Research On Online Control Of CSTR Process Based On Adaptive Dynamic Programming

Continuous stirred tank reactor(CSTR)is one of the most common industrial equipment in petroleum and chemical industry.It has been widely applied in pharmaceutical synthesis,fermentation industry and additive preparation.In industrial production,especially the production of chemicals,chemical reactions in the reactor are accompanied by a large number of exothermic or endothermic chemical reaction,which makes the temperature in the reactor increased or decreased sharply in a short period of time,so the reactor temperature control has been a hot research topic for experts.When the controller is unable to control the temperature accurately,many factories will use artificial control,which will increase the labor cost,but there are also serious security risks.Lead to reactor temperature control difficult is caused not only by the endothermic and exothermic chemical reaction,uncertain factors in the production process(such as environmental impact or external interference)makes CSTR often exhibit strong nonlinear,strong,large time delay and strong coupling,the properties of guided process model induced by reactor is difficult to establish or the low precision,so as to control the great trouble.In this paper,petroleum engineering filtration reducing agent carboxymethyl starch sodium,(CMS)production process as an example,based on the domestic and foreign experts and scholars on the modeling method of using data mining driven CSTR model,and the use of adaptive dynamic programming(ADP)design of control strategy,focuses on the research of modeling and the temperature control problem CSTR.The main contents of this paper are as follows:(1)A data driven model of CMS production process is established.By analyzing the CMS production process and the working characteristics of CSTR,selecting the key factors affecting the temperature of the reactor,and using Elman neural network to memorizing the historical data,a CSTR model based on data driven modeling is proposed.Further,considering the influence of neural network weights of each layer of neurons weighted calculation,the Elman model optimizing weights method using unscented Calman filter(UKF),update gradient neural network inherent decline,many off-line learning so that the accuracy of the model is higher.(2)Using ADP to control the temperature of CSTR.The Elman-UKF model replaces the model in ADP network,by comparing the heuristic dynamic programming(HDP),dual heuristic programming(DHP)and globalized dual heuristic programming(GDHP)three ADP basic structures,fully consider the influence,control accuracy and computation time of the controller GDHP is chosen as the structure of the CMS production process control.The evaluation module of GDHP is the double output of the cost function and the co function,which has the best global nature,but there is the problem of long computing time.(3)Using event triggering GDHP to solve the problem of long GDHP computing time.The event triggering conditions are set first,and the control structure of the event triggering GDHP is designed.This control scheme only takes system sampling under the violation of specific rules.The evaluation module and execution module are updated only under event triggering,and the rest time is maintained by the retainer.The event triggered GDHP can reduce the system sampling amount in a large amount,thus reducing the amount of communication and calculation of the system.(4)Using interval GDHP to enhance the robustness of the control system.Considering the imprecision of the industrial data,the input and output variables of the model are transformed into interval variables,and the interval neural network model of CSTR is established based on the interval algorithm.On this basis,using the interval control strategy,when the controlled variable exceeds the preset range,using two single point controller each switching of the controlled variable quickly and smoothly back to the range,to ensure stable production process at the same time,reduced to the maximum extent during the execution of the action times.