首页 | 本学科首页   官方微博 | 高级检索  
     


Faster integer-feasibility in mixed-integer linear programs by branching to force change
Authors:Jennifer Pryor  John W Chinneck
Affiliation:Systems and Computer Engineering, Carleton University, Ottawa, Ontario, Canada K1S 5B6
Abstract:Branching in mixed-integer (or integer) linear programming requires choosing both the branching variable and the branching direction. This paper develops a number of new methods for making those two decisions either independently or together with the goal of reaching the first integer-feasible solution quickly. These new methods are based on estimating the probability of satisfying a constraint at the child node given a variable/direction pair. The surprising result is that the first integer-feasible solution is usually found much more quickly when the variable/direction pair with the smallest probability of satisfying the constraint is chosen. This is because this selection forces change in many candidate variables simultaneously, leading to an integer-feasible solution sooner. Extensive empirical results are given.
Keywords:Mixed-integer programming  Branching  Integer feasibility
本文献已被 ScienceDirect 等数据库收录!
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司  京ICP备09084417号