Intelligent Environment for Training of Power Systems Operators

B. Gabrys, R.J. Howlett, and L.C. Jain (Eds.): KES 2006, Part I, LNAI 4251, pp. 943 – 950, 2006. © Springer-Verlag Berlin Heidelberg 2006

Intelligent Environment for Training of Power Systems Operators

G. Arroyo-Figueroa1, Yasmin Hernandez1, and Enrique Sucar2

1 Gerencia de Sistemas Informaticos, Instituto de Investigaciones Electricas Reforma # 113, Col. Palmira,

69042 Cuernavaca, Morelos, Mexico {garroyo, myhp}@iie.org.mx

http://www.iie.org.mx/ 2 Instituto Nacional de Astrofísica Optica y Electrónica

Luis Enrique Erro #1, 72840 Sta. María Tonantzintla, Puebla, Mexico

[email protected]

Abstract. Training of operators has become an important problem to be faced by power systems: updating knowledge and skills. An operator must compre-hend the physical operation of the process and must be skilled in handling a number of normal and abnormal operating problems and emergencies. We are developing an intelligent environment for training of power system operators. This paper presents the architecture of the intelligent environment composed by computer based training components labeled as reusable learning objects; a learning object repository; concept structure map of the power plant domain, a tutor module based on course planner, operator model based on cognitive and affective components and operator interface. The general aim of our work is to provide operators of complex industrial environments with a suitable training from a pedagogical and affective viewpoint to certify operators in knowledge, skills, expertise, abilities and attitudes for operation of power systems.

1 Introduction

Training of operators is an important problem faced by thermal power plants: updat-ing knowledge and skills. The process of learning how to control, maintain and diag-nose in a complex industrial environment takes years of practice and training. An operator must comprehend the physical operation of the process and must be skilled in handling a number of abnormal operating problems and emergencies. The problem increases when the complexity of the system obstructs the efficiency, reliability and safe operation. So the novice and experienced operators need continuous training in order to deal reliably with uncommon situations.

Distributed control system (DCS) and management information systems (MIS) have been playing an important role to monitor the plant status. These systems are generally employed to deal with information during normal operation. However, in abnormal operations such as equipment failures and extreme operation, human opera-tors have to rely on their own experience. During disturbances, the operator must

944 G. Arroyo-Figueroa, Y. Hernandez, and E. Sucar

determine the best recovery action according to the type and sequence of the signals received. In a major upset, the operator may be confronted with a large number of sig-nals and alarms, but very limited help from the system, concerning the underlying plant condition. Faced with a vast amount of raw process data, human operators find it hard to contribute a timely and effective solutions.

For safety efficiency reasons it is not recommend to train operators on real equip-ment. Several ways to cope with operators training has been proposed. Most pro-posals for operator training are based on Computer Based Training Systems (CBT) and simulators. CBT is a computer system that groups a set of programs and train-ing methods that uses the computer capabilities to present the operator with the instructional material and to provide an environment for practical training. The main advantage of CBTs are their unlimited availability: they can be used whenever the trainee wants.

Due to necessity of training of operation skills in power plants, simulators have been a common alternative. A power simulator is composed of a set of quantitative and qualitative models which simulate the physical behavior of the plant. The aim of training with a simulator is to get trained operators not only in performing the common procedure, but in analyzing abnormal situations and performing emer-gency procedures as well. The next generation in training systems are computer-based training systems that provides instruction just like a human trainer does. The training requirements ask for powerful interfaces, a more efficient and better adap-tive training, by means of incorporating artificial intelligent (AI) techniques, adap-tive interfaces, simulations tools, learning objects based on multimedia and virtual reality components.

This paper describes the architecture and development of an intelligent environ-ment (IE) for the training of power systems operators. In contrast with a traditional training system, the main goal of the intelligent environment is to certify operators in knowledge, skills, expertise, abilities and attitudes for operation of power systems.

2 Architecture of the Intelligent Environment

The architecture of the intelligent environment is based on dynamic course generating systems proposed by Brusilovsky [1]. The architecture of the intelligent environment, shown in Figure 1, is composed by five main modules: Learning Content Module, Domain Knowledge Module, Tutor, Operator Model, and Operator Interface.

2.1 Learning Content Module

This module contains tools for edition of teaching and testing materials based on learning objects. The teaching and testing materials consider both theoretical and practical concepts contained in: electronic books, simulation tools, multimedia and virtual reality tools, to present to the operator pedagogical actions such as explana-tions, exercises, tests, and so on. Each material is labeled as Reusable Learning Ob-jects (RLO) or Shared Content Objects (SCO) according with the SCORM (Sharable

Intelligent Environment for Training of Power Systems Operators 945

Content Object Reference Model) terminology [2]. Learning objects are self-contained learning components that are stored and accessed independently. RLO are any digital resource that can be reused to support Web-based learning and learning management system (LMS). The authors or learning developers can create, store, reuse, manage and deliver digital learning content. The same object may be used as many times and for as many purposes as is appropriate. XML servers to this function by separating content from programming logic and code [3].

Concept Structure

Learning Objects

Repository

Domain

knowledge

Learning Content

Editor

Concept Structure

Learning Objects

Operator Interface

Operator Model

Cognitive

Profile

Afective

Concept Map

Course Planner

Tutor

Course

Pedagogical Component

Concept Structure

Learning Objects

Repository

Domain

knowledge

Learning Content

Editor

Concept Structure

Learning Objects

Operator Interface

Operator Model

Cognitive

Profile

Afective

Concept Map

Course Planner

Tutor

Course

Pedagogical Component

Fig. 1. Architecture of the intelligent environment

The learning objects are either dispensed to users individually or used as compo-nents to assemble larger learning modules of full courses, depending on individual learning needs. The instructional output many be delivered via the Web, CD-ROM or printed materials. We have developed reusable learning objects for two areas: valve maintenance and maintenance of energized power-lines, based on virtual reality [4]. Some examples are depicted in figure 2.

2.2 Domain Knowledge Module

The domain knowledge module has two main components: the concept structure map and the learning objects repository. The domain knowledge contains the expert’s operation abilities in procedure and malfunction operations and its representation considers both theoretical and practical concepts. It is possible to organize the domain concepts/topics into a set of smaller, possibly interrelated AND/OR graphs, represent-ing relatively independent sub-areas of the knowledge, different views, or different levels of granularity.


Fig. 2. Examples of RLO based on virtual reality for valves (left) and power lines (right) [4]

The concept structure contains the concept/topic structure of the subject knowledge to be taught (see figure 3). It is represented as an AND/OR graph, where nodes repre-sents the concepts domain or elements of knowledge, such as electrical topics, com-ponents of control board, rules, procedures and so on; and arcs represents relation between concept, such as a prerequisite for learning a concepts or a sequence. Every node is associated with a set of teaching and testing materials labeled as RLO, which instantiate different ways to teach the concept/topic (e.g. introduce, explain, give an example, give a simulation, exercise, or test).

For the training of power plant operators, the concept structure map is made based on the structural decomposition of the generation process of electricity. The genera-tion process can be divided in unit, structure, systems, subsystems, equipment and component. The concept structure has a higher expressive power because it allows representing not only prerequisites, but many different types of relations between concepts; and it enables the use of AI planning techniques for the generation of alter-natives courses. Therefore, it guarantees a wide variety of different teaching goals and several courses for achieving these goals.

The learning content module has a Learning Object Repository (LOR). The LOR is a central database in which learning content (RLO) is stored and managed.

2.3 Tutor

The tutor is the component that generates the sequence of learning objects (teaching and testing materials) to be presented to operator in a form of course. Taking as a basis the AND/OR graph representing the concept map, the pedagogical component and the operator model, the training course planner generates a specific course. The concept plan is a sub-graph of the graph of knowledge domain (concept structure). The pedagogical component provides a set of teaching rules for the selection of con-tent and presentation of the concept plans according with the cognitive style, affective state or learning preferences of the operator. The pedagogical component contains the learning goals and several pedagogical strategies to teach operators and select


the pedagogical strategy to be used, based on operator model. This strategy represents the specific necessities of the operator from both, affective and pedagogical point of view.

Unit

Structure

Unit 1 Unit 2

Boiler1

Condenser Generator

Turbine1 preboiler

Fuel-oil

Air-gas Water-steam

hp-turbine

Ip-turbine

lp-turbine

feeding

water

System

ECO

DRUM

WW1

DCOMER

DEAE

FPUMP

FVALVESub-system

Equipment PIcontrol

Component LAGaditivesl

pipe

valve

pump themometer

Unit

Structure

Unit 1Unit 1 Unit 2Unit 2

Boiler1Boiler1

CondenserCondenser GeneratorGenerator

Turbine1Turbine1 preboilerpreboiler

Fuel-oilFuel-oil

Air-gasAir-gas Water-steamWater-steam

hp-turbinehp-turbine

Ip-turbineIp-turbine

lp-turbinelp-turbine

feedingfeeding

waterwater

System

ECOECO

DRUMDRUM

WW1WW1

DCOMERDCOMER

DEAEDEAE

FPUMPFPUMP

FVALVEFVALVESub-system

Equipment PIcontrolPIcontrol

Component LAGLAGaditivesladitivesl

pipepipe

valvevalve

pumppump themometerthemometer

Fig. 3. Concept structure map for training of power plant

The training course planner (TCP) is a compromise between an Intelligent Tutorial (ITS) system and a traditional CBT system. A course generated by TCP looks like a traditional structured course. However, this course is generated individually for every operator to achieve a certain training goal (a concept or topic that has to be learned). The course planner is an AI for planning system based on the well known algorithm for graph planning, AO [5]. Each course is generated in a dynamic way based on particular characteristics of the operator, and it can be changed also dynamically to reflect changes in operator model. During the presentation of the course to the opera-tor, if the operator answers the test items or simulator task correctly, i.e. demonstrates that she/he has acquired the concepts or skills, no changes to the course are necessary. However, if the operator fails to demonstrate knowledge of a concept, a re-planning of the course follows. The TCP generates a new sequence of concepts and teaching materials leading to the goal concept, starting from the current state of student knowl-edge as recorded in the operator model.


2.4 Operator Model

The operator model is used to adapt the intelligent environment (IE) to each specific operator. The operator model is built from observations that the IE makes about the operator. These can come in the form of responses to questions, answers to problems, behavior, etc. The operator model is divided into three subcomponents: cognitive component, the operator profile and affective component.

The cognitive component is the state of the domain knowledge of the operator. The cognitive component is constantly updated during the training/tutorial session as the cognitive state of operator changes. The cognitive component is an overlay model, where the operator’s knowledge is a subset of the concept structure. The cognitive model is active during the trainee session and collects all the information related to the performed of the training in the current session: instructional plan, errors, objec-tives requested, actions performed, etc.

The operator profiles are the relevant operator’s features which are important for the training process: capacity of cognitive development, motivation, learning style, expertise, and so on. The profile component is active during all the instructional proc-ess for each operator. It is update after each training session and it contains informa-tion relate to: operator curriculum, expertise, skills, learning characteristics, operator’s errors history, didactic material used with the trainee and a history of the whole in-structional process.

The affective component of the operator is modeled using a OCC cognitive model of emotion [6]. To determine the operator affective state we use the following factors: 1) operator personality traits, 2) operator knowledge state, 3) goals and 4) tutorial situation. The goals for our domain are: 1) to learn the topics related to the operation, 2) to perform tests and simulations successfully, and 3) to complete the experiment as fast as possible. According to the OCC model, the goals are fundamental to determine the affective state; we infer them by means of personality traits (profile).

The OCC model establishes emotional state as a cognitive appraisal between goals and situation. We represent this comparison with nodes goals and training situation, propagating evidence to nodes goals satisfied, and with these last nodes propagating evidence to the nodes affective state. According to the OCC model, the goals are fundamental to determine the affective state; we infer goals by means of personality traits and the knowledge state.

We based personality traits on the five factor model [7], which considers five di-mensions for personality: openness, conscientiousness, extraversion, agreeableness, and neuroticism. Currently, we use only two of them (conscientiousness, neuroticism) to establish goals because a relationship exists between these two dimensions and learning [8].

In order to know if a goal has been satisfied, we include the nodes goals satisfac-tion (one for each goal), and they are influenced by the nodes goals and tutorial situa-tion. The nodes goals satisfied represent the comparison between goals and situation as established by the OCC model. The nodes tutorial situation have the outcomes of the user actions, and their values are obtained from the performance of the student on the intelligent environment.


From the emotion set proposed by the OCC model [6], we consider four possible emotional states: joy, distress, pride, shame. We use these emotions because they emerge as a consequence of the actions of the students. A representation of the affec-tive operator model based on a probabilistic network [10] is presented in Figure 4.

The affective state is used in combination with the knowledge state (in the cogni-tive component) to select the “best” tutorial action based on the current course.

To complete the experimentas fast as possible

To perform experimentsuccessfully

To learn the topics relatedto the experiment

Neuroticism

Acquiredknowledge

Conscientiousness

To do experiment as fastas possible satisfied

To perform experimentsuccessfully satisfied

To learn the topics includedin experiment satisfied

Pride-ShameJoy- Distress

Personality traits

Goals

Goals satisfied

Affective state

Tutorial situationExperiment

resultsExperiment

duration

Knowledge

Fig. 4. Affective Operator Model

2.5 Operator Interface

The operator interface controls the interactions with the operator, including the dia-logue and the screen layout The main purpose of this module is to present to the op-erator the teaching materials in the most effective way. One of major challenge of this module is to take advantage of new technologies such as pedagogical agents, virtual reality, natural language processing, affective wearables, among others. The user interface must provide the motivation for the operator to continue. The application of new technologies is promising, for example the use of animated pedagogical agents give students the sensation of someone behind the program and it can be used trans-mit enthusiasm and motivate them. The use of virtual reality provides to operator a mock up of the real work environment. This last one enables to train operator how to perform a task, guide him in the work and show him, for instance, where the work tools are.


3 Conclusions

In this paper we have presented the architecture of an Intelligent Environment for training of Power Systems Operators. The aim is build an advance training system which includes: computer based training components labeled as Reusable Learning Objects; a learning object repository; concept structure map of the power plant do-main; a tutor module based on course planner and operator model based on cognitive and affective components.

We presented preliminary results for each component of the architecture; such as-RLO based on virtual reality, concept structure map for the operation of power plants, and cognitive and affective operator model. Our IE combines the advantages of clas-sical CBT with those of Intelligent Tutoring Systems, providing a general framework for training of power plant operators with reusable components, that can be adapted to individual needs. The proposed learning environment has several advantages: flexi-bility, it allows to consider different areas of power plant operation and different training requirements; adaptability, it can be adapted to generate instruction according to individual needs; and modularity, it can be easily extended to include more train-ees, and more experiments and other domains. The next phase is to integrate the indi-vidual components into a working environment, and validate it with actual operators.

References

1. Peter Brusilovsky and Julita Vassileva: “Course sequencing techniques for large-scale web based education”, Int. Journal Cont. Engineering Education and Lifelong Learning, Vol 13, No. 1/2, (2003) 75-94.

2. ADL. Sharable Content Object Reference Model Version 1.2:The SCORM Overview. Ad-vanced Distributed Learning, (2001) URL http://www.adlnet.org.

3. W3C. Extensible Markup Language (XML) 1.0. World Wide Web Consortium 2 ed. W3C Recommendation, (2000). URL http://www.w3c.org/TR/2000/REC-xml-2001006.

4. M. Perez, Virtual Reality: introduction, prototypes, applications and tendencies (In Span-ish): Boletin IIE, Vol. 28, No. 2, (2004), Abril-Junio.

5. Darwyn R. Peachey and Gordon I. McCalla: “Using planning techniques in intelligent tutor-ing systems”, Int. Journal Man-Machines Studies, Vol. 24, (1986) 77-98.

6. Ortony, A., Clore, G.L., Collins, A.: The Cognitive Structure of Emotions. Cambridge Uni-versity Press (1988).

7. Costa, P.T., McCrae, R.R.: Four Ways Five Factors are Basic. Personality and Individual Differences, 13(1) (1992) pp. 653-665.

8. Y. Hernández, J. Noguez, Enrique Sucar and G. Arroyo-Figueroa, “A probabilistic model of affective behavior for Intelligent Tutoring Systems”, A. Gelbukh, A. de Albornoz, and H. Terashima (Eds): MICAI 2005, LNAI 3789, (2005) 1175-1184.

9. J. Pearl: “Causality: Models, Reasoning and Inference”,Cambridge University Press, (2000).

Intelligent Environment for Training of Power Systems Operators

Documents

Transcript of Intelligent Environment for Training of Power Systems Operators