Integration Prcess Design And Control With Uncertainty Based On Transfer Learning

Chemical process industry is an important part of modern society,it is closely related to eco-nomic development.In order to promote the competitiveness of enterprises and high-end manu-facturing levels,and achieve flexible production,integration process design and control has be-come the most important way to enhance the level of automation and intelligence in the chemical industry.This paper focus on the "transfer learning,algorithm design,uncertainty analysis and the op-timization strategy".Mining the related knowledge and dynamic characteristic from the complex process,proposed the surrogate modeling approach with incorporated process steady state knowl-edge,and came up with transfer learning algorithm based on the process dynamic data.System-atically analyzed the uncertainties that affect process design and control from the perspective of optimization strategies.The main contents and major contributions of this thesis are described as follows:1.A novel surrogate model for dynamic process is developed to enhance the modeling effi-ciency and computation expense in the IPDC problem.Based on the surrogate model,a new dynamic modeling approach is proposed with transfer different source process dynamic data to the new design process.The algorithm attempts to mining the process dynamic relation from the data,transfer and reuse the dynamic data to the new process.A novel active select strategy is proposed to find the most useful data to the new target process,which proved to be efficient in discarding the invalid data and avoid the negative transfer.The proposed data transfer algorithm greatly improved the efficiency and the fidelity of the process model,and the computation cost in IPDC optimization is further decreased2.A novel transfer base vector is proposed to extract multiple dimension knowledge from the source process model.Furthermore,the propagation of uncertainty in the multiple step ahead prediction and in the model transfer arc detailed analysis,which provides a theoretical basis of improving the credibility of the transfer learning model.The simulation results show that the base vector transfer algorithm is highly improved the extrapolation ability of the process model and the modeling efficient is further improved3.A chance constrained programming approach with varying time scale uncertainty is devel-oped to cope with the uncertainty that affects both process steady state design and control system design.This paper proposed a new form of uncertainty that varies both in long time scale and short time scale.Analyze in detail the impact on multiple sources and varying time scale uncertainties during long-period operation of the process.Then a novel inter-active solution approach is proposed to solve the varying time scale uncertainty.Based on preset confidence level,the IPDC optimization can be solved in a flexible trade-off between economic benefits and the uncertainty risks.4.A novel fuzzy decision optimization formulation is proposed to cope with complex dynamic characteristics and discrete state transitions.This paper analyzed the impact of steady-state point selection on process dynamics and process economic performance.Through fuzzy decision-making,the uncertainty in the constraints and objective functions are fuzzified with membership function,and a compromise design between the worst-case conservative design and the radical design are optimized.The designer can avoid design decisions that are too conservative or too radical,and take into account of the economic benefits,system robustness,and uncertainty risks.