Distributed Parallel Optimization Of Ethylene Cracking Furnace Based On Linux Cluster System

Ethylene is one of the important basic raw materials in the petrochemicalindustry. Production and technical level of ethylene has been as a sign of anational petrochemical industry. Cracking furnace is key component ofethylene plant. Process simulation and optimization of cracking furnace hasbecome an effective adjunct to improve economic benefit of ethylene industry.Due to complexity of pyrolysis furnace, the mathematic model of pyrolysisfurnace based on process mechanism and rigorous physical properties hasbeen characteristics of large-scale and non-linear. Therefore, Solving theoptimization proposition based on mechanism model of cracking furnace istime-consuming. With the rapid development of the computer science andtechnology, improvement of solution efficiency such as the kind ofoptimization problem utilizing parallel computing technology has becomenoticeable hot spots. The research in the field of parallel optimization hasmainly focused on development of large-grained parallel algorithms becauseof the trend of distributed computing. Particle swarm optimization (PSO) algorithm is an important non-numerical parallel algorithm, the excellentnatural parallelism of which provides advantage for development oflarge-grained parallel algorithms. The research topic of the paper is distributedparallel optimization of ethylene cracking furnace, the major researchachivements include:(1) Combined with the Kumar molecule reaction kinetics and cokingkinetics model, a periodic steady-state mathematical model with parametersoptimization for radiant section of SRT-III type naphtha pyrolysis furnace wasdeveloped based on pseudo-steady assumption. The simulation results of themechanism model can relect real commercial operation of cracking furnace.(2) The high-performance computing platform based on Linux clustersystem was builded through comparing a variety of parallel computerarchitectures. The specific design scheme of the cluster system was given. Theoptimization method for improving performance of the cluster system isproposed on the basis of the HPL(High Performance Linpack) test results.(3) A parallel collaborative PSO algorithm was proposed based onMaster-Slave communication mode to meet the need of development oflarge-grained algorithm. The performance of proposed algorithm was testutilizing five common benchmark functions based on Message PassingInterface(MPI) parallel environment. The simulation results demonstrate thevalidity and reliability of LPC-PSO algorithm, especially for complexhigh-dimension functions, and choice of communication period has a great impact on convergence performance of the algorithm.(4) The proposition to adjust operating variables in the different stageduring the whole cracking period was stdudied. The objective function wasdefined as average sum of ethylene yield and propylene yield of the wholeperiod; Coil outlet temperature (COT) of every quasi-steady-state section werechoosed as decision variables; the optimization model of periodic operationwas established, which was solved by the way of distributed paralleloptimization. The simulation results show distributed parallel optimization ofthe cracking furnace can improve the efficiency of the optimization calculationcompared with serial optimization, and the value of objective function hasincreased by3.81%, indicating that the economic benefits of the naphthacracking furnace can be increased significantly through periodic operationoptimization.