SAT based Enforcement of Domotic Effects in Smart Environments

The emergence of economically viable and efficient sensor technology provided impetus to the development of smart devices (or appliances). Modern smart environments are equipped with a multitude of smart devices and sensors, aimed at delivering
of 15
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
  ORIGINAL PAPER SAT based enforcement of domotic effects in smart environments Fulvio Corno  • Faisal Razzak Received: 7 December 2012/Accepted: 10 April 2013   Springer-Verlag Berlin Heidelberg 2013 Abstract  The emergence of economically viable andefficient sensor technology provided impetus to the devel-opment of smart devices (or appliances). Modern smartenvironmentsareequippedwithamultitudeofsmartdevicesand sensors, aimed at delivering intelligent services to theusers of smart environments. The presence of these diversesmart devices has raised a major problem of managingenvironments. A rising solution to the problem is the mod-eling of user goals and intentions, and then interacting withthe environments usinguser defined goals. ‘domoticeffects’is a user goal modeling framework, which provides ambientintelligence(AmI)designersandintegratorswithanabstractlayer that enables the definition of generic goals in a smartenvironment, in a declarative way, which can be used todesign and develop intelligent applications. The high-levelnature of domotic effects also allows the residents to pro-gram their personal space as they see fit: they can definedifferent achievement criteria for a particular generic goal,e.g., by defining a combination of devices having someparticularstates,byusingdomain-specificcustomoperators.This paper describes an approach for the automaticenforcement of domotic effects in case of the Booleanapplication domain, suitable for intelligent monitoring andcontrol in domotic environments. Effect enforcement is theability to determine device configurations that can achieve aset of generic goals (domotic effects). The paper also pre-sents an architecture to implement the enforcement of Boolean domotic effects, and results obtained from carriedout experiments prove the feasibility of the proposedapproach and highlight the responsiveness of the imple-mented effect enforcement architecture. Keywords  Ambient intelligence    Domotic effects   Domotic effect enforcement    SAT based enforcement   High level modeling    DogEffects ontology 1 Introduction The last decade saw the emergence of economically viableand efficient sensor technology which can be integratedwith appliances, enabling them to sense different parame-ters of their respective environments, i.e., temperature,luminosity, pressure, etc. It helped realizing the vision of smart environments (Weiser 1995) by developing hetero-geneous dynamic ensembles: groups of co-located devicesof different types which evolve over time (Bader andDyrba 2011). Such environments promise to offer addi-tional intelligent capabilities that go beyond the integratedand remote control of appliances present in the environ-ment. But the presence of diverse devices and the associatedcomplexity has given rise to a major problem in the pastyears, i.e., the problem of providing users with the ability tocontrol and manage their respective environments.State of the art revolves around the issues related tocommunication protocols and technologies (Dey et al.1999; Rashidi and Cook  2009; Kawsar et al. 2008). Many approaches are furthermore based on abstract modeling of smart devices, resorting to some knowledge representation Electronic supplementary material  The online version of thisarticle (doi:10.1007/s12652-013-0183-x) contains supplementarymaterial, which is available to authorized users.F. Corno    F. Razzak ( & )Dipartimento di Automatica ed Informatica,Politecnico di Torino, Corso Duca degli Abruzzi 24,10129 Turin, Italye-mail: raja.faisal@gmail.comF. Cornoe-mail:  1 3 J Ambient Intell Human ComputDOI 10.1007/s12652-013-0183-x  tool [e.g., ontologies (Heider and Kirste 2002; Encarnac¸aoand Kirste 2005; Bonino and Corno 2008)], but the research trend is moving from a traditional device-centricvision (bottom-up) to a vision of providing higher leveldesign for user interaction and control (Ducatel et al. 2003;Amigoni et al. 2005; Cheng et al. 2009; Kaldeli et al. 2010), i.e., user goal modeling. However, this researchtrend has received little attention as acknowledged in Chenet al. (2012). One such, user goal modeling approach is‘domotic effects’ (DE) framework (Razzak  2013).DE framework models user intentions or goals (called domoticeffects )(Razzak 2013;CornoandRazzak 2012).The context is thateverydeviceina smart environment is capableof providing certain visible (perceivable) effects for a user.Theseeffects are fulfilledbypossiblestatesofthedevice.Forexample, an effect of   illumination  can be provided by a lampin ‘ON’ state. However, modern devices are complicated innature and a single device can have a composite state, whichmaybemodeledasconcurrentsub-states.Thesesub-statesareorthogonal regions combining multiple descriptions of adevice.For example,a TVset mayhaveanon–offstate (withpossible values ON or OFF), a volume state (with possiblevalues 0 through 100), a channel state (with possible valuesdependingonthesetofprogrammedchannels).Adevicestateisthereforecompositeinnatureandtherefore it is modeledasthe parallel composition of different sub-states.There might be cases in which an effect can only be ful-filled by a combination of devices having particular states.Forexample,the effectof  securing a building mayrequire allthe exit doors and windows tobe closed. Inthe context ofDEframework, an effect that depends upon a single device(having a state or sub-states) is called a simple effect (SE)and an effect dependent on a combination of devices (havingparticular states and sub-states) is called a complex effect(CE). A CE is described by combining SEs and other CEs.The DE framework is logically organized in a three-tieredarchitecture as shown in Fig. 1. The  core layer   contains thebasic class definitions for expressing domotic effects. Eachdomotic effect (DE) is expressed as a function of existingdevice states or sensor values. Such function is expressedusing a set of operators that can be extended by a AmIdesigner. The  AmI layer   encodes the set of operators definedor customized by the AmI designer, depending on theapplication domain. Finally, the  instance layer   represents thespecific DEs being defined in a specific environment.The DE framework addresses the concerns from perspec-tives of the AmI designer and the residents. It provides AmIdesignerswithanabstractionlayerthatenablesthedefinitionof genericgoalsinsidetheenvironment,inadeclarativeway,andthatcanbeusedtodesignanddevelopintelligentapplications.It provides a general framework for expressing functionalproperties, in a domain-dependent way: for each applicationdomain, the AmI designer may choose the most suitablerepresentation, and define suitable functional operators. Usingtheseoperators,varioususergoalsarethendefinedinaspecificenvironment. The high-level nature of the DEs, on the otherhand,alsoallowstheresidentstoprogramtheirpersonal,officeor work spaces as they see fit: they can define differentachievement criteria for a particular generic goal, by using thedomain-specific operators defined in the previous phase.This paper discusses the control aspect of the DEframework. The control aspect involves the ability of usersto manage and control their environments with the help of user-defined intentions or goals. This amounts to correctlymapping user goals in terms of a combination of deviceshaving particular states. In this paper, the control aspect of the  DE   framework restricted to Boolean applicationdomains is discussed. Boolean application domains aresuch domains in which the values of user goals (effects)can either be true (active) or false (inactive). They covermost control applications and many monitoring use casesin smart homes, offices and industrial plants.The paper is divided into seven sections. Section 2provides the formal and conceptual modeling for domoticeffects specialized to the case of Boolean applicationdomains, and then defines the problem tackled in the paper.The general approach adopted for enforcement is describedin Sect. 3, and later Sect. 4 defines its architecture and implementation. Section 5 shows results of the experimentscarried out on effects enforcement. Section 6 compares ourapproach to some related works and Sect. 7 concludes thepaper and highlights future work. 2 Modeling 2.1 FormalismGiven an intelligent environment, we define as  D  the set of installed  controllable devices d   2 D :  Each device is Fig. 1  Logical architecture of the domotic effects framework (Cornoand Razzak  2012)F. Corno, F. Razzak   1 3  characterized by a  device category  that, among the otherthings, defines the allowed  sub-states  for a device.Depending on the device category, for each device  d   wedefine the set of allowed sub-states  S  ( d  ); this set may bediscrete (e.g., {ON, OFF} for a lamp) or continuous (e.g.,[0, 100] for a volume knob). During system evolution, theactual state of each device is a time-dependent function s ð d  ; t  Þ 2  S  ð d  Þ :  The whole environment therefore possessesa  global state space  G  ;  represented by the Cartesianproduct of all device state spaces:  G ¼ Q d  2D  S  ð d  Þ ;  thusdefining a global environment state  g  2 G  : Formally, a domotic effect  ð  DE  Þ  is defined as a functionof the global state space:  DE   :  G ! V  ;  where  V   is anapplication-dependent value space. For example, for con-trol applications,  V ¼ f 0 ; 1 g  since each domotic effectrepresents the activation of a given state configuration;conversely, when dealing with energy savings,  V ¼ < þ since domotic effects may be used to represent consumedpower.AmI designers and end users may define custom do-motic effects by working with a domain-specific set of operators. Such operators work on the value space  V   rel-evant to the specific application domain. 1 The specificationof a  DE   function requires three levels of formalization:1. defining simple effects (SE), to extract a  V  -valuedquantity from a single device state. Formally, SE is afunction that considers the state on only one device, SE   :  S  ð d  Þ ! V  ;  such function is also time-dependentsince it depends on  s ( d  ,  t  ).2. defining  effect operators  working within  V  -spacealgebraic semantics, suitable for composing newfunctions in the application domain. Formally, anoperator  op  is a function  op  :  V   N  ! V  ;  where  N   represents the number of operands of the specific  op .3. defining complex effects (CE), by applying effectoperators to the values computed by other SE or CE.Formally, a CE is represented by a couple ( op , ð  DE  1 . . .  DE   N  ÞÞ  composed of an operator name  op and a list of domotic effects  DE  i  whose values are usedas operands.For each application domain, there would be a set of pre-defined SE and a set of operators that the users maycombine to compute the values of interest. For example, if we consider control applications ( V ¼ f 0 ; 1 g ), then the SEfunctions that may be adopted are:– for discrete-valued states, a SE detects whether a devicecurrently is in any given state. E.g.,  SE  ON ð d  ; t  Þ ¼ð s ð d  ; t  Þ ¼¼  ON Þ .– for real-valued sensors, a SE usually compares thecurrent sensor data with a threshold. E.g., SE  HOT ð d  ; t  Þ ¼ ð s ð d  ; t  Þ [ 30  C Þ .The instance layer defines a set  I   of all defined domoticeffects (instances), i.e.,  I ¼ f  DE  1 ;  DE  2 ;  . . . ;  DE   N  g .2.2 Conceptual modelingIn order to provide a formal knowledge-base for the DEframework, a modular ‘DogEffects’ ontology is developed.The DogEffects ontology is organized in a three-tierstructure, corresponding to the logical architecture of theDE framework (Fig. 1). It models the user defined goalsand their mapping to devices and their correspondingstates. The DogEffects ontology requires the concepts of devices and their states, the modularity pattern was adoptedfor designing the ontology. Modularity allows the ontologyto easily integrate with other ambient ontologies that modelenvironments. In our case, DogOnt (Bonino and Corno2008) is adopted. Three modeling layers of the DogEffectsontology are explained below: 2.2.1 Core layer  The core layer defines concepts lying at the foundation of the DE framework. The main concepts are illustrated inFig. 2. The core layer consists of three main classes, i.e.,Effect, EffectOperator and Operand. Every  domotic effect  is formally organized into a concept hierarchy inheritingfrom the  dogEffects:Effect  class. Effects can eitherbe simple ( dogEffects:SimpleEffect ) or complex( dogEffects:ComplexDeviceEffect ). For bothkinds of effects, domain-dependent subclasses are definedat the AmI layer.Simple effects (SEs) are the terminal nodes of the repre-sentation and compute a value depending on a device state orsensor value. SEs act as interface points between the Dog-Effects ontology and some device description ontology (e.g.,DogOnt). The  dogEffects:effectOf  and  dogEf-fects:functionOf  open relations (i.e., relations with-out range restrictions) permit to identify the device and thedevice state for which a given SE is computed, respectively.Every complex effect (CE) represents a functionalexpression of SEs and other CEs declared by usingdomain-dependent operators defined at the middle-layer of the domotic effects framework. Effect operators take eithersimple or complex effects as operands (through the  dog-Effects:hasOperand  relation) and generate new CEsas result, identified by means of   dogEffects:hasRe-sult  relation.Two main disjoint families of operators are modeled:unary operators ( dogEffects:UnaryOperator ) and 1 For cross-domain applications,  V   would be the union of all relevantvalue spacesSAT based enforcement of domotic effects  1 3  non-unary operators ( dogEffects:NonUnaryOper-ator ). Unary operators only involve one  dogEf-fects:Operand  (cardinality restriction on the dogEffects:hasOperand  relation), which can eithermap to a SE or a CE (by the  dogEffects:operand-Effect relation). Typical examples of unary operators arethe NOT operator (in the boolean control domain). Non-unary operators are further specialized (disjoint union) intocommutative ( dogEffects:CommutativeOpera-tor ) and not commutative ( dogEffects:NotCom- mutativeOperator ) operators. According to themathematical definition of commutative (not-commutative)operators, the result produced by the former ( dogEf-fects:CommutativeOperator ) is independent onthe order in which operands are evaluated while, the latteroperator provides a results depending on the order of theeffect operands. In such a case the  dogEffects:Or-deredOperand  subclass of operands shall be used toaccount for the operand order, expressed as an ascendinginteger number by the  dogEffects:hasPositionN property. Typical examples of non-unary effects are theAND, OR and EX-OR logic operators, in the booleancontrol domain. 2.2.2 AmI layer  The AmI layer allows AmI designers to define the domain-dependent effect operators that govern the combination of domotic effects. Every application domain will definedifferent operator classes for this layer by sub-classing thegeneral operator classes defined in the core layer. InBoolean application domain, simple effects correspond todevices (sensors) being in specific states (measuring spe-cific range of values). SEs and CEs can only evaluate totrue or false, and Boolean logic is sufficient to computeCEs and to implement rather advanced activation scenar-ios. From the modeling point of view, mapping operators toBoolean logic requires a minimum set of logic operators,e.g., AND ( ^ ), OR ( _ ) and NOT ( : ) (see Fig. 2, bottom);however, AmI designers may choose to define more com-plex, user-intelligible operators such as the ONE_OFoperator (true iff exactly one of the operands is true), theNONE_OF operator, and so on. Some of them areexplained below:1.  ImpliedOperator:  This operator represents the ‘‘logicalimplication’’ relationship.2.  AlternateOperator   ( A ): This operator represents afunction whose value is active when  exactly one  of its operands is active. It is commutative and non-unary.Mathematically, the alternate effect operator can bedefined as Eq. 1.3.  ExactlyMOperator:  This non-unary operator repre-sents a function whose value is active when  exactly M  number of its operands are active. Suppose there are n  operands, i.e.,  OP  ¼ f 1 ; 2 ;  . . . ; n g :  Then the  Exact-lyMOperator   effect operator can be defined as Eq. 2. Fig. 2  The DogEffects ontology (core and AmI layer)—Boolean application domainF. Corno, F. Razzak   1 3   Alt  ð  x 1 . . .  x n Þ ¼ X i  x i   Y  j 6¼ i  x  j ! :  ð 1 Þ Exactly  M  ð  x 1 . . .  x n Þ ¼ X O  OP ; j P j¼  M  Y i 2 O  x i   Y  j 62 O  x  j "#  ð 2 Þ 2.2.3 Instance layer  The instance layer of the DE framework represents specificDEs defined in a given smart environment. They aremodeled as instances of the classes defined in the core andAmI layer. The DEs are defined according to user prefer-ences, i.e., can be defined by the users using a GUI or canbe designed by AmI designers.Consider a trivial example in which a user wants toachieve illumination in the room using lamps, called‘Lamp Illumination’ use case. 2 The ‘Lamp Illumination’use case is illustrated in Fig. 3. The ‘Lamp Illumination’use case will be represented as a ‘Lamp Illumination’ CE(an instance of   ComplexEffect ). Suppose the ‘‘LampIllumination’’ can be achieved using ‘Left Wall LampIllumination’ SE and ‘Right Wall Lamp Illumination’ SE.The combination is governed by the ‘‘And1’’ instance of  And Operator ) class. The ‘Left Wall Lamp Illumina-tion’ SE represents the ‘Lamp4’ in ‘OnState_lamp4’ state,while the ‘Right Wall Lamp Illumination’ SE representsthe ‘Lamp5’ in ‘OnState_lamp5’ state. The informationabout specific devices and their states comes from theDogOnt ontology. In order to provide easy comprehensionFig. 4 shows the simplified graphical representation of the‘Lamp Illumination’ CE and Table 1 outlines the func-tional representation of the ‘Lamp Illumination’ CE. InTable 1, a SE is represented as  SE  [device, sub-state(s)].For instance, the ‘Right Wall Lamp Illumination’ SE isequivalent to  SE  (Lamp5, OnState_lamp5). On the otherhand, a CE is represented as  Operator  (  DE  1 ;  DE  2 ;  . . . ). Forinstance, the ‘Lamp Illumination’ CE is equivalent to  And (Left Wall Lamp Illumination, Right Wall LampIllumination).The reader interested in more details about the DogEf-fects ontology is referred to (Razzak  2013).2.3 Problem statementConsider a smart environment with an AmI system man-aging it. A user can define several domotic effects (simpleand complex) on top of the domotic structure, based on theeffect operators defined for the environment. At anyinstant, each domotic effect has a value associated with it.The user has the ability to request  R  the AMI system toenforce a set of domotic effects on the environment. ‘effectenforcement’ addresses the problem of finding at least oneconfiguration that satisfies the user’s request  R . The con-figuration refers to the combination of devices havingparticular states and sub-states.The user request  R  is defined as a subset of the domoticeffects present in the instance layer:  R   I  :  In simpleterms, the user request  R  is the subset of   DE  i  that the userwants to be active (true) at a given instant.Given a request  R , effect enforcement consists of find-ing a global domotic state  g  2 G   where  all  the domoticeffects  DE  i  2  R  are true. This is equivalent to computingthe  satisfiability  of the function shown in Eq. 3. F   R ð g Þ  : Y  DE  i 2  R  DE  i  ð 3 Þ 3 Approach In order to enable the user to enforce particular values of domotic effects on the environment, at least a configurationneeds to be found which fulfills the user request  R , asdefined in Sect. 2.3. To solve this problem the paper pro-poses to transform the user’s request into a Boolean sat-isfiability problem (SAT). SAT is a decision problem todetermine whether a given propositional formula is a tau-tology (Cook  1971). In other words, it establishes if thevariables of a given formula can be assigned in such a wayas to make the formula evaluate to true. Equally importantis to determine whether no such assignments exist, whichwould imply that the function expressed by the formula isidentically false for all possible variable assignments.To transform the user’s request into a SAT problem,each domotic effect defined in the instance layer is mappedas a Boolean variable. The functionality of each effectoperator defined in the AmI layer is mapped in terms of aBoolean sub-expression in the SAT problem. The value of the variable corresponding to the Simple Effect is true(active) if and only if the device is in a particular sub-state(s). Meanwhile, complex domotic effects can dependupon the values computed by multiple simple or complexdomotic effects and, therefore, the value corresponding totheir variables are dependent on the values of their oper-ands. As a consequence, the Boolean expressions for acomplex domotic effect are constructed over its dependentdomotic effects using the effect operator defined for it. Theprocess is recursive, as the Boolean expressions for all theoperands are constructed and conjuncted.For example, consider a trivial user request  R  to enforcethe ‘Illumination’ use case on the environment. The ‘Illu-mination’ use case represents the user intention to achieveillumination in a room. The ‘Illumination’ use case will berepresented as an ‘Illumination’ CE inside the ‘DogEffects’ 2 A use case represents a scenario or the overall effect that a userwants to achieve.SAT based enforcement of domotic effects  1 3
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks