Modeling on Blend Scheduling with Recipe Optimization based on Improved Genetic Algorithm in Oil Refinery
According to the character of oil-refinery process, the overall oil-refinery scheduling involves the scheduling of crude oil unloading with inventory control, the production unit scheduling, and the finished product blending and shipping scheduling. The finished product blending scheduling is a crucial effect in the oil-refinery operations scheduling. The finished product blending scheduling optimization problem involves recipe optimization and blend scheduling. The recipe optimization problem is formulated as a nonlinear programming because of the nonlinear product properties. The scheduling problem is formulated as a mixed integer programming. Continuous variables are defined to represent flow rates as well as starting and ending times of processing tasks. Binary variables are principally related to assignment decisions of tasks to event points, or to some specific aspect of each problem. Based on the background of one domestic oil-refinery, an integrated finished product blend scheduling optimization model is given. The model in which recipe optimization and blend scheduling are considered simultaneously is a mixed integer nonlinear programming. It is difficult to be solved by general method because it is NPHard. An improved general algorithm is proposed to solve this model. As a kind of high efficiency algorithm, genetic algorithm has been applied to many fields. However, traditional genetic algorithm has the shortcomings of converging easily and the limitation on local search. In the paper, an improved genetic algorithm is proposed. Based on the analyses of simplex search and arithmetic crossover and combining both together, an improved crossover operator is presented to improve the local searching capability of genetic algorithm and lead gradually the population to the extreme point so as to implement the rapid searching.At the same time , to lead infeasible individuals to approach the feasible region so as to improve their feasibilities for the better , the penalty and repair strategies are associated with each other to form a repair operator for repairing infeasible individuals , accelerating the speed of the individuals to approach the feasible region and improving the searching efficiency and the capability in solving the nonlinear constraint. As a whole, the performance of the algorithm is therefore improved. A case study based on realistic data provided by one domestic oil-refinery is given in the paper. It is shown that the MILNP model proposed in this paper can be solved efficiently even for realistic large-scale problems. The improved genetic algorithm can converge to the solutions faster and more reliably.
genetic algorithm simplex search mixed integer nonlinear p programming penalty strategy recipe optimization blend scheduling
XU Jian-you TANG Li-xin GU Shu-sheng
College of Information Science and Engineering,Northeastern University,Shenyang,110004,P.R.China
国际会议
西安
英文
2007-08-15（万方平台首次上网日期，不代表论文的发表时间）