A Multiagent Model for Intelligent Distributed Control Systems.
ABSTRACT One approach for automating industrial processes is a Distributed Control System (DCS) based on a hierarchical architecture.
In this hierarchy, each level is characterized by a group of different tasks to be carried out within the control system and
by using and generating different information. In this article, a reference model for the development of Intelligent Distributed
Control Systems based on Agents (IDCSA) inspired by this hierarchical architecture is presented. The agents of this model
are classified in two categories: control agents and service management environment agents. To define these agents we have
used an extension of the MAS-Common KADS methodology for Multi-Agent Systems (MAS).
- SourceAvailable from: Juan Carlos Terán Picón
Conference Paper: Cultural Algorithms-Based Learning Model for Multi-Agent Systems[Show abstract] [Hide abstract]
ABSTRACT: This paper aims to evaluate the learning model for coordination schemes in multiagent systems (MAS) based on Cultural Algorithms. The model is applied to a case of study in industrial automation, related to the Agents-based System for Fault Management System. The instantiation occurs on the conversations that are defining in the MAS's coordination model, which are characterized by type of conversation that have been previously defined. A conversation can have sub-conversations, and in this case the sub-conversations are characterized by a particular type of conversation. Additionally in these conversations can occur some type of conflict, that can be solved by using different coordination mechanisms existing in the literature. For this, it is developed a model based on cultural algorithms, which is used by the MAS as a learning way in the process to determine which coordination mechanism is more suitable for a given conversation and a given scenario. The results show that the obtained model through this learning guides the MAS to determine which mechanism is better suited for a given conversation. I. INTRODUCCIÓN Un sistema multiagente (SMA) está formado por un grupo (comunidad) de agentes que interactúan entre sí utilizando protocolos y lenguajes de comunicación de alto nivel, para resolver problemas que están más allá de las capacidades o del conocimiento de cada uno . Estas interacciones entre agentes pueden ser vistas como conversaciones, que a su vez pueden tener sub-conversaciones. Para caracterizar estas sub-conversaciones se utilizan tipos de conversación (TCs), los cuales han sido definidos previamente , éstos permiten generalizar las interacciones o conversaciones entre agentes de cualquier comunidad. Ahora bien, estas sociedades de agentes pueden enfrentar conflictos a la hora no sólo de comunicarse, sino también a la hora de administrar recursos entre los individuos o a la hora de asignar tareas, etc. Para manejar dichos conflictos, existen los esquemas de coordinación (mecanismos de coordinación, MC) que permiten la resolución de los mismos. En este trabajo se propone un modelo de aprendizaje y optimización de esquemas de coordinación para SMA basado en Algoritmos Culturales (AC). Estos algoritmos permiten compartir experiencias entre los individuos, ya que uno de los componentes principales de estos algoritmos es un espacio común de experiencias, proveyendo así la capacidad de un aprendizaje colectivo basado en el intercambio de conocimientos. Dentro del marco de la coordinación en SMA existe una gran cantidad de trabajos orientados a su estudio, e. g., en el trabajo de  se enfocan en diseñar agentes que logren una óptima, eficiente y flexible coordinación. Para lograr esto basan su modelo en la teoría de juegos y derivan una solución llamada Harsanyi-Bellman Ad hoc coordination (HBA), la cual utiliza el equilibrio de Nash Bayesiano para planear procedimientos que lleven a encontrar acciones óptimas en el sentido del control óptimo de Bellman (programación dinámica). En  estudian la coordinación distribuida en sistemas multiagente modelando dinámicas no-lineales que caracterizan las interacciones entre los agentes con grafos no dirigidos. En  desarrollan mecanismos de coordinación para la implementación del control autónomo en procesos logísticos, con tecnología multiagente. El paradigma de logísticas autónomas reduce la complejidad computacional, y se enfrenta a la dinámica local, delegando procesos de control a los objetos participantes. Finalmente, el trabajo de  aplica algoritmos culturales a una comunidad de agentes, para estudiar la población prehispánica de de la región Central de Mesa Verde en EEUU. En ese trabajo utilizan los AC cómo método de aprendizaje para que los agentes puedan reconocer a aquellos individuos que estén más dispuestos a colaborar en proveer recursos (alimentos).Conferencia Latinoamericana en Informática CLEI 2013, Club Puerto Azul, Naiguatá, Vargas, Venezuela; 10/2013
- [Show abstract] [Hide abstract]
ABSTRACT: It is usually agreed that a system capable of learning deserves to be called intelligent; and conversely, a system being considered as intelligent is, among other things, usually expected to be able to learn. Learning always has to do with the self-improvement of future behavior based on past experience. In this paper we present a learning model for Multi-Agent System, which aims to the optimization of coordination schemes through a collective learning process based on Cultural Algorithms.2014 IEEE Congress on Evolutionary Computation (CEC), Beijing, China; 07/2014
- [Show abstract] [Hide abstract]
ABSTRACT: The good performance of a set of computer systems based on agents depends on the coherence degree and coordination between their activities. The study of coordination problem is an important topic for designers and researchers in the multi-agents systems field. There are several coordination mechanisms in the current literature, being the auction and the contract net the most popular ones. These mechanisms allow the agents to allocate resources and tasks to achieve their objectives. This paper aims to present formal models of the auction and the contract net as coordination mechanisms in multi-agents systems based on FIPA (Foundation for Intelligent Physical Agents) Protocols. Mathematical equations describe the different parameters characterizing the auction and the contract net mechanisms; they allow define a generic structure of each mechanism and groups of agents can create several instances of them to coordinate their needs.clei electronic journal. 08/2013; 16(5).
R. Khosla et al. (Eds.): KES 2005, LNAI 3681, pp. 191–197, 2005.
© Springer-Verlag Berlin Heidelberg 2005
A Multiagent Model
for Intelligent Distributed Control Systems
José Aguilar, Mariela Cerrada, Gloria Mousalli,
Franklin Rivas, and Francisco Hidrobo
CEMISID, Dpto. de Computación, Facultad de Ingeniería,
Av. Tulio Febres. Universidad de los Andes, Mérida 5101, Venezuela
Abstract. One approach for automating industrial processes is a Distributed
Control System (DCS) based on a hierarchical architecture. In this hierarchy,
each level is characterized by a group of different tasks to be carried out within
the control system and by using and generating different information. In this ar-
ticle, a reference model for the development of Intelligent Distributed Control
Systems based on Agents (IDCSA) inspired by this hierarchical architecture is
presented. The agents of this model are classified in two categories: control
agents and service management environment agents. To define these agents we
have used an extension of the MAS-Common KADS methodology for Multi-
Agent Systems (MAS).
Presently, there is great interest in the development of integrated automation systems
that permit monitoring different plant operation variables in a broad and dynamic way
and to transform such variables in control commands that are later integrated into the
plant through actuators. Moreover, they should take into account production and
economic criteria that can be applied as control commands or as part of a plant pro-
gramming function. The integrated structure should permit the flow of information at
all levels (management, operation, etc.) concerning the plant, the products obtained,
and all the relevant information. On the other and, a MAS can be defined as a net-
work of “problem- solvers” that work together to solve problems that could not be
worked out individually . The main preoccupation of the MAS is the coordination
between the groups of autonomous agents, perhaps intelligent, to coordinate their
objectives, skills, knowledge, tasks, etc.
In this work, a reference model for Intelligent Distributed Control System (IDCS)
based on Agents is proposed. Our model will be made up of entities called agents,
that work together dynamically to satisfy the control systems local and global objec-
tives and whose design can be made completely independent of the system could be
developed. The description of the agents in this reference model is based on the
MASINA methodology  which has an extension of MAS-Common KAD method-
ology to incorporate other characteristics of agents such as emerging behavior, the
reasoning, and the possibility of using intelligent techniques (expert systems, artificial
neuronal networks, genetic algorithms, fuzzy logic, etc.) for carrying out their jobs.
192 José Aguilar et al.
2 Reference Model for Intelligent Distributed Control System
Based on Agents (IDCSA)
The control jobs and information management needed in automation processes can be
distributed and expressed through a hierarchical logical structure. Figure 1 shows
hierarchical reference architecture to develop a Distributed Control System (SCD),
that permits the automation of an industrial plant .
Fig. 1. Reference Model for Distributed Control Systems
At the business and planning levels decisions are made at the managerial level, and
the control process jobs are carried out at the lowest levels. The Supervisory level
adjusts the parameters of the controllers, the control signal is obtained at the Local
Control level, to later be incorporated at the plant at the Process level. This SCD
model is complemented with a group of agents in each one of the levels of the hierar-
chy, this is the IDCSA model shown in Figure 2. These agents carry out diverse jobs
looking towards reaching the specific control objective.
Fig. 2. IDCSA Reference Model
A Multiagent Model for Intelligent Distributed Control Systems 193
The incorporation of the MAS to the reference model permits the control to
emerge from the interactions of those entities (agents). The IDCSA is seen as a net-
work of autonomous agents, with different responsibilities according to the level to
which it belongs. As such, the details of each agent (objectives, communication, jobs,
intelligence, etc.) come according to the levels to which the given agents belong in
the IDCSA model. This way, the agents are distributed through the control hierarchy
and can be geographically dispersed. The dynamic interaction of the multiple agents
happens between levels and in the interior of the levels. In the IDCSA model two
categories of agents exist:
Control Agents: Carry out the jobs of control, measurement, control decision mak-
ing, and putting the decisions into practice, among others. The agents are:
• Coordinating Agent: make decisions and plan control jobs
• Specialized Agent: carry out specific jobs that serve to support the coordinating
• Controller Agent: obtains control action
• Actuator Agent: executes the control action
• Observation Agent: measures and processes the variables of the plant
• Human Agent: supervises the SCIDA
Agents of the Service Management Environment (SME): It is the base of the dis-
tribution system given that it manages the communication of the IDCSA model and
permit the distribution of the control system and heterogeneity among geographically
dispersed agents. The agents of the SME are :
• Agent Administering Agents: coordinates the multi-agent society
• Resource Management Agent: administers the use of the resources.
• Applications Management Agent: administer the use of applications (software) of
• Data Base Agent: manages the information storage means.
• Communication Control Agent: establishes communication with other MAS.
3 Specifications of the IDCSA Reference Model
The models proposed in the MASINA methodology have been used according to the
needs of IDCSA .
3.1 Organization Model
The main objective of the organization model is to specify the environment in which
IDCSA will be developed. In this stage of modeling, a human organization is ana-
lyzed to determine the areas susceptible to the introduction of the IDCSA model. The
model is conceived in three stages:
• Modeling of the organization environment: Describes the environment where we
are going to introduce the IDCSA model.
• Evaluating the organization environment: The viability of introducing the IDCSA
model to the organization is evaluated.
194 José Aguilar et al.
• Model of the Multi-Agent Society: In this stage, the following activities are de-
– Identify the IDCSA levels in the organization: some applications can consider
the levels proposed in the IDCSA; others only require a group of them .
– Agent’s proposal at each IDCSA level: after identifying the levels needed, the
group of IDCSA agents adapted to the functions that are carried out at each
level is defined.
– Identify the environment objects: in each one of the levels, the surrounding ob-
jects with which agents will interact to achieve their objective should be identi-
3.2 Agent Model
The agent model serves to specify the generic characteristics of the IDCSA agents.
The agent model proposed is oriented towards services. Agents are capable of carry-
ing out jobs which are offering to other agents (called services). An agent will be
selected, the Observation Agent, to discuss the templates of the Agent Model (to see
the remaining templates, refer to ).
1. General Information: i) Name: Observation Agent, ii) Type: Software Agent and
Physical Agent, iii) Paper: Environmental, iv) Position: Lowest level agent of the
Multi-agent System, v) Description: The Observation Agent makes up the data
acquiring system: the sensor, the conditioning system and the transmitter.
Through them changes in the process can be detected.
2. Agent Objectives: i) Name: Process Variable Measurement, ii) Type: Persistent,
direct objective, iii) Entry Parameters: Variables that need to be measured, reposi-
tory where data is stored, iv) Exit Parameters: Measured variable, v) Activation
Condition: Order of measurement (non-periodic case) or period of time available
for measurement, vi) End Condition: Measured variable, vii) Representation Lan-
guage: Natural language, vii) Ontology: Control ontology, viii) Description: is as-
sociated with the capture of process variable values through carrying out direct or
indirect measurements on given variables. The obtained signals should be filtered
to eliminate measurement errors, and transported and stored in systems.
3. Agent Services: Process measurement variables, Conversion of variables, and
Detection of Quick Changes
4 Agent Capabilities: contact with the process, make available to the rest of the
MAS changes occurring in the environment
3.3 Job Model
The elements of this model are ingredients, capacity, environment and method. A job
has associated entry ingredients that permit the job to be executed and at the same
time, can produce exit ingredients as a product of the execution job. Jobs can be made
using a specific technique (classic, intelligent) or hybrid techniques. Below a list of
IDCSA jobs is presented. In  the use of each one of the agent’s cases is presented,
in which the jobs that are carried out is inferred.
A Multiagent Model for Intelligent Distributed Control Systems 195
1 Measurement Jobs: i) Measure: Sense and condition, ii) Process the variable
2 Control Jobs: i) Decision-making, ii) Obtain control action, iii) Process control
action, iv) Execute control action
3 Specialized Jobs: i) Process specialized agent
4 Information Management Job: i) Information storage, ii) Information up-date, iii)
Look for information, iv) Management of the BD
5 Location Jobs: i) Locate resources, agents or applications, ii) Assign
The Observation agent will continue to be utilized to describe the templates that
specify the jobs (to see the remaining refer to ).
Sensing: i) Objective: Process variables are registered in a device sensor, ii) Pre-
condition: The sensor is an element of the lowest level; the only thing that is required
is to have an associated process variable, iii) Control Structure: Sensing does not have
sub-jobs but rather belongs to a group of jobs that correspond to measuring agents,
iv) Execution Time: Continue, v) Frequency: Absolute frequency (constant), vi) Con-
tent: The variable is a signal coming from the process, which can be measured and is
associated with the state of the process, vii) Surrounding: Process level, viii) Regula-
tions: the sensor can be calibrated, ix) Method: Classic techniques for capturing data
3.4 Intelligence Model
This model is implemented in those intelligent agents which have reasoning capabili-
ties to decide about the jobs they are going to carry out to solve a situation. For that,
the agent can use previous experience or accumulated knowledge through the learn-
ing process. According to agent jobs and services defined by IDCSA, their intelligent
agents are: Coordinating Agents, Specialized Agents and Controller Agents. The
intelligent model for these agents is generically defined around three elements:
• Experience: i) Representation: Rules, ii) Type: Based on cases
• Learning Mechanism: i) Type: Adaptive, ii) Representation Technique: Rules, iii)
Learning Source: Process, exit and or failure conditions occurring in the proposal
of a control action, proposal of new control and decision-making parameter values
in coordination jobs, iv) Up-Dating: Feedback (using previous experiences for up-
• Reasoning Mechanisms: i) Information source: Previous results obtained for sys-
tem control agents, ii) Activation Source: control and coordination jobs, iii) Type
of Inference: Based on rules, iv) Strategic Reasoning: Coordinate strategies for the
proposed selection of appropriate control algorithms and for decision-making in
the supervisory and planning level. Confront unknown situations to enrich the ex-
3.5 Coordination Model
In general, this model describes the coordinating mechanisms between agents. Basi-
cally, this model describes the way activities are organized in the IDCSA to reach the
objectives. It deals with the planning process and conflict resolution mechanisms that
occur between agents. The model is complemented with a description of protocols,
196 José Aguilar et al.
ontologies, and communication mechanisms (direct and indirect). In the IDCSA, only
predefined planning mechanisms are used. The proposed coordinating model is ori-
ented to services. A service can have associated specific properties (cost, duration,
etc). For the IDCSA case, conversations generated from the predefined planning
process are: i) Obtaining control action, ii) Manage level objectives, iii) Sound alarm,
iv) Coordinate humans. Below, the first conservation alone is described:
1 When communication exists with the coordinator, the control agent is only in
charge of generating the control action (the associated interactions are shown in
2 When no communication exists with the coordinator, the control agent has the
possibility of auto-regulating itself. In this case, the control agent, making use of
the intelligent model, can identify parameters of the current situation and attempt
to define the parameters that remedy said situation.
Table 1. Interactions of the Conversation “Obtaining Control Action”
Transmit: Current Control (level of local
Search for information
Conversion of Control Signals to physical
3.6 Communication Model
This model describes each interaction between agents (speech acts). In the case of
IDCSA, a predefined planning process was used, which determines direct communi-
cation carried on between agents. Each interaction between two agents is carried out
through sending a message, and has a speech act associated with it (see table 1).
This work proposes a reference model for distributed control systems based on MAS
whose architecture is inspired in a hierarchical scheme. The use of MAS incorporates
collaborative, organizational and intelligent aspects that permit the control system to
have emerging behavior. The use of MASINA methodology permitted the specifica-
tion of the IDCSA model. The IDCSA model allows model distributed control system
in a way that we can include intelligent components in our system. Two main aspects
can be considered in our model: we can distribute the different control tasks among
the control agents of IDCSA, and the intelligence is part of the control agents.
1. Wooldridge M., Jennings N.: Intelligent Agents: Theory and Practice. Knowledge Engineer-
ing Review (1994).
A Multiagent Model for Intelligent Distributed Control Systems 197
2. Aguilar J., Cerrada M., Hidrobo F., Mousalli G., Rivas F., Díaz, A: Application of the
Agents Reference Model for Intelligent Distributed Control Systems. Advances in Systems
Sciences Measurement, (Ed. N. Matorakis, L. Pecorelli), WSES Electrical and Computer
Engineering Series (2001) 204-210.
3. Franek B., Gaspar C.: SMI++ object Oriented Framework for Designing and Implementing
Distributed Control Systems. IEEE Trans. on Nuclear Science, Vol. 45 (1998), 1946-1950.
4. Haring G., Kotsis G., Puliafito A., Tomarchio O.: A Transparent Architecture for Agent
Based Resource Management. Proc. Of IEEE Intl. Conf. On Intelligent Engineering (1998)
5. Hidrobo F., Aguilar J.: Resources Management System For Distributed Platforms Based on
Multi-Agent Systems. Proc. of 5th International Conference on Artificial Intelligence, Simu-
lation and Planning-AIS' 2000 (2000) 60-69.
6. Aguilar J., Cerrada M., Hidrobo F., Mousalli G., Rivas F., Díaz, A: Especificaciones de los
Agentes del SCDIA usando MASINA. Technical Report # 23-2004, FONACIT, Ministerio
de Ciencia y Tecnologia, (2004).