MAS-Based Intelligent Pressure Decoupling And Coordinate Control Of Gas Collectors In Coke Ovens

The coke oven is a nonlinear, time-varying, multi-variable, strongcouple-controlled plant with distributed parameters. It is difficult to set upaccurate mathematic models under conditions of several coke ovensparallel working due to complex coupling of the pressure of gas collectorsof coke ovens (GCCO), which brings large obstacles for the design ofcontrol system of GCCO. Stable and accurate pressure control of gascollectors is rarely achieved. The MAS(Multi-Agent System) is featuredas autonomous, distributed, coordinated, strongly adaptable, robust,reliable, and highly efficient to solve real problems. In this dissertation adecoupling and coordinated control structure and method based on MAStechnology with the background of intelligent pressure control of GCCOis proposed. The basic idea of the strategy is to take the whole controlsystem as a MAS and decompose goal of control into some sub-goals,then dispatch different Agents to implement these sub-goals, coping withintelligent decoupling and coordinated control problems with Agentreinforcement learning algorithm and realizing long-time stable controlfor the pressure of GCCO through integration of various software andhardware platforms.In this dissertation the MAS-based intelligent decoupling andcoordinate pressure control of the GCCO is discussed from three aspects:theoretic model and basic structure, decoupling and coordinated controlof the pressure of GCCO based on task decomposing and reinforcementlearning, the system implementation and application effect based onJADE platform and multi software and hardware integration.For such complex industrial process as the pressure control of theGCCO, a concept model based on MAS technology for distributedintegrated intelligent control system is proposed and the whole processfrom the design of structure to the commission of system is described inrespect of both space and time. The concept model can be used forguiding the design of intelligent decoupling and coordinated control. Thecontrol system is composed of human system integration unit, control unit and virtual simulation unit. Each unit consists of hierarchical agentsor agencies. Special agents and supervisory agent can combine to form anagency. The system is restructured and the mode of control is switched bychanging the active or inactive state of agents, and on this basis, thestructure of integrated intelligent coordinated decoupling control systemof the gas collectors of coke ovens is set up.According to the concrete object of intelligent decoupling control ofthe pressure of GCCO, the structure of distributed intelligent decouplingcontrol for gas collectors is set up, the real-time control Agent is realizedthrough fuzzy neural network. In order to solve the learning problemsunder dynamic and uncertain conditions, Agent reinforcement learning isintroduced, fuzzy reinforcement learning based on GA and multi-Agentcooperation are studied. A coordinated system and algorithm based onfuzzy reinforcement learning of distributed co-evolution is proposed, theoptimal rules with efficient decoupling performance are obtained throughcontrol Agents' coordinate reinforcement learning, and distributeddecoupling control of GCCO is realized. The technics model of thepressure of gas collectors is discussed and simulation software isdeveloped on Matlab, the simulation results show that the proposedmethod is efficient.In order to solve the control problems of strong disturbance producedby high pressure ammonia and variety of external environment, anintelligent multi-level coordinated control strategy for the pressurecontrol of gas collectors based on reinforcement learning is proposed.Multi-level coordination architecture is structured by the gas collectorsAgency, primary cooler Agency and blast blower Agency. Thecoordination of agents is employed to solve the set point of pressurebefore the primary cooler and the blast blower in order to keep the controlvalves in sensitive position in every level. The system can be switched todifferent modes through the state change of agents in order to operate inrapidly time-varying environments. For the control under impulse of highpressure ammonia, a distributed reinforcement learning structure ofactor-critic which is realized by the TS type recurrent fuzzy network(TSRFN) is adopted. The agents in system are optimized coordinately through the distributed reinforcement learning algorithm. The simulationresults show that the proposed control strategy can successfully solve thecoordinate control problem of the GCCO with the strong disturbanceproduced by high pressure ammonia.The proposed structure and algorithms are realized based on theJADE (Java Agent Development Environment) platform which is incompliance with the FIPA specifications through the Agent-orienteddevelopment method and integration of universal industrial controlsoftware and PLC (Programmable Logic Controller). The implementationtechnology of control Agent, control algorithm and communicationamong Agents based on JADE platform are discussed, the intelligentcontrol software of the pressure of GCCO based on MAS is developed.Application results show that long-time and stable pressure control ofGCCO is realized.The application of MAS-based intelligent control techniquesimproves the global control level of GCCO, restrains the influence ofcoupling and disturbance produced by high pressure ammonia, protectsthe environment, reduces the labor intension of workers, promotes theoutput and quality of chemical production, prolongs the life of coke ovens,and achieves distinct economic benefit and social benefit. Meanwhile aset of practical efficient method of design and implementation is providedfor the MAS-based control of complex industrial process.