Cooperative Validation in Distributed Control Systems Design (PDF ...

Cooperative Validation in Distributed Control Systems Design Dariusz Choinski, Mieczyslaw Metzger, Witold Nocon, Grzegorz Polakow Faculty of Automatic Control, Electronics and Computer Science, Silesian University of Technology, ul. Akademicka 16, 44-100 Gliwice, Poland {dariusz.choinski, mieczyslaw.metzger, witold.nocon,grzegorz.polakow}@polsl.pl

Abstract. The team of engineers designing and implementing distributed control system software must communicate within a multidisciplinary environment. One of the main problems is the interaction between hardware and software solutions. Software project presumptions may not be based only on minimum hardware requirements and on technology rules. Modern distributed control systems embrace all aspects of a complex and widespread object. Hence, every modification within any discipline requires interference into the system and validation of its new features, which in turn constraints effectiveness of designing. This paper discusses how to increase effectiveness and speed up validation, in a standardised CAD environment, by using: MultiAgent System in order to limit the number of interactions between particular subsystems, ontology for assisting topology description and properties of system entropy for assessment of introduced solutions. The proposed system was implemented and worked out in a biotechnological pilot plant. Key words. Collaborative design, multi-agent systems, knowledge ontology, web environment for collaborative working, multiple location collaborative design, industrial applications

1. Introduction Design of modern industrial plants creates problems that arise from the fact, that the automation and information control system running this plant needs to be taken into account. Such a system usually comprises a great amount of different components such as control instrumentation, control software and communication networks. Design and integration of the process control system and finally an operation of the process during normal exploitation as well as in emergency situations are difficult tasks. Another problem arises from a significant difference between obsolescence times of the mechanical and electronic plant components (the instrumentation for control and information, should be frequently upgraded). Therefore, during normal exploitation of the whole plant the control and information instrumentation should be redesigned. A multi-agent-based system for cooperative design, validation and operation of industrial processes proposed in the paper can be very helpful for these tasks. The proposed system includes several modern ideas such as multi agent

2

Dariusz Choinski, Mieczyslaw Metzger, Witold Nocon, Grzegorz Polakow

systems (MAS) – see for example [1][2], control algorithms of hybrid systems [3],[4],[5] hybrid I/O automata [6], knowledge ontology [7], cooperative systems for design [9],[10],[11],[12] as well as network-based collaborative design systems [13],[14],[15].

2. Problem under consideration Our research so far, using a biotechnological pilot-plant as an example [16], has proven, that the control system may be represented by a hybrid system model. This system consists of an automaton having a finite number of states. Transition conditions between those states are described by two sets defining controllable and uncontrollable events. Control of a system modelled in this way is realized within MAS. The control agent tries to maintain the given state despite disruption caused by some uncontrollable events, while the supervisory agent tries to change the current state into another desirable state, by applying a sequence of controllable events. Any transition functions that are missing or not specified yet, may by developed by an expert. Because of technological constraints and limited capabilities of measurement, control and powering devices as well as of the information structure of the distributed communication equipment, the system has been divided into subsystems. This division is based on ontology that takes the semantics used in CAD systems into account. Apart from the subsystems, the system possesses defined functions, the taxonomy of which is based on phenomenological models. The main technological concepts are based on those phenomenological models. Architecture of the proposed system is presented in Fig. 1. Subsystems hierarchy

Technology rules

Web Engineering constraints

Cooperative validation

Subsystems Subsystems Subsystems

Component ontology

Engineering

Corrected functions

Taxonomy

Functions

System architecture

Functions

System software

States

Corrected states transitions

Fig. 1. Concept of the hybrid system.

Expert validation

Expert validation

Cooperative Validation in Distributed Control Systems Design

3

An impartial assessment of the distributed control system operation is very difficult and depends on the goal of this assessment. One of the important aspects is determination, whether the system’s user, who is not familiar with control theory, may utilize all the capabilities of the technological object. That is in relation to the possibility of access to different states by a minimum number of intermediate states and remembering the once-carried-out organization of subsystems. Such operations must be robust, with means avoiding the system to be moved into a state, from which the currently used controls may be insufficient for a save return of the system into the previous state.

2. Developed system description As practice of automation of biotechnological [16],[17] processes shows, automated plants may be represented as hybrid state machines [6] i.e. state machines augmented with sets of differential equations. Such hybrid systems are described with two sets:  Ω – set of continuous state variables. Range of those variables depends on the specifics of the considered system i.e. process constraints, measurements, activator’s capabilities, etc.;  Φ – set of events conditions enabling transitions between states. The Φ set can be divided into two subsets:  Φu – subset of uncontrollable events – transitions caused by internal process behaviour or external environment, the control system can not prevent nor cause;  Φc – subset of controllable events – transitions which can be induced and/or prevented by the control system, so those transitions can be used to perform a control task. Changes of system’s state can happen as discrete events when certain conditions defined in Φ are fulfilled or as a continuous trajectory in a space state according to the differential equations (which are a mathematical description of an object-specific function f of inputs and outputs of the system). Control of a hybrid automaton is a complex task, which is best performed with multi-agent systems. The proposed structure of the control agents, which proved to be effective, is hierarchical and consists of three layers: control, supervisory and expert. Each layer performs tasks at a designated level of abstraction, and actively cooperate to achieve control goal in a robust manner. The lowest layer consists of Control Agents. A Control Agent is bound directly to control instrumentation of the controlled plant. It implements all the required control algorithms, takes care of presumed sequential production cycles and performs all the other functions of typical control task (e.g. emergency shutdowns). All control layer agents work in time-driven mode to be able to implement typical time determined control algorithms with signals sampling. Uncontrollable events are disturbances Control Agents try to cope with by standard means of control algorithms. The middle layer of the control system structure is formed by Supervisory Agents. Each agent supervises a number of Control Agents and monitors quality of control in

4


a broader scope than Control Agents are able to. Supervisory Agents are capable of more advanced state recognition and trajectory planning, so in case of worsening control quality, a Supervisory Agent may decide that it is desirable to switch the system into some Ωi state (or even through some Ωi, …, Ωn states sequence planned ahead) to fulfil a given control task. The plan is then performed by the specific Control Agent as a proxy. At last, the top layer contains Expert Agents. Expert Agents are a system’s interface to external sources of knowledge such as human and/or artificial overseers and experts. Expert Agents’ role is to supply Supervisory Agents with additional knowledge on processes in the system. Data which is missing at the moment but is required to fulfil control task may include for example transfer functions or possible sequences of transitions between states of the automaton. It should be noted that implementation details of control system (i.e. power supply lines, control loops realisation, effectors physical location, etc.) should not burden remote experts. It is desired for them to work on abstract knowledge, not the specific realisation. This makes it possible to implement the idea of universal experts for a specific process class. Experts are able to serve their general knowledge to many physical instances of the class, if only Expert Agent supports data exchange with spatially distant location over long range communications link (including Internet network). Both supervisory layer and expert layer work as event-driven, which in consequence means, that the multi-agent layered structure decomposes control system into two parts: time-driven (Control Agents) and event-driven (Supervisory and Expert Agents). Such decompositions allow for separate analysis of both modes, which can significantly decrease complexity level of the control system. To make design, verification and maintenance phases of the system’s cycle of work easier and more regular, all of the system elemental components should be logically grouped into subsystems. Such classification can be done according to the IEC 61346 standard, which defines rules for structuring system’s components into hierarchical subsystems and referencing them. The norm proposes three kinds of hierarchies, according to components’: function, location, and product. In effect, system’s logical organisation consists of three different trees with system’s components as vertices, where each of the components belongs at the same time to each of the trees. Edges of the trees are designated by membership relations (smaller subsystems forming larger ones). In distributed control systems, usual function criteria are for technical reasons: power supply, process, and control task. In a real complex control system many subsystems of varying complexity levels can be distinguished on various levels of abstraction. Additionally, each of the control instruments and each of the subsystems belong to many structures depending on the assumed grouping criterion. Independently of subsystems distinction, whole control system has functions defined. Those functions are consequence of control systems designed task and their taxonomy is determined by phenomenological models of processes in the specific controlled plant. A proposed approach to this problem is visualised in figure 2. The figure systemises the knowledge of design, verification and maintenance processes of the complex distributed control system and digests this knowledge into an algorithm which may be a base for development of software applications supporting this procedure. Upper part of the figure represents the mentioned earlier process of logical


5

decomposition of the complex system’s structure into elemental components and subsystems according to the IEC 61346 norm. Reference designation sets for function-related, location-related and product-related obtained in this process are a base for constructing ontology for determination of subsystems. The ontology is expressed as a Calculus and describes topology of the system [18]. Fenomenological models

Boundary conditions

Process under design IEC 61346 Set of the reference designations

Subset of function hierarchy

Subset of location hierarchy

Subset of product hierarchy

process power supply

Component ontology

control

Subsystem taxonomy Hybrid system Functions

Control System Software

Control System Architecture

Deterministic finite state automaton

OPC Self-orginizing Relational Data Base WEB XML

XML

XML

Mediator Mediator Mediator Observed Observed Observed finite finite Observed Observed functions Observed finite states functions states functions states Corrected state Corrected Corrected state Corrected state Corrected Corrected transitions functions transitions functions transitions functions

Fig. 2. Architecture of hierarchical Web based system implementation.

This structure of subsystems hierarchy taken together with boundary conditions of the systems’ variables defines the system as a deterministic finite state automaton. On the other hand, the ontology is also taken into account when definition of functions taxonomy is built. Taxonomy of functions performed by the system is derived from phenomenological models of processes being automated. Finally, both state automaton definition and set of system’s functionalities define the architecture of hybrid system which is the final product of the design process. There is a need to verify the resulting functionality (being an effect of subsystems configuration and system’s states) of the complex distributed control system for a

6


compliance with technological presumptions. It can be done by employing remote experts connected to the system by means of Expert Agents’ interfaces. Remote experts are able to use their knowledge (e.g. experience, mathematical simulations and phenomenological models) to design (at design phase) or modify (at maintenance phase) taxonomy of system’s functions. The proposed interface for Expert Agent is Web-based, because of the particular popularity of the Web environment, thus greatly simplifying the process of remote expert employment. As is shown in the figure 2, it is assumed that the system is automated using OPC-capable instrumentation. Thanks to this assumption, the system is connectible with OPC-standard compliant software and can be linked to many external applications, and such broad possibilities of interconnectibility are highly desired when implementing Expert Agents’ interfaces. To make Web–based application capable of data interchange with industrial–grade software, specialised application was developed. The self-organisational relational data base [19] is a data mining application actively probing designed control system and storing the variables and their historical trends in Web-standards compliant MySQL-driven database. The database is accessed by a custom web server-side application, which transparently encodes and decodes incoming queries and outcoming replies using the XML notation. Such encoding enables quick and easy deployment of Webservices enabling access to the control system’s internal information. End client application for system–human expert interaction is developed using Flash technology and can be embedded in webpage and executed in nearly any Web browser. The Flash application uses the Web service to communicate with the system (interface part of the expert’s application is designated as a Mediator). The role of the remote expert is correction of the possible state transitions and correction of system’s functions. The task is performed based on the data received form the database through the Expert Agents and Mediator interfaces. The structure of the presented system allows for easy and quick connecting of multiple remote experts. However, in case where multiple remote experts are employed in one control system, inter-expert cooperation becomes problematic. There are a few existing platforms for multiple software agents’ cooperation (e.g. FIPA Foundation) but they focus mainly on technical issues like software development, data flow and organisation, and software threading (control flow). Although these methods are known, the problem under consideration is non-trivial and needs to be researched as a subject of cooperative problem solving before it can be physically distributed among multiple remote Expert Agents. As Smith and Davis suggest [20], cooperative distributed problem solving consists of three steps: problem decomposition, sub-problem solution, and solution synthesis. In the system under consideration second and third steps are already implemented. Sub-problem solution consists of correcting states transitions sequences and functions by a remote expert, based on his knowledge, observed finite states and functions (received from system’s database). Solution synthesis is a simple process of storing corrected data in the database. However, first step of problem decomposition requires additional effort. It is required to describe the control system in such way, that each of the experts works in his own problem domain without interfering with other experts. Since natural division of problem in hybrid systems is into automaton’s particular states, it is desired that each system’s state is as independent from other states as possible [5]. When this assumption is fulfilled, each of the experts may focus on a specific subset of the states


7

space, and modify the subset without disturbing other experts. To enable this possibility and to maximise control capabilities of the system, innovatory definition of the control system’s design and verification goals are introduced in the following section.

3. Ontology-based subsystem topology semantics Even the not complicated system design has a multilevel tree of reference designations, which is usually incomprehensible without associated diagrams. The main aim of the proposed ontology is to make physical meaning of the subsystems as a collection of formal axioms based on primitive relation incorporated only in the reference designation tree and the distances in hierarchy The set of reference designations prepared in accordance with the IEC 61346 standard is a basis for the division into subsystems according to the component ontology. The mereology starts taking a relation ‘Cx,y’ to express that individuals x and y are connected, as was introduced by Clarke [18] as a Calculus. This Calculus may state ontology describing topology based on reference designation according to IEC 61346 For particular individuals we can present the following mereological definitions (where superscript ‘*’ is for all references designation, ‘-‘ is for subset of product, ‘+’ is for subset of location, ‘=’ is for subset of process function hierarchy and ‘==’ is for subset of power supply function hierarchy respectively): D 1: Common root in hierarchy of technological equipment set: Cx+,y+ D 2: Source of the product information connected to technological individual: Cx+yD 3: Common source of power supply: Cx==,y== D 4: Source of the power supply connected to technological individual: Cx==,y+ D 5: Interconnected technological functions: Cx=,y= D 6: ‘x is a Part of y’: Px * , y *  (z * )(Cz * , x *  Cz * , y * )

 

D 7: ‘x Overlaps y’: Ox * , y *  z * Pz * , x*  Pz * , y *



D 8: Independent functions: ‘x is Discrete from y ’: DRx  , y   Ox  , y  D 9: Connected to externally subsystem: ‘x+ is Externally Connected to y+’: ECx  , y   Cx , y   Ox  , y  D 10: Remote subsystem function: ‘x= is Externally Connected to y+’: ECx  , y   Cx , y   Ox  , y  D 11: Local subsystem function: ‘x= is Tangential Part of y+’: TPx  , y   Px  , y   z  ECz  , x   ECz  , y  D 12: Supervised function: ‘x= is Non Tangential Part of y+’: TPNx , y   Px  , y    z  ECz  , x   ECz  , y  Fig. 3 presents a subsystem composition based on this ontology. ==

==

 

 





8

Dariusz Choinski, Mieczyslaw Metzger, Witold Nocon, Grzegorz Polakow Subsystem Z

=Z1

Subsystem X

NTP O -X1

TP

=X1

EC

P +X2

+X1

Subsystem Y

P -X2

O

+Y1

==X2

== Y1

DR component ontology Location X1 is a part of location X2 in subsystem X State variable X2 is a part of location X1 Location X2 is connected to location Y1 in externally subsystem State variable X1 overlaps location X1 Function X1 is a local function of location X1 Function Z1 is a supervision function in location X1 Power supply X2 overlaps location X2 Power supply X2 is a independent to power supply Y1

Fig. 3. Ontology based subsystem composition

3. Entropy functions-based validation Variables characterizing the system in general and the individual subsystems in particular, may be divided into intensive variables and extensive variables. Extensive variables ex depend on the system’ scale, while intensive variables in are independent of the system’ scale. Any subsystem state may be characterized by values of intensive variables that depend on each other. Balance equations that serve as a basis for phenomenological models describe flux flows of the extensive variables. Those flows are forced by the difference in intensive variables that are in this case the moving force of the process. In a controlled system, some intensive variables posses defined ranges of values, while other intensive variables, with unspecified limits are observed or bindings of those variables with other intensive variables are known, usually by applying Le Chatelier-Braun principle. The relation between flows and forces characterizes the system kinetic. Conditions for reaching a steady state of the subsystem are specified by extensive variables bound with each other in balance equations. Because the intensive variables used in control of the process are independent of the system scale, only binding those variables with extensive variables enables the steady state of the system to be determined. It is especially important for biotechnological objects, where stable values of intensive variables (substrate concentration for example) may


9

correspond to unstable values of extensive variables (biomass quantity for example). Any intensive variable acting between two subsystems must obviously posses the same value. MAS for the hybrid control system enables determination of all reachable states. For every subsystem, the state is specified by the set of extensive variables that are part of the state equations of the base class and by intensive variables with limits specified. For every subsystem a probability pi of reaching the outcome of the state characterized by  i a posteriori may be determined and the discrete form of Shannon information entropy function [21] my also be determined:

S  p  

N

 p log  p  i

2

(1)

i

i 1

where: N – number of states for the subsystem, and p={pi}. A normalization condition is defined as: N

p

i

1

(2)

i 1

For an isolated subsystem, the possibility of reaching particular states with the same probability is connected to entropy maximization. Because of the hierarchical structure of MAS, the Control Agent maintains the states desired by the Supervisory Agent. In order to check the behaviour of the particular state i, minimization of partial Kullback-Liebler cross-entropy function is used.

DiKL  pi log 2

pi qi

(3)

where: qi is the a priori probability of income for state i. Fig. 4 presents values of the DKL function for particular states in case when reduction, maintaining and production of takes place for the subsystem. In should be noted, that when entropy is maintained, the DKL function has low values in a wide range of states. In addition, averaging values of states increases entropy. In case when entropy increases, it is possible to find a set of states in the region of minimal values of DKL. This region however is divided by states with low probability of occurrence, hence finding such a minimum requires precise knowledge about constrains, because change in those constrains may be burden with a penalty for exceeding the regions with low probability of occurrence. Opposite situation occurs when entropy is reduced. In such a case the region of high probability is very small, while the low probability region is large. Such situation may for example take place when fault or wrong state changing function is determined. For application of Kullback-Liebler [22] cross-entropy in the cooperative validation on-line estimation this entropy value is necessary. When the Gaussian distribution of subsystem states probability is assumed, for the window of P observed income states {q1,..,.qi,...,qP} enforced by Control or Expert Agent and outcome states {p1,..,.pi,...,pP} by Control Agent realized, standard deviation  can be estimated [23].

10


Fig. 4. Three dimensional plot of the partial Kullback-Leibler cross entropy used to examine the behaviour of each state of particular subsystem.

The cross-entropy is an unsymmetrical measure and subsystems have a variety of mutual interconnections dynamically changing during designing and validation, hence estimator of entropy value should avoid this disadvantage and is defined as follows:

Dˆ PKL

 ˆ p log 2   ˆ q  

2

  ˆ   q   ˆ p   2

2

  ˆ q  log 2    ˆ p   

2

  ˆ   p   ˆ q   2

2

    1

(4)

In this paper a system is considered, that exchanges entropy with the surroundings in a necessary way. A system consisting of many mutually connected subsystems may serve as an infinite source of entropy for every subsystem. In such a case, entropy may be dissipated and the system may transit into a state of irreversible processes. For this case it should be checked whether the subsystem is in a stable state with a concurrent minimization of the entropy value. Hence, despite decreasing the probability of particular states occurrence, the subsystem is in a steady state, in which there are bindings between different state variables. For this situation, a Onsager condition [24] may be used. If it is possible to find the following dependence, for the intensive variables moving forces and fluxes of extensive variables:

 iex 

L  ij

in j

(5)

j

where: L is a matrix of phenomenological kinetics parameters, than it may be assumed, that the subsystem is in a steady state, if:


L  LT

11

(6)

Such situation testifies that an irreversible process has taken place, hence is it not possible to change this state by the currently available control variables that are intensive. Cooperation of other subsystems is necessary in order to reverse the moving forces. The presented equations can be used only for predicting the possibility of equilibrium state with irreversible process. During designing, process technology rules are elaborated for avoiding such states. However, dynamically configured subsystems can overlap state variables with symmetric matrix of phenomenological kinetics parameters. This requires very expensive validation of subsystem functions related to the technology rules. For that reason, especially during cooperative validation, automated procedures are advisable. Such automated procedures are based on component ontology and taxonomy of function performed by the particular subsystem. Taxonomy is derived from phenomenological process models and boundary conditions, both based on technology rules. Component ontology ensures description of multidimensional relationship subsystems describing mutual hierarchy of products, locations, process, power supplies, and controls.

4. Concluding remarks - application for the biotechnological pilotplant For over three years, a biotechnological pilot-plant has been operated continuously at the Faculty of Automatic Control, Electronics and Computer Science. It serves as a platform for investigations regarding activated sludge processes in aquatic environment. One of the processes that may be investigated is bioaugmentation of biomass for different purposes. Hence, depending on the goal of research, the structure of the pilot-plant is changed. Therefore, operation and maintaining of this pilot-plant is a difficult task, and participation of external experts is crucial. The proposed system evidently enables these tasks. The proposed system has been beneficially functioning for two years. Acknowledgements. This work was supported by the Polish Ministry of Scientific Research and Information Technology.

References 1. Wooldridge, M., Jennings, N.R.: Intelligent agents: theory and practice. The Knowledge Engineering Practice, Vol. 10, No.2 (1995) 115-152 2. Jennings, N. R., Sycara, K., Wooldridge, M.: A Roadmap of Agent Research and Development. Autonomous Agents and Multi–Agent Systems 1, Kluwer Academic Publishers Boston (1998) 7–38

12


3. Leduc, R.J., Lawford, M., Dai, P.: Hierarchical Interface-Based Supervisory Control of a Flexible Manufacturing System. IEEE Transactions on Control Systems Technology 14, No. 4 (2006) 654-668 4. Cassandras, C. G., Pepyne, D. L., Wardi, Y.: Optimal control of a class of hybrid systems. IEEE Transactions on Automatic Control 46, No. 3 (2001) 398–415. 5. Schaft van der, A.J., Schumacher, J. M. Compositionality issues in discrete, continuous, and hybrid systems. International Journal of Robust and Nonlinear Control 11, No. 5 (2001) 399-539 6. Lynch, N., Segala, R., Vaandrager, F.: Hybrid I/O Automata. Inf. and Comp.185 (2003) 105–157 7. Guarino, N.: Understanding, building and using ontologies. Int . J. Human – Computer Studies 46 (1997) 293 – 310 8. Luo, Y., Dias, J.M.: Development of a cooperative integration system for AEC design. CDVE2004, Springer Lecture Notes in Computer Science, Vol. 3190 (2004) 1-11 9. Roller, D., Eck, O., Dalakakis, S.: Integrated version and transaction group model for shared engineering databases. Data & Knowledge Engineering 42 (2002) 223–245 10. He, F., Han, S.: A method and tool for human-human interaction and instant collaboration in CSCW-based CAD. Computers in Industry 57 (2006) 740–751 11. Anumba, C.J., Ugwu, O.O., Newham, L., Thorpe, A.: Collaborative design of structures using intelligent agents. Automation in Construction 11 (2002) 89–103 12. Korba, L., Song, R., Yee, G., Patrick, A.: Automated Social Network Analysis for Collaborative Work1. CDVE 2006, Springer Lecture Notes in Computer Science, Vol. 4101 (2006) 1-8 13. Huang, G.O., Huang, J., Mak, K.L.: Agent-based workflow management in collaborative product development on the Internet. Computer-Aided Design 32 (2000) 133–144 14. Huang, G.: Web-based support for collaborative product design review. Comp. Ind. 48 (2002) 71-88 15. Zhao, G., Deng, J., Shen, W.: Clover: an agent-based approach to system interoperability in cooperative design systems. Computers in Industry 45 (2001) 261–276 16. Choinski, D., Nocon, W., Metzger, M.: Real-time control strategy for sequentially operated continuous WWTP. Proceedings of the 10-th IEEE International Conference on Methods and Models in Automation and Robotics, Miedzyzdroje (2005) 451-456 17. Davidsson, P., Wernstedt, F.: Software agents for bioprocess monitoring and control. Journal of Chemical Technology and Biotechnology, Vol. 77 (2002) 761-766 18. Clarke B.L.: Individuals and Points. Notre Dame J. of Formal Logic, vol. 26, No. 1 (1985) 61-75 19. Choiński, D., Nocoń, W., Metzger, M.: Multi-agent System for Hierarchical Control with Self-organising Database. The 1-st KES Symposium on Agent and Multi-Agent Systems – Technologies and Applications, Springer Lecture Notes in Computer Science, Wroclaw (to appear in May 2007) 20. Smith, R.G., Davis, R.: Frameworks for cooperation in distributed problem solving. IEEE Transactions on Systems, Man and Cybernetics, 11(1), (1980). 21. Shannon, C.E.: A Mathematical Theory of Communication. The Bell System Technical Journal, Vol. 27 (1948) 379-423, 623-656 22. Niven, R.K.: The constrained entropy and cross-entropy functions. Physica A 334 (2004) 444–458 23. Chendeb, M., Khalil, M., Duchêne, J.: The use of wavelet packets for event detection, in: Proceedings of the 13th European Signal Conference EUSIPCO, Antalya, Turkey, 4-8 September 2005 24. Tsirlin, A. M.: Optimal Processes in Open Controllable Macrosystems. Automation and Remote Control, Vol. 67 (2006) No. 1 132-147