Integrated Optimization Model For Feedstock And Processing Conditions Selection In The Thermal Cracking

Ethylene is the most significant raw material of petrochemical industry. Ethyleneproduction is often used as an important measure of a country’s petroleum chemicaldevelopment level.Along with the high-speed development of petrochemical industry in our country,the production capability of ethylene increases sharply. The consumption rate ofethylene, however, is still much larger than the growth rate of ethylene production.Meanwhile there is an obvious lack of the feedstock of ethylene production. Hence, thecritical thing to the petrochemical industry is reaching the ethylene plant’s full potentialand increasing the revenue of ethylene production based on the existing situation.Ethylene production is a complicated reaction process influenced by lots of factors.Most researches, simulate the cracking process, are based on the molecular reactiondynamics model. However, these models could not give the effective results for mostpractical problems, since the feedstock always have complex compositions and theactual production process is hard to control. Especially, much data has been involved inthe simulation process, which gives rise to the enormous computations, slowperformance and makes the inspection and debugging process difficult. Hence, a fastand effective model is needed to optimize the thermal cracking plant’s production.Based on the project with SINOPEC, this thesis develops a MILP to optimize theproduction of thermal cracking plant. A discretization method is used to deal with thenonlinear part during the cracking process and Lagrangian relaxation algorithm isproposed to solve the MILP. In order to make the model more close to the actualproduction process, this thesis considers the yield’s uncertainty and transforms thedeterministic optimization model into the robust counterpart.The final part of the thesis performs some computational experiments, whichshows the high efficiency and effectiveness of the optimization model and associatedLagrangian relaxation algorithm. Meanwhile the robust model shows its ability ofreducing the impact of yield’s uncertainty on the final result and makes the researchmeaningful to the real-world production.