Q-Learning Based Control Of Roller Kiln Temperature

Lithium-ion batteries,as the main power sources of battery electric vehicle,are receiving more and more attention with the rapid expansion of the new energy automobiles market.Cathode materials are the key components of LIB,and Li Nix Coy Mn1-x-y ternary cathode materials are considered to be one of the most promising cathode materials for the reason of low cost,high specific capacity,good cycle performance and excellent thermal stability.The main method of industrial large-scale preparation of ternary cathode materials is high temperature solid phase method whose core is roller kiln sintering process.Roller kiln temperature is the most critical monitoring parameter during sintering process,control kiln temperature precisely is extremely important to reducing energy consumption,improving materials performance and heightening unity of products.However,the difficulties such as unknown system dynamics,severe temperature coupling among different regions in roller kiln and frequent disturbance make it very difficult to control kiln temperature accurately.Therefore,it is urgent to study control methods of roller kiln temperature.The main work of this paper is as follows:Firstly,isolated region temperature model cannot reflect severe temperature coupling among different regions in roller kiln which causes the problem of inaccurated modeling,to solve this problem,a new roller kiln temperature modeling method accounting effects of different regions temperature coupling is proposed in this article.First,by analyzing the mechanism of thermal convection of atmosphere and thermal radiation of silicon carbide rod,the factors causing heat exchange among adjacent regions in roller kiln are summarized.Then,the roller kiln temperature interconnected model whose interconnected terms is constructed by heat exchange among different regions is established based on conservation of energy and temperature-energy formula.Secondly,a roller kiln temperature optimal control method based multistep Q-learning is proposed to overcome the controlling difficulties caused by temperature coupling among different regions and optimize control performance.First,by centralizing the system dynamics and interconnected terms,roller kiln temperature optimal control problem is turned to the optimal control problem of a centralized large-scale systems.Then,the multistep Q-learning algorithm is applied to learn the optimal temperature control policies.Because of the temperature coupling among regions,control policies of any region is interconnected with other regions.Thus,a novel policy improvement method is proposed which make the updated control information of other temperature regions can be applied in time when improving policies of any temperature region.Simulation result shows that roller kiln temperature can reach the set under proposed control method,and control performance is optimized.Thirdly,In order to deal with the frequent disturbance,an decentralized H_∞control method for roller kiln temperature via policy iteration Q-learning is proposed.First,decribing the H_∞control problem of roller kiln temperature.Then,a bounded function is constructed to describe the maxium effects caused by temperature coupling to the control performance of regions in roller kiln.And the minimax design problems of different regions can be established based on bounded function above,thus,the large-scale H_∞control problem of roller kiln is turned to small-scale H_∞control problem of regions.The target roller kiln temperature H_∞control policy can be obtained by solving minimax design problems above,so the control method is competeley decentralized.Then,the policy iteration Q-learning algorithm is applied to learn roller kiln temperature decentralized H_∞control policy.Simulation result shows that the proposed control method can not only control roller kiln temperature to reach the set point precisely,but can also overcome the negative effects caused by disturbance.