articles describe the adaptation or development of appropriate materials ^{[31][33][35][36][42][46][47][48][49][50]}. Smart materials have been developed for material interaction integrating sensing and actuation into the fabric of architecture ^[42][48]. In contrast, biomimetic behaviors responding to environmental changes use, for example, hygromorphic materials ^[31][47][49] or wood-moisture relations arising from the shrinkage or swelling of wood ^[49]. This naturally responsive mechanism is not only predictable and reproducible but also reversible ^[47]. Biomimetic behaviors can be further developed using responsive programmable biofunctionality and microscale patterning using a fluidic system (glaze chemistry, texturing and geometry) and light-responsive ceramic bio-tiles (fluorescent or colour-changing) ^[50]. Such examples enable metabolically independent movement or materially-embedded responsiveness in the smart environment.

The IBs of responsive architecture can accommodate both spatial adjustments and qualities. Spatial adjustments involve linear displacement of a partition, increasing available surface area, while spatial qualities are related to geometry, colour and lighting, acoustics and ventilation ^[23]. Such IBs can arise from bio-inspired elastic kinetic shading systems, soft textile shadings (changes of bending curvature), changes in length of the membrane strip, a self-supporting shell structure or the intrinsic capacity of wood (wood cones) ^[36]. Unexpectedly, although natural interactions and simulated behaviors can be a key behavior in responsive architecture ^[8], they were rarely identified in the dataset for responsive architecture. Natural interactions, like embodied interactions, are generally developed using motion tracking, gesture recognition, emotional detection, facial expression identification and speech processing ^[41]. This embodied IB can support healthcare-related functions (e.g., early detection, diagnosis, monitoring and documentation, prevention, treatment, alleviation of disease, rehabilitation, wellness management and motivation) ^[41].

2.1.2. CBs of Responsive Architecture

CBs of responsive architecture typically correspond with IBs in an architectural system because a certain behavior is triggered in response to a particular situation or stimulus. For example, to appropriately manage the climate-responsive behaviors of a building façade, preset programming is typically used to track the sun movement ^[40][44]. Furthermore, sensing behaviors capture temperature and humidity, sun tracking and daylight harvesting, light levels, movement and local environmental conditions ^[36][40]. Origami-based behavior is also based on these environmental sensing behaviors using a network of micro-sensors, detecting, for example, an a priori defined noise threshold (interior) and optimal light conditions (exterior) ^[33]. These sensor data and embedded computational elements then regulate the quality of light ^[11]. Temperatures and optimization criteria—including energy-related indicators and fluid-dynamic analysis—can also be used for this CB ^[30]. Energy-related indicators include glare probability, daylight and illuminance uniformity, and factors of external view. In addition to sensor-integrated automation, climate-responsive behaviors can be triggered by user interaction with an app-based remote control ^[35][44].

In general, there are two types of automated control behaviors in the research in this dataset: intrinsic and extrinsic controls. Extrinsic controls deal with distributed CBs via embedded computation in local processors, and/or centralized CBs triggered by a supervisory control unit to fulfil global target values ^[34]. In contrast, intrinsic controls tend to be self-adjusting, or automatically triggered by environmental stimuli (e.g., temperature, relative humidity, precipitation, wind speed and direction, solar radiation, cloud cover or CO₂-level) ^[34]. In addition to environmental and infrastructure sensing, wearable sensors can be implemented for natural interactions based on real-time processing and data transmission via wireless body communication networks ^[41]. Park ^[37] furthermore identified various aspects of CBs, including user engagement (active and passive), number of users, input devices (e.g., sensors, Radio frequency identification (RFID)), local and global network, expressive, responsive (linear) and interactive (single loop) intelligence. A special case is “sensponsiveness”, where ambient intelligence imbues space with cognitive capacity ^[23].

The automation and control aspects of CBs (digital electronics) can have a significant impact on the performance of a building. They include control strategies and scenarios (e.g., open-loop or closed-loop systems, single- or multi-variable systems), controlling technologies (e.g., self-sensing and self-actuating technology) and controlling algorithms ^[43]. Control strategies can further be categorized into single variable-man, multivariable-man, multivariable automatic and multivariable heuristic controls ^[38]. Intelligent control behaviors address these control strategies and mechanisms with feedback. Control mechanisms and feedback simply consider both open-loop and closed-loop systems. An open-loop system responds to a signal in a predefined way, while a closed-loop system uses a feedback system ^[38]. In this way, advanced lighting controls can moderate levels of natural and artificial illumination, using digital dimming ballasts and programmed and user operable controls ^[39]. In summary, a control system or inference system translates sensor signals from user interaction or climate change into actuation commands. Actuators then produce reactions in the smart environment, converting energy into movement. In this process, the interactive or collective stimuli are captured by sensors and elaborated by the computation of the control system ^[36].

In contrast, as discussed in the previous section, smart materials have self-powering and self-actuating systems. For example, hydrogel biomaterials can innately respond to pH, glucose, oxidants, antigens, enzymes, ligands, temperature, pressure and light ^[50]. “Hygroexpansion” is also based on moisture-induced opening and closing ^[31]. Thus, CBs are rarely exhibited in the sort of smart environments enabled through the use of smart materials. Smart materials can also be pre-programed ^[31] or use shape-changing rules integrating material transformation into shape computation ^[49]. Thus, this self-actuating behavior is like an SMA’s self-organizing behavior ^[32][35], which is a research topic that has not yet been explored in the literature.

2.2.1. IBs of Kinetic Architecture

The phrase “kinetic architecture” typically refers to a mechanical and movable structure or organism, as architecture is often described metaphorically in biological terms ^[10]. Developments in robotic technologies as well as digital design processes have allowed architecture to become more flexible and adaptable in response to changing needs ^[51]. Ramzy and Fayed ^[24] provides a historical, evolutionary overview of kinetic architecture, from early primitive kinetic systems (e.g., pivoting, sliding openings) to advanced kinetic systems using AI. Twentieth-century kinetic systems already adopted a variety of mechanical, lightweight and flexible structural systems to create movable designs, while recent kinetic systems developed have employed computerized systems to create intelligent architecture ^[24]. Holden ^[52] provides a unique way of looking at this dataset, through an analysis of Jean Tinguely’s kinetic movements. Tinguely’s interactive constructions and artworks in the 1960s provided a different model of the relationship between art and architecture. Tinguely’s kinetic structures, or “meta-mechanicals”, can be regarded as participatory art, presenting an operative concept in both art and architecture. His large-scale kinetic (also interactive and collaborative) projects provided new types of platforms for cultural activity in the urban environment. For example, Dylaby using scaffold-like frames (1961) represented the unpredictable movement of the audience as well as the kinetic movement of architectural elements ^[52]. Tiluzi (1967) also exhibited the “regulated movement of the Ferris wheel” and the “erratic movement of a long circuitous slide or ramp” ^[52]. Drawing from this work, it is possible to identify kinetic systems as those which can automatically fold, slide and expand in both size and shape, exhibiting various IBs such as intelligent circulation (vertical and horizontal), environmental response (e.g., revolving buildings and responsive roofs) and flexibility for inner spaces (e.g., movable partitions) ^[24].

Despite such definitions drawn from art, kinetic architecture remains a contested concept in architecture, and there are multiple, sometimes conflicting terms and typologies in use ^[51]. One view is that kinetic systems embedding computational intelligence in architecture to enable it to be “adaptable, collapsible, deployable, enabling, evolutionary, flexible, intelligent, kinetic, mobile, performance-based, reconfigurable, responsive, revolving, smart, transformable, and transportable” ^[51] (p. 132). “Adaptable” structures (e.g., movable-wall systems) are easily altered or modified to meet different social functions, while “deployable” structures are capable of automatic configuration changes. “Intelligent” structures learn and respond to the information collected from the exterior or interior environments and “performance-based” structures use digital technologies to support the environment, users and society ^[51]. That is, each variation in terminology represents a specific IB of kinetic architecture.

The kinetic mechanisms identified in this literature also include two types: spatial-real movement and non-spatial-material deformation ^[51]. The former presents basic movement—folding, sliding, rolling, expanding and transforming—by changing the axis, strength and direction of kinetic elements. The latter uses smart materials driven by their molecules’ ability to change form, function or appearance ^[51]. In contrast, Ramzy and Fayed ^[24] argue that kinetic systems may be classified into four types: (i) skin-unit systems, (ii) retractable elements, (iii) revolving buildings and (iv) biomechanical systems. Skin-unit systems include responsive and interactive facades, flare skin and movable louvers. These systems typically support climate-responsive behaviors, although they can also respond to other information, such as pedestrian movement. The mechatronic behaviors using deployable kinetic structures are clearly seen in “retractable elements”, which fold or expand creating entire architectural element (roofs, walls, floors, etc.). Whilst “revolving buildings” respond to wind-power or solar energy, “biomechanical systems” using embedded or dynamic kinetic structures adjust themselves in respond to inner or outer forces ^[24]. The last category, biomimetic behavior, is mostly reliant on smart materials as described in the previous section.

The most dominant IB in the kinetic architecture literature is a structural adaptive behavior using folding mechanisms ^{[53][54][55][56][57][58][59]}. For example, linkage mechanisms for kinetic architecture include Watt-I linkage as a one degree-of-freedom (1-DOF) mechanism ^[53], spherical linkages ^[55][59] and a two degree-of-freedom (2-DOF) 8R (revolute joint) linkage ^[56]. The Watt-I “finger linkage” used in robotic, anthropomorphic fingers has been adopted for a convertible stadium roof structure, providing a wide range of structure flexibility and shading options ^[53]. Spherical linkages with Miura-ori folding mechanisms have been used to create deployable surfaces ^[55][59]. Their translational motion behavior uses scissor-like structures that develop the target curvature by changing the length of the bar and creating foldable assemblies ^[55]. Lastly, from the Bennett linkage using a 4R spatial linkage, Korkmaz et al. ^[56] suggest a 2-DOF 8R linkage for transformable hyperbolic paraboloid (“hypar”) structures. The 2-DOF system allows various configurational structures and wider form flexibility, responding to dynamic and constantly changing activities. This kinetic, structural adaptive behavior is most often applied to roof structures ^{[53][54][56][57]}.

Research has also proposed the development of lightweight roof oculus structures to integrate two cooling strategies for desert climates: evaporative cooling (day) and radiative cooling (night) ^[54]. The prototype for this roof oculus uses a slab as a thermal mass, storing coolness which influences geometric kinetics (constricting or releasing the opening). In contrast, another example proposes a pliable structure based on curved-line folding and origami-based behavior (the combination of folding and bending paper) ^[58]. The pliable structure can present origami-like self-organizing behavior ^[32][35], while curved-line folding creates a more complex 3D shape through a curved crease as well as an elegant folding motion. This example uses both material-dependent behaviors and Finite Element Analysis (FEA) to enable its kinetic and structural behavior ^[58]. Developable surfaces are also explored in the literature for their capacity to exhibit such behaviors ^[55][58][59].

The last IB of kinetic architecture develops ASF using a kinetic PV shading system ^[60]. This type of ASF is described in the “responsive architecture” dataset as mechatronic behavior ^[39][44]. In contrast, Jayathissa et al. ^[60] address performative design environments that enable a solar radiation model, PV electricity production, a building energy model, a daylighting model and their optimization. In summary, past research in kinetic architecture typically deals with various aspects of kinetic hardware and its structural kinetic behaviors.

2.2.2. CBs of Kinetic Architecture

Kinetic systems in this dataset are typically associated with sensing and actuation systems that are largely reliant on environmental sensing behaviors. One example, a kinetic roof structure, uses “DHT-22 temperature and humidity” sensors and a cable-driven actuator network ^[54]. However, the second dataset barely describes any other CBs, and mostly just highlights the physical changes or movements of architectural components. One exception ^[61], like ^[43] in the previous dataset, describes the use of control scenarios. It argues that metamorphic architecture can be developed through scenario-based design, addressing problems, activity, information, interaction and usability scenarios ^[61]. “Activity”, which is the transformation of a current setting into a new configuration, and “interaction” scenarios are closely related to IBs, while “information” and “usability” scenarios support the development of effective CBs.

2.3.1. IBs of Adaptive Architecture

Adaptive architecture “has the ability to alter its physical properties (form, shape, colour, texture, acoustic, porosity, etc.) in a predefined/programmed/designed way to adapt to changing external and internal environmental stimuli (temperature, relative humidity, precipitation, wind, sound, solar radiation, CO₂-level, etc.), user activities and needs, and social contexts” ^[62] (p. 557). This definition emphasizes the synergies of overlaps with the broader concept of responsive architecture. The definition also stresses the importance of pre-programed elements which facilitate “adaptivity” or “adaptiveness” in architecture. Such elements often include smart materials, and it is not surprising that one of the papers in the third dataset, Abdullah and Al-Alwan ^[63], contains a review of past research into smart material systems that can create adaptive architecture. Their survey classifies smart materials into two types (property change and energy exchange) and smart material systems into three types (passive, active and hybrid). Since a smart material has multiple functions, including sensing changes that trigger actuation, its classification considers the way it responds to stimuli. Furthermore, they argue that combining different types of smart material systems can produce a higher level of adaptivity in architecture. Thus, biomimetic behaviors of smart materials play an important role in adaptive architecture, using smart materials’ hygroscopic behaviors ^[63][64][65] and even plants’ biological adaptation ^[66].

The hygroscopic behavior of wood, in response to changes of relative humidity (e.g., false ceiling opening), is one example of an adaptive hygrothermal comfort system ^[64]. This sort of IB, which was discussed in the previous section, has additional pertinent qualities when considered under the heading of adaptive architecture. Both structures of plywood and “unplywood” using active and passive layers are examined in the literature ^[64]. The bending reaction of wood bilayers is also developed in a multi-element wood-GFRP (glass fibre reinforced polymer) bilayer ^[65]. The wood and wood-hybrid bilayers can accommodate controlled and reversible shape changes in reaction to relative humidity. Furthermore, the hygroscopic behavior, the properties of curvature (e.g., specific sizes, shapes and aspect ratios), can be controlled or designed to achieve a particular outcome ^[65]. A different approach is to use organisms’ survival, evolutionary or natural behavior in a building façade system ^[66]. For example, plants in a façade can react to light, temperature or water changes and support building performance (energy saving) as well as occupants’ comfort levels at the macroscopic and microscopic scales. Like some other smart materials, there is no clear CB in the plant’s biological adaptation, because an organism’s shape, size, pattern or structure naturally depends on its surroundings.

As it was for kinetic architecture, the dominant IB of adaptive architecture is a structural adaptive behavior ^{[67][68][69][70]}. The structural adaptive behavior relies on a controllable dynamic system consisting of sensor-actuators, structural elements and skins. It then develops a transformation from a load-bearing behavior to a dynamic one ^[71]. These dynamic interactions between occupants and buildings enable, for example, the reduction of energy consumption and emission rates as well as occupants’ comfort ^[71]. As for the adaptive building skin, the load distribution of structural elements is regarded as an IB of intelligent machines ^[67], supporting real-time activation as well as self-learning behaviors. Building components identified in past research include façade elements, canopies and other structural features, where load distribution parameters and spatial parameters are used to adapt to unpredictable forces ^[67].

In addition, an interactive and optimized behavior can be applied to deployable building structures using tensegrity and scissor-like systems ^[67][68]. In such systems a lightweight linkage structure provides a generic 1-DOF, an effective crank-slider mechanism and a n-bar linkage with direct or cable-driven actuation ^[67][68][69]. Such a system is also flexible, expandable and controllable through modularity and actuation. In this way, structural elements become reconfigurable, and structures become self-erectable ^[67][68][69]. Structural reconfigurations involve morphological changes, manipulation and locomotion ^[69]. When considering all of these features, Phocas et al. ^[68] and Christoforou et al. ^[69] identified multiple IBs of reconfigurable architecture, such as optimizing the performance of a PV roof; optimizing distribution of structural and minimizing aerodynamic loads; optimizing occupants’ comfort by adjusting ventilation and lighting; improving space utilization; harvesting sun, wind and rain water; removing snow from roofs and producing unique aesthetic effects. Pruitt et al. ^[72] also describe historical ideas about the comfort-ensuring behaviors of adaptive architecture (e.g., climate-responsive façade design and mechanical ventilation systems).

2.3.2. CBs of Adaptive Architecture

Intelligent CBs, such as self-learning algorithms or artificial neural networks (ANNs), created by a genetic algorithm develop various structural adaptive behaviors. Active control systems also use a database (knowledge) of pre-calculated equilibrium solutions. In this way, structural adaptive behaviors not only enable real-time measurement and optimization of the environment, but also improve their adaptation processes over time through self-learning behaviors ^[67]. In addition, an irregular self-bearing structural system, such as a tensegrity-membrane structure, can be used for wind-adaptive architecture, changing the aerodynamics of a building ^[70]. CFD simulations can be used to model its potential CB.

In another example, the biofeedback-driven system “ExoBuilding” provides the immersive effect of adaptive architecture (specific interactive effects), using recent technologies such as pervasive computing and a tent-like fabric structure ^[73]. The system presents biofeedback in an immersive, evolutionary fashion, in terms of embodied behavior. Thus, its CBs address the capturing of various types of personal data (e.g., location information, activity, social networking data, reactivity around one specific type of personal data and physiological data), controlling its central flexible spine via servomotors ^[73]. In addition, AI (e.g., ANN, multi-agent system (MAS) and EM algorithm) can be used for lighting control in smart cities ^[74]. Its goal is efficient energy management, detecting foot traffic patterns, managing ANN to predict consumption from light intensities and estimating energy consumption. Thus, the intelligent street lighting system collects information such as pedestrian and traffic flow and weather data ^[74]. Such methods are also suggested for climate-adapted architecture for energy saving ^[75], which is closely related to the theme in the following section.

2.4.1. IBs of Intelligent Buildings

The most dominant IB found in the intelligent building literature is concerned with either “energy efficiency behavior” ^{[81][82][83][84][85][86][87][88][89]} or “energy saving behavior” ^{[77][79][80][82][83][89][90][91][92][93][94][95][96][97]}. Ding et al. ^[77], for example, identify research trends in building energy saving using a text mining methodology. In their survey, heating, ventilation and air-conditioning (HVAC) systems, energy technologies and lighting systems have been the major topic of 1600 articles on energy saving from 1973 to 2016. Thus, HVAC and lighting behaviors should be a fundamental IB of intelligent buildings. The most common topics of recent articles (2010–2016) are green building envelopes, building retrofitting, system operations and building information. These reflect recent academic interests in the integration of a solar energy system or a life-cycle management system using building information modelling (BIM) ^[77]. Nguyen and Aiello ^[80] offer an alternative definition, energy intelligent buildings, which refers to “buildings equipped with technology that allows monitoring of their occupants and/or facilities designed to automate and optimize control of appliances” (p. 247).

In contrast, there are two dominant comfort-ensuring IBs: “thermal comfort behavior” ^{[79][81][84][85][94][95][96][98][99][100][101]} and “visual comfort behavior” ^{[79][83][94][100]}. In comparison with many exterior-oriented behaviors identified in the previous sections of this paper, research about “intelligent building” often examines multiple indoor comfort-ensuring IBs and CBs including indoor daylight ^[81], environmental quality (IEQ) ^[82][102], thermal comfort ^[98][99] and air quality ^[94][103], navigation ^[104], positioning ^[105] and even indoor electrical IoT ^[105]. A comprehensive review by Mofidi and Akbari ^[76] of intelligent buildings categorizes six EMS topics: (i) occupant comfort conditions, (ii) occupant productivity, (iii) building control, (iv) computational optimization, (v) occupant behavior modelling and (vi) environmental monitoring and analysis ^[76]. As for the first two comfort-ensuring IBs, an intelligent EMS not only addresses thermal comfort, lighting and daylighting, visual comfort and indoor air quality (IAQ), but also supports the occupants’ productivity and well-being. The Leadership in Energy and Environmental Design (LEED) certification program, the WELL building standard and the Building Research Establishment Environmental Assessment Method (BREEAM) can be used for the productivity standards and guidelines ^[76]. Interestingly, whilst the second dataset, kinetic architecture, was largely focused on IBs, articles in the last dataset are more commonly concerned with CBs than IBs. This characteristic is to be expected, because the term, “kinetic” strongly indicates a physical movement in a product, whereas the term, “intelligent” is related to “thinking” as the process in the operations of smart environments.

2.4.2. CBs of Intelligent Buildings

First of all, for an intelligent building to achieve energy savings it typically controls lighting, HVAC and “plug loads” (energy used by appliances), depending on occupant presence and behavior ^[80]. For example, Aftab et al. ^[93] present an occupancy-predictive HVAC control system using embedded system technologies (e.g., real-time occupancy recognition, dynamic analysis and prediction of occupancy patterns and a model of predictive control). The real-time occupancy recognition is achieved using video-processing and machine learning (ML) techniques, while the HVAC system is supported by a real-time building thermal response simulation using EnergyPlus ^[93]. A recent cloud-based system for energy information management also monitors, analyses and controls the energy use of a building. The cloud forecasting system uses a hybrid AI model—seasonal autoregressive integrated moving average (SARIMA) and metaheuristic firefly algorithm-based least squares support vector regression (MetaFA-LSSVR)—to characterize energy usage patterns, and to predict energy demand in real time ^[90]. To improve energy efficiency and thermal comfort, a model predictive control (MPC) design has adopted a tuning methodology that takes account of process disturbances, temporal parameters and weights on the objective function ^[96]. Energy-optimizing CBs are fundamentally involving such intelligent control systems that consist of numerous sensors and computational intelligence.

A BECM system evolves with intelligent control CBs based on AI ^[106]. Thus, multiple articles in this last dataset examine AI ^{[90][98][106][107][108]} and ML ^{[83][90][106][109]}. Panchalingam and Chan ^[106] conduct a literature review of research on AI technologies for smart buildings, focusing on nine topics: ML, natural language processing, deep learning, pattern recognition, machine vision, expert systems, ANN, fuzzy logic and genetic algorithms. Interestingly, energy saving behaviors in their survey largely adopt ML, supported by ANN, fuzzy logic and genetic algorithms. These computing systems also support structural adaptive behaviors (described in the previous section).

Multiple expert systems for reducing energy consumption have been developed at various scales. For example, an ANN control model for optimized distribution and heat trading effects can be used for responding to occupant characteristics, optimizing supply air condition and maximizing energy cost savings ^[99]. A two-layer ANN is also used for inferring occupancy counts from existing ICT system data ^[110]. Non-linear models based on fuzzy logic and ANN have been applied to predict electricity consumption and develop energy efficiency strategies ^[86]. As such, the intelligent building can be integrated using a micro-grid based on renewable energy resources (RERs) and energy cost coefficients (ECC) ^[111]. An energy-efficient outdoor lighting control system is also based on an expert system that uses knowledge-based rules for real-time control and monitoring function ^[107]. In this context, Aduda et al. ^[112] suggest the creation of an “energy and comfort active building” using a MAS, which interacts with electrical smart grids. Its EMS involves four levels of informational flows (communications): use level, building management level, agents/agent platform-utility grid and utility grid side communications ^[112]. A building EMS can also use power line communication ^[97]. Importantly, energy-optimizing behaviors are based on real-time occupancy information about preferences, patterns or use and activities ^[80]. Again, in order to recognize occupants’ activities, energy intelligent buildings adopt various technologies and approaches: logical inference from sensor data, ANN, fuzzy-logic-based incremental synchronous learning (ISL) systems, Bayesian Networks (BNs) and multivariate Gaussian and agent-based models ^[80].

In addition to these AI technologies, a cost-benefit evaluation addressing life cycle net present value (NPV) can be applied to support energy consumption of building intelligence systems ^[78]. Specifically, to develop nearly zero energy buildings (NZEBs), energy consumption standards can adopt energy-efficient measures based on efficient thermal insulation systems, high-performance window systems (heat transfer coefficient, solar heat gain coefficient and window-to-wall ratio), good airtightness and fresh air heat recovery systems. Furthermore, NZEBs use various renewable energy technologies such as solar thermal systems, solar PV systems, ground source heat pumps (GSHP), air source heat pumps (ASHP) and wind power systems ^[113].

Intelligent controls with smart sensing and self-learning behaviors have been used for both energy and comfort. However, there are some interesting characteristics of thermal comfort-ensuring behaviors. For example, Cheng et al. ^[84] address human thermal comfort measurement using a contactless measurement algorithm and Peng et al. ^[91] developed a learning-based (using ANN) temperature preference control (LTPC) as an occupant-centric climate control system. Yoganathan et al. ^[88] introduce an optimal sensor placement strategy using clustering algorithms that optimize the number and location of sensing points. A recommender system using distributed sensing, context-awareness and ML can also be applied for personalized visual comfort ^[83], while a decentralized stochastic control using a Markov decision process can be used for comfort-ensuring behaviors ^[100]. An indoor localization system based on ANN and particle filters is also proposed for customized comfort service ^[105]. In addition to thermal and visual comfort, acoustic comfort has also been considered in smart environments ^[94][97].

In addition, there are three comfort control strategies—conventional methods, intelligent control and multi-agent-based modelling (MABM) techniques—which enable an intelligent BECM system ^[76]. To develop comfort-ensuring CBs in smart environments, intelligent building control systems have adopted computational optimized operational methods: occupant behavior modelling, and data collection, analysis and feedback. Modelling occupant behavior involves deterministic, stochastic and agent-based behavioral modelling techniques, while computational optimization is achieved by single-objective optimization (SOOP), multi-objective optimization (MOOP) and classical methods such as the weighted sum method (WSM) and evolutionary algorithms ^[76]. These AI methods and techniques are used to simultaneously optimize energy and comfort-related behaviors in buildings, supported by the self-learning behaviors discussed in the previous sections. This intelligent aspect of smart environments also links to adaptive comfort behaviors including psychological and physiological adaptation as well as behavioral adjustment ^[76].

A smart HVAC system should be a long-term research topic for smart environments, but recent studies adopt intelligent HVAC controls using real-time occupancy recognition ^[80][82][93], an MPC ^[96][109], a MOOP method ^[94], a fuzzy supervised neural-control (FSNC) ^[103] and hybrid learning ^[101]. These indoor comfort-ensuring CBs also require sensing systems. For example, to determine an occupant’s thermal comfort preference, temperature and humidity, velocity and heart rate and skin temperature sensors can be used in the building system. In contrast, individual visual comfort can be determined using photometric and mobile pupilometer sensors ^[79].

The final observation of this last dataset involves safety, design and maintenance behaviors in smart environments ^[106]. Safety research is concerned with reducing the risk of harm for occupants, although it can consider crowd safety ^[108], privacy and security issues ^[79] and health and safety requirements ^[114]. Design (e.g., architectural, electrical, mechanical or layout design) can be improved by the integration of automation and control systems in a building ^[106]. Thus, from a design perspective, smart homes continue to be a research topic ^{[90][115][116]} along with façade design ^{[114][117][118][119][120]}. Automated adaptive façade functions ^[117] and occupant–facade interaction ^[118] not only present energy and comfort related behaviors, but also impact on building design. Furthermore, solar PV systems ^[113], smart materials ^[114] and phase change materials ^[120] can be investigated for intelligent building design. Recently, intelligent building design is linking to its life-cycle maintenance, significantly supported by BIM ^{[77][104][121]}. BIM also supports indoor navigation ^[104] and intelligent disaster prevention ^[121]. In addition, the management and maintenance of an intelligent building can adopt cognitive facility management ^[122], real-time digitalization ^[82] and even autonomous robots ^[123].