Decoupling Control Of Gas Collector Pressure Of Coke Oven Based On Reinforcement Learning

With the development of coke-chemical industry, how to doeffective control on gas collector pressure has become a bottleneck whichrestricts the development of the industry. In coking process, the stabilityof gas collector pressure influences the coke quality, the life-time ofovens and the producing environment, and its control have a direct impactto the operating condition of the whole coke oven system. However, thecontrol system of gas collector pressure is a complicated multivariable,nonlinear, time-varying and big time-delay control object, which havemany disturbance factor and the coupling phenomenon is serious.Therefore, it is difficult to establish a mathematical model to reflectworking situation of the control system of gas collector pressureaccurately and obtain the desired control effect using traditionalautomatic control methods.In order to solve the above problems, the research purpose of thepaper is to adopt advanced intelligent control strategy based on FuzzyQ-learning algorithm to effectively control gas collector pressure andsolve the coupling problem of pressure in the process of gas gathering, sothat gas collector pressure can be stable in the scope-assigned, and it alsocan ensure the normal production and operation of coke oven. At thesame time, PLC control technology, WinCC configuration software andOPC technology are used to design an intelligent control system of gascollector pressure, which integrates Fuzzy Q-learning algorithm, so as toimprove the production efficiency and achieve energy-saving andemission reduction.The process of coke oven gas gathering and control requirements areanalyzed in the paper, and find out the relevant factors and couplingrelationship which influence the pressure. On this basis，a dynamicmechanistic model of coke oven gas collector is established. The keyparameters such as butterfly value opening and gas production areanalyzed by model simulation. The simulation laid the foundation for thestudy on control strategy.In the control strategies, the fuzzy control strategy based on Fuzzy Q-learning algorithm is designed, which based on the related theory ofreinforcement learning and the fuzzy control theory. And the problem thatfuzzy control rules are difficult to obtain is solved by strategy. Moreover,the strategy will map continuous state space of pressure values tocontinuous action space of butterfly value opening by fuzzy inferencesystem, and then get a complete decoupling control rule base through theonline study. The rule base can provide prior knowledge for butterflyvalue’s action selection, so as to achieve the decoupling control of gascollector pressure. MATLAB simulation results show that the controlstrategy is effective.According to the system control strategy, the hardware and softwareof intelligent control system are designed in the paper. In the hardware,the system selects S7-300PLC to collect field data and control fileddevices. In the software, Fuzzy Q algorithm program is implemented inthe MATLAB platform, and PLC fuzzy control program is realized, andalso the user interface of the monitoring is designed by using WinCC. Inthe meanwhile, OPC communication, the Industrial Ethernet and Fieldbustechnologies is adopted to realize the communication and data transfer onthe entire system.