Modeling Hierarchical Resources Within a Uniﬁed Ontology

A Position Paper

Alexander Schiendorfer

, Yves Wautelet

and Wolfgang Reif

Institut f

ur Software & Systems Engineering, Universit

at Augsburg, Augsburg, Germany

Faculteit Economie & Management, KU Leuven, Leuven, Belgium

Keywords:

Resource Management, Resource Allocation, Uniﬁed Ontology, Smart Grids, Hierarchical Modeling.

Abstract:

Resource-Intensive Software Ecosystems (RISE) can mainly be found in production management but also

in virtually any socio-technical environment. RISE appear prominently in the form of smart grids or cloud

environments where optimizing resource utilization and allocation becomes the most important aspect for com-

petitive service provision. In such a context, the need for uniﬁed ontologies supported by adaptive software

(i.e., software able to learn from and act on its environment) is highly attractive. Indeed, resources are mostly

not monolithic entities but active and collaborative agents often organized in a hierarchical manner. A hierar-

chy implies multiple levels of abstraction leading to resource allocation on different levels of organization –

with abstractions being relevant for both inter- and intra-organization resource management. Once adequately

deﬁned, the use of constraint-based optimization algorithms on those multiple levels can provide efﬁcient re-

source allocation. We apply, in this paper, ontological elements to model resources in a uniﬁed manner on

multiple levels onto an example taken from distributed energy management. Then we present algorithmic

ideas to organize the hierarchy of these resources.

1 INTRODUCTION

We deﬁne Resource-Intensive Software Ecosystems

(RISE) as environments where resources take a

prominent role in the realization of the services pro-

vided to the end-user. In other words, they are cru-

cial for the proper execution of business processes.

(Wautelet et al., 2012) presents a uniﬁed ontology for

the modeling of resources into RISE; within the ontol-

ogy, resources are modeled in function of their usage

(see Section 2). Explicitly modeling resources in a

uniﬁed and standardized manner (i.e. independent of

their type, domain and interface) can provide us with

ﬂexible and reusable components.

(Cabanillas et al., 2013) identiﬁes a gap between

resource assignment at design time and resource al-

location at run time; this paper is part of the willing-

ness to dispose of a tool to make resource allocation

at run time only on the basis of resources designed

uniformly on the basis of a same ontology. This thus

helps ﬁlling this gap.

Resource allocation is indeed a task frequently

encountered in industrial environments and often re-

solved through agent-based software solutions imple-

mented within a distributed context (Chevaleyre et al.,

2006; Seebach et al., 2010). It can be achieved by the

use of different levels of organization leading to a syn-

thesis and abstraction problem in order to deal with

each level and impact on the other levels accordingly

(see (Schiendorfer et al., 2014)).

This paper illustrates the use of hierarchical re-

source composition as depicted in (Schiendorfer et al.,

2014) with the ontology presented in (Wautelet et al.,

2012). The main contribution is to show the rep-

resentation is feasible and to explore the beneﬁts of

a disciplined resource modeling process that leads

to algorithmic advantages in terms of hierarchies.

Also by resource composition, we address inter-

organizational problems in an efﬁcient manner. For

instance, a virtual power plant is a power plant acting

on behalf of real power plants. For an energy provider

it is thus much more convenient to sell a single value

at a certain time to their contractors (managing yet

another set of physical plants) which they in turn re-

distribute. Useful abstract models leaving aside or-

ganizations’ internal details have to be developed for

this purpose.

614

Schiendorfer A., Wautelet Y. and Reif W..

Modeling Hierarchical Resources Within a Uniﬁed Ontology - A Position Paper.

DOI: 10.5220/0005289006140619

In Proceedings of the International Conference on Agents and Artiﬁcial Intelligence (ICAART-2015), pages 614-619

ISBN: 978-989-758-074-1

 2015 SCITEPRESS (Science and Technology Publications, Lda.)

FunctionalityParameter

Functionality

ResourceObject

homogenous : Boolean

unit_of_measure : Single

quanitty : Single

1..n 0..n1..n 0..n

provides

SoftwareAgent

StatusParameter

Configuration

CostParameter

name : String...

type : String

ResourceList

ResourceStatus

CostOfUse

billing_model : String...

Usage Resource

1..n

0..n

1..n

0..n

owns

0..n

owns

Has

Contract

begin_of_contract : Date...

end_of_contract : Date

QualityParameter

name : String

type : String

QualityLevel

0..1

0..n

0..1

0..n

defines

HumanAgent

CompetencyParameter

ServiceRealization

0..n

uses

Competency

0..n

requires

ResourceAgent

virutal : Boolean

1..n

0..n

1..n

0..n

provides

FunctionalityParameter

Functionality

ResourceObject

homogenous : Boolean

unit_of_measure : Single

quanitty : Single

1..n 0..n1..n 0..n

provides

SoftwareAgent

StatusParameter

Configuration

CostParameter

name : String...

type : String

ResourceList

ResourceStatus

CostOfUse

billing_model : String...

Usage Resource

1..n

0..n

1..n

0..n

owns

0..n

owns

Has

Contract

begin_of_contract : Date...

end_of_contract : Date

QualityParameter

name : String

type : String

QualityLevel

0..1

0..n

0..1

0..n

defines

HumanAgent

CompetencyParameter

ServiceRealization

0..n

uses

Competency

0..n

requires

ResourceAgent

virutal : Boolean

1..n

0..n

1..n

0..n

provides

Figure 1: A meta-model representing an ontology for resource representation.

2 ABOUT THE

UNIFIED-RESOURCE

ONTOLOGY

There are numerous deﬁnitions of the concept of re-

source. The Merriam-Webster dictionary refers to it

as a person, asset, material, or capital which can be

used to accomplish a goal. While remaining basic,

this deﬁnition has the interesting aspect of identify-

ing the fact that a resource is not a process (some-

thing functional/operational) but should be used for

the proper achievement (the goal) of such a process

through action(s). This consequently leads to the in-

tuition that resources do own particular skills that are

required in a functional context (for the realization of

processes or services). Within a pool of resources, the

requesting service is thus only interested in resources

helping it achieving its functional requirements. Nev-

ertheless, depending on their type, resources are inan-

imate and do have a functional utility (this is the case

of Resource Objects) or are able to behave (this is the

case for Resource Agents). Resources’ offer and de-

mand can consequently not be centralized into a sin-

gle concept so that Resource Objects are functional-

ity providers while Resource Agents are competency

providers. Resources offer functionalities and com-

petencies while services demand them.

In (Wautelet et al., 2012), we model resources as

components encapsulating functionalities and compe-

tencies and illustrate it on a case study in the steel in-

dustry. Figure 1 depicts the uniﬁed meta-model using

a UML class diagram (OMG, 2011). Details about

the concepts and their interdependencies can be found

in (Wautelet et al., 2012). The present paper focuses

on the hierarchical resource allocation problem within

this ontology rather than on functionality and compe-

tency modeling.

ModelingHierarchicalResourcesWithinaUnifiedOntology-APositionPaper

615

500

10060140 160 40 140 60 100 160 40

300

200

500

Figure 2. Resource allocation problems can be solved using a hierarchical decomposition struc-

ture. Inner nodes representing virtual agents are marked by double circles.

Implementing this situation typically results in solving an optimization problem,

the objective being to allocate energy to the subordinates as cost-effectively as possible

such that the amount to be traded at the energy market is met. This corresponds to the

general one-good resource allocation problem without externalities [5], i.e. given a total

quantity x

of a resource, ﬁnd an allocation hx

, . . . , x

i of the resource to n agents to

solve

minimize

,...,x

i=1

) subject to

i=1

= x

We illustrate how the uniﬁed-resource ontology can be mapped to this scenario.

Concretely, we pick the service that a virtual power plant vp has to offer when com-

mitted to a certain set of operating reserve for the following day. Assume that we bid

to sell 500 KW at 14:00, 600 KW at 14:15, 700 KW at 14:30, and 400 KW at 14:45.

For the sake of simplicity, these quantities representing the bid for one hour of pro-

duction are represented as vectors. In total, this amounts to a provided power of 2200

KWh resulting in costs of 3300 e at 1.5 e per KWh. Assume further, that vp has two

plants available: a biomass power plant bm, and a hydro plant hy. All power plants, be

it virtual or not, are represented by the concept ResourceProvider which is a spe-

cialization of a resource agent that contains a set of intervals feasReg, representing

the feasible regions that a resource provider can contribute in.

Service Realization Provide energy at deﬁned cost and time slots.

Input contract: h500, 600, 700, 400i at 14:00 at costs ≤ 3300 e.

bm : ResourceProvider hy : ResourceProvider

Requires Resource e

0 at 14:00, e

0 at 14:15, etc.

The bm and hy power plants can ﬁrstly be deﬁned as ResourceAgent. We deﬁne:

Resource ResourceProvider bm

Type Resource Agent

Attribute feasible Regions : [0, 0], [100, 400]

costs : 1.7 e per KWh

Having the [0, 0] interval present indicates that the agent can be disconnected from

contributions and if it does contribute, it has to do so above a minimal level. The same

Figure 2: Resource allocation problems can be solved using a hierarchical decomposition structure. Inner nodes representing

virtual agents are marked by double circles.

3 DEFINING THE

HIERARCHICAL RESOURCE

ALLOCATION PROBLEM

The ontology presented in (Wautelet et al., 2012)

induces a possible hierarchy among resources

in the sense that a ResourceObject or a

ResourceAgent can either be an assembly –

i.e., a resource made of other resources advertising

functionalitites/competencies – or be atomic – i.e.,

not made of other resources advertising functionali-

tites/competencies. This is represented into the meta-

model of Figure 1 by the composition association

on the ResourceObject and ResourceAgent

concepts themselves.

A Resource Agent taking the form of an assem-

bly that is issued of the combination of other resource

agents is called a Virtual Agent (VA). The VA then

manages the tasks of its subordinates representing an

organization. This is shown in the ontology by the

boolean ﬁeld virtual in the ResourceAgent class.

As it is important for the case study we also high-

light the case of an homogeneous ResourceObject. In-

deed, a resource like electrical energy is said to be

homogeneous because it cannot be counted but is the

available in a deﬁned quantity. Within the case study

of Section 4, a quantity x

of an homogeneous re-

source is to be provided by n providers x

such that

∑

i=1

= x

. Figure 2 shows this decomposition and

how a certain quantity of such a resource can be hier-

archically provided.

4 CASE STUDY: ENERGY

DISTRIBUTION IN A VIRTUAL

POWER PLANT

In conventional energy systems, a small number of

rather large and inﬂexible power plants – such as coal

or nuclear power plants – come up for the majority

of the power production. Environment considerations

as well as the limited availability of fossil fuels in

the future motivate to use a larger number of smaller

but renewable energy providers. Examples of the lat-

ter providers are wind, hydro, photovoltaic or biogas

power plants. The resulting complexity of the system

puts stress on the control of the production. Indeed,

the deviation between demand and production must

be coordinated to leave the utility frequency in a well-

deﬁned band (e.g., [49.8, 50.2] Hz in Europe).

Clearly, the higher coordination demand between

production and demand is a resource-intensive pro-

cess and calls for IT-based solutions: the so-called

“Smart Grid” systems (Ramchurn et al., 2012). A

ﬁrst step in this direction is the introduction of “Vir-

tual Power Plants”, i.e., a single energy provider takes

the responsibility of providing a certain amount of en-

ergy at speciﬁc time steps by issuing a contract at an

energy market

. This top-level service, provide (cus-

tomers with) energy at deﬁned cost and time slots, is

divided into sub-services that the contractors (which

can be organizations or plant providers such as farm-

ers) offer. The contracts are typically traded one day

in advance and are binding for all involved parties.

Implementing this situation results in solving an

optimization problem, the objective being to allocate

energy to the subordinates as cost-effectively as pos-

sible such that the amount to be traded at the energy

market is met. This corresponds to the general one-

good resource allocation problem without externali-

ties (Van Zandt, 1995), i.e. given a total quantity x

of a resource, ﬁnd an allocation hx

, . . . , x

i of the re-

source to n agents:

minimize

,...,x

∑

i=1

) subject to

∑

i=1

= x

We illustrate how the uniﬁed-resource ontology

can be mapped to this scenario. Concretely, we pick

the service that a virtual power plant vp has to offer

see, e.g., https://www.regelleistung.net for the German

energy market

ICAART2015-InternationalConferenceonAgentsandArtificialIntelligence

616

when committed to a certain set of operating reserve

for the following day. Assume that we bid to sell 500

KW at 14:00, 600 KW at 14:15, 700 KW at 14:30, and

400 KW at 14:45. For sake of simplicity, these quan-

tities representing the bid for one hour of production

are represented as vectors. In total, this amounts to

a provided power of 2200 KWh resulting in costs of

3300 e at 1.5 e per KWh. Assume that vp has two

plants available: a biomass power plant bm, and a hy-

dro plant hy. All power plants, virtual or not, are rep-

resented by the concept ResourceProvider which is

a specialization of a resource agent that contains a set

of intervals feasReg for the feasible regions that a re-

source provider can contribute in.

As discussed, the aim of the paper is to highlight

the representation of resources on multiple levels of

abstraction. Figure 3 shows two levels of granular-

ity within the uniﬁed ontology, of course there could

virtually be an inﬁnity of levels. The service to be

realized is the service Provide energy at deﬁned cost

and time slots. The contract

Service Realization Provide energy at deﬁned cost

and time slots.

Input contract: h500, 600, 700, 400i at 14:00 at

costs ≤ 3300 e.

bm : ResourceProvider hy :

ResourceProvider

Requires Resource e

500 at 14:00, e

600 at

14:15, etc.

The bm and hy power plants can ﬁrstly be deﬁned

as ResourceAgent. We deﬁne:

Resource ResourceProvider bm

Type Resource Agent

Attribute feasible Regions : [0, 0], [100, 400]

costs : 1.7 e per KWh

Having the [0, 0] interval present indicates that the

agent can be disconnected from contributions and if it

does contribute, it has to do so above a minimal level.

The same description can be made for the Resource-

Provider hy. The feasible regions and costs for that

other power plant (hy) is given by {[0, 0][50, 300]} at

1.2 e.

Next, we deﬁne the energy resource provided by a

plant as ResourceObject, more speciﬁcally:

Resource Energy e

Type Resource Object

Attribute quantity and unit

Provides Electrical Energy

The same description can be made for Energy e

Finally, we illustrate the aggregation of the resources

like:

Resource AggregateEnergy e

Type Resource Object

Provides Energy equal to

∑

a∈{bm,hy}

Resource VirtualPlant vp

Type Resource Agent

Aggregates bm, hy

Attribute virtual: true, feasible Regions: [0,0],

[50, 700], cost: *piecewise linear function*

Figures 2 and 3 show the modeling of the two lev-

els of hierarchy using object diagrams. In the lower

side of the picture the lowest level of abstraction is

represented while the upper side of the picture shows

how a virtual agent can be built out of a composition

of resource agents.

After the description and the illustration on the

Figure, it is important to highlight that we can express

constraints on production (of energy in this particular

case) within instances between the association class

between the ResourceAgent instance and the Compe-

tency one. Typically the Contract instance is part of

the Service Level Agreement (SLA) that is expressed

at Service Realization instance level as a composition

of Contract instances for the realization of the ser-

vice. The building of the global SLA can thus also

be expressed on multiple levels and be rolled-up and

drilled-down on multiple levels.

5 SOLVING HIERARCHICAL

RESOURCE ALLOCATION

Once the resource agents are modeled adequately, we

can use model abstraction to reduce the complex-

ity of the optimization problem faced when solving

cost-effective resource allocation. First, for each in-

ner node in the hierarchy, a set of models of un-

derlying resource agents is collected and abstracted

into a single one (representing the group’s capabili-

ties) and can be handed over to the subordinate agent.

Second, the resource allocation can be performed in

a top-down and recursive manner. Algorithms as

described in (Schiendorfer et al., 2014) come into

play. We illustrate the concept with the case study

and show the simplest of all techniques, ﬁnding the

general feasible regions of one virtual power plant.

To do so, we essentially combine all feasible inter-

vals of the underlying agents and merge overlapping

intervals. If we activate only bm, {[100, 400]} is

possible, if we activate only hy, {[50, 300]} is vi-

able. Switching both off enables the [0, 0] interval

and switching both on means that the resulting in-

terval is given by [150, 700], hence we obtain the in-

ModelingHierarchicalResourcesWithinaUnifiedOntology-APositionPaper

617

Figure 3: Resource Modeling on Various Levels of Abstraction.

tervals {[0, 0], [50, 300], [100, 400], [150, 700]}. How-

ever, from an outside perspective the concrete con-

ﬁgurations do not matter (e.g., if 150 is provided by

(0, 150) or (50, 100)) but only the feasible and infeasi-

ble regions. This abstracted view is achieved by merg-

ing overlapping intervals yielding {[0, 0], [50, 700]}.

The abstracted model does not show particulars

of the underlying agents and can hence be seen as a

means to perform efﬁcient intra-organization commu-

nication as superordinate agents need only to worry

about a deﬁned interface (e.g., consisting of the avail-

able feasible regions of an abstracted ResourceAgent

and its cost function). This cost function can be found

by a an approach based on sampling, i.e., repeatedly

solving a combinatorial problem to build an approxi-

mation of the cost function. For our concrete exam-

ple, we sketch the steps involved for mapping pro-

duction to costs (input and output pairs are collected

to form piecewise linear functions):

1. Solve for minimize costs, given that the combined

production is 0

2. Solve for minimize costs, given that the combined

production is 50

3. Solve for minimize costs, given that the combined

production is 55

4. · · ·

6 CONCLUSION

RISE can beneﬁt from uniﬁed modeling mechanisms

particularly for environments requiring to be ex-

pressed on multiple hierarchical levels. We have,

in this paper, illustrated hierarchical modeling of re-

sources within the context of smart grids as well as

the impact onto the resource allocation problem. Ex-

ICAART2015-InternationalConferenceonAgentsandArtificialIntelligence

618

pressing the problem on multiple levels allows the fol-

lowing beneﬁts:

• express constraints, contracts and SLA on multi-

ple levels. Ultimately, the SLA is a composition

of contracts and is located at top abstraction level,

i.e. the service level;

• deal with resource allocation algorithms at multi-

ple levels with direct impact on the other levels as

depicted in (Schiendorfer et al., 2014);

• by nature, we can treat multi-organizational prob-

lems at once. Indeed, when multiple organizations

are depicted at lower level, the application of opti-

mization algorithms at upper level treats each peer

uniformly for an efﬁcient resource use and alloca-

tion at global level;

• by the associated development approach, it inher-

ently ﬁlls the gap between resource assignment at

design time and resource allocation at run time.

The work presented in the paper remains how-

ever preliminary; its main aim is to argue for the use

of a uniﬁed model in a multi-hierarchical and multi-

organizational context. The full instantiation and im-

plementation of the ontology onto the smart grid case

as well as on other case studies is currently under im-

plementation.

REFERENCES

Cabanillas, C., Garc

ıa, J. M., Resinas, M., Ruiz, D.,

Mendling, J., and Cort

es, A. R. (2013). Priority-based

human resource allocation in business processes. In

Basu, S., Pautasso, C., Zhang, L., and Fu, X., edi-

tors, Service-Oriented Computing - 11th International

Conference, ICSOC 2013, Berlin, Germany, Decem-

ber 2-5, 2013, Proceedings, volume 8274 of Lecture

Notes in Computer Science, pages 374–388. Springer.

Chevaleyre, Y., Dunne, P. E., Endriss, U., Lang, J.,

Lema

ıtre, M., Maudet, N., Padget, J., Phelps, S.,

Rodr

ıguez-aguilar, J. A., and Sousa, P. (2006). Is-

sues in Multiagent Resource Allocation. Informatica,

30(1):3 – 31.

OMG (2011). OMG Uniﬁed Modeling Language (OMG

UML). Version 2.4. Technical report, Object Man-

agement Group.

Ramchurn, S. D., Vytelingum, P., Rogers, A., and Jennings,

N. R. (2012). Putting the ’Smarts’ into the Smart Grid:

A Grand Challenge for Artiﬁcial Intelligence. Com-

mun. ACM, 55(4):86–97.

Schiendorfer, A., Stegh

ofer, J.-P., and Reif, W. (2014).

Synthesis and Abstraction of Constraint Models for

Hierarchical Resource Allocation Problems. In

Proc. 6

Int. Conf. Agents and Artiﬁcial Intelligence

(ICAART’14), Vol. 2, pages 15–27. SciTePress.

Seebach, H., Nafz, F., Stegh

ofer, J.-P., and Reif, W.

(2010). A Software Engineering Guideline for Self-

Organizing Resource-Flow Systems. In Proc. 6

Int. Conf. Self-Adaptive and Self-Organizing Sys-

tems(SASO’10), pages 194–203.

Van Zandt, T. (1995). Hierarchical Computation of the Re-

source Allocation Problem. European Economic Re-

view, 39(3-4):700–708.

Wautelet, Y., Heng, S., and Kolp, M. (2012). A usage-based

uniﬁed resource model. In SEKE, pages 299–304.

Knowledge Systems Institute Graduate School.

ModelingHierarchicalResourcesWithinaUnifiedOntology-APositionPaper

619