System Integration And Process Optimization Based On Q-learning

In recent years,with the rapid development of information technology,computer science has been applied in many fields,or has become a cross-discipline related to other fields,which is no exception in the field of chemical industry.Nowadays,chemical engineering design is increasingly inseparable from the assistance of computer software and technology.However,in the field of chemical process,the commonly used simulation program has a variety of limitations.In the process of putting the theoretical model into the actual production,the complexity of equipment and construction,the specificity of debugging and operation,are often difficult to be calculated by the program and cannot meet all the needs of the process.On the basis of existing application software,it is not easy to implement and unstable to extend the functions of other application software.Therefore,in order to integrate the excellent functions of various application software to form a new and powerful application software without spending too much time and money,it is of great significance to integrate and redevelop the software.Aiming at this problem,based on the existing technology,Python language and Aspen Plus software are employed,combines optimization algorithm and reinforcement learning algorithm Q-learning for secondary integration,and develops a chemical process optimization system.The system realizes the interaction between Python and Aspen Plus,integrates univariate search technique,NSGA-II algorithm and Q-learning algorithm,so that the computer software is more suitable for the actual chemical engineering design.The work of this paper is as follows:(1)The interface connection and information exchange of Aspen Plus simulation software are realized by using Python programming language.A chemical process optimization system is developed on this basis.The system integrates univariate search technique,multi-objective genetic algorithm and Q-learning algorithm.It can also be updated to integrate other optimization algorithms to cope with different optimization needs.(2)In this chemical process optimization system,the annual total cost TAC calculation model is taken as the objective function.The parameters of the double-tower pressure-swing distillation process of methanol-acetone azeotropic system were optimized by univariate search technique.Under four optimization strategies,the optimization of different variables and different order of variables in the simulation is realized.Optimal operating parameters are obtained,which improve the efficiency of process development.(3)In this system,the non-dominated sorting genetic algorithm(NSGA-II)is integrated to solve the multi-objective synthesis problem.In the case of acetone recovery,a variational convergence criterion strategy is used for the convergence of process recurrent flows.Calculations show that the proposed solution strategy is efficient and can be used to solve multi-objective synthesis problems for general processes.(4)This work studies the reinforcement learning algorithm Q-learning and integrates it on this system.In the chemical process optimization system,the optimization problem with two-dimensional and three-dimensional variables is solved with TAC as the objective function.With the given upper and lower limits and the use of the TAC objective function,an effective return function is constructed.The operating parameters of the process when the objective function TAC is the smallest are obtained.The two-dimensional variable and three-dimensional variable optimization of the pressure swing separation process of the methanol-acetone azeotrope system were realized,which laid a good foundation for other researchers to carry out multidimensional optimization.