Message Board

Dear readers, authors and reviewers,you can add a message on this page. We will reply to you as soon as possible!

2020 Volume 42 Issue 1
Article Contents

Shen-zhong WANG, Xin-gong ZHANG. The Supply Chain Scheduling Problem Based on Truncated Learning Effect and Time Dependence[J]. Journal of Southwest University Natural Science Edition, 2020, 42(1): 44-50. doi: 10.13718/j.cnki.xdzk.2020.01.007
Citation: Shen-zhong WANG, Xin-gong ZHANG. The Supply Chain Scheduling Problem Based on Truncated Learning Effect and Time Dependence[J]. Journal of Southwest University Natural Science Edition, 2020, 42(1): 44-50. doi: 10.13718/j.cnki.xdzk.2020.01.007

The Supply Chain Scheduling Problem Based on Truncated Learning Effect and Time Dependence

More Information
  • Corresponding author: Xin-gong ZHANG
  • Received Date: 29/10/2017
    Available Online: 20/01/2020
  • MSC: O223

  • This paper studies truncated learning effect and time-dependent effect on the supply chain scheduling. The truncated learning effect means that the actual processing time of a job is a function of its position and a control parameter. The objection functions are to minimize the makespan, the (weighted) total completion time and the maximum lateness. For the first two scheduling problems, we show that their optimal sequence can be obtained by the non-decreasing normal processing time scheduling. For the total weighted completion and the maximum lateness, we present the worst-case bounds by the heuristic sequencing rules of the classic scheduling algorithms. The latter two problems can also be solved by polynomial time algorithms under some special conditions between the normal processing times and job weights or due date. Finally, we give the results of some numerical experiments to elaborate the reasonability of these algorithms.
  • 加载中
  • [1] BISKUP D. Single-Machine Scheduling with Learning Considerations[J]. European Journal of Operational Reserch, 1999, 115(1):173-178. doi: 10.1016/S0377-2217(98)00246-X

    CrossRef Google Scholar

    [2] YANG D L, KUO W H. Some Scheduling Problems with Deteriorating Jobs and Learning Effects[J]. Computer and Industrial Engineering, 2010, 58(1):25-28. doi: 10.1016/j.cie.2009.06.016

    CrossRef Google Scholar

    [3] YANG D L, KUO W H. Some Scheduling Problems with Deteriorating Jobs and Learning Effects[J]. Computers & Industrial Engineering, 2010, 58(1):25-28.

    Google Scholar

    [4] ZHANG X G, YAN G L, HUANG W Z. Single-Machine Scheduling Problems with Time and Position Dependent Processing Times[J]. Annals of Operational Research, 2011, 186(1):345-356.

    Google Scholar

    [5] 张新功.时间相关的单机排序的最坏竞争比分析[J].重庆师范大学学报(自然科学版), 2013, 30(5):5-10.

    Google Scholar

    [6] KOULAMAS C, KYPARISIS G J. Single-Machine Scheduling Problems with Past-Sequence-Dependent Delivery Times[J]. International Journal of Production Economics, 2010, 126(2):264-266. doi: 10.1016/j.ijpe.2010.03.016

    CrossRef Google Scholar

    [7] 刘洋, 唐恒永, 赵传立.同时具有学习效应和退化效应的单机排序问题[J].运筹与管理, 2012, 21(3):81-86. doi: 10.3969/j.issn.1007-3221.2012.03.013

    CrossRef Google Scholar

    [8] WU C C, YIN Y, CHENG S R. Single-Machine and Two-Machine Flowshop Scheduling Problems with Truncated Position-Based Learning Functions[J]. Journal of Operational Research Society, 2012, 64(1):147-156.

    Google Scholar

    [9] 白静, 刘璐, 王吉波.具有截断学习效应和工件带准备时间的单机排序问题[J].运筹与管理, 2014, 23(6):152-156. doi: 10.3969/j.issn.1007-3221.2014.06.021

    CrossRef Google Scholar

    [10] 王申重.基于一般时间相关和位置相关的单机排序问题研究[J].重庆师范大学学报(自然科学版), 2017, 34(2):6-10.

    Google Scholar

  • 加载中
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Article Metrics

Article views(884) PDF downloads(196) Cited by(0)

Access History

Other Articles By Authors

The Supply Chain Scheduling Problem Based on Truncated Learning Effect and Time Dependence

    Corresponding author: Xin-gong ZHANG

Abstract: This paper studies truncated learning effect and time-dependent effect on the supply chain scheduling. The truncated learning effect means that the actual processing time of a job is a function of its position and a control parameter. The objection functions are to minimize the makespan, the (weighted) total completion time and the maximum lateness. For the first two scheduling problems, we show that their optimal sequence can be obtained by the non-decreasing normal processing time scheduling. For the total weighted completion and the maximum lateness, we present the worst-case bounds by the heuristic sequencing rules of the classic scheduling algorithms. The latter two problems can also be solved by polynomial time algorithms under some special conditions between the normal processing times and job weights or due date. Finally, we give the results of some numerical experiments to elaborate the reasonability of these algorithms.

  • 学习效应和时间相关的供应链排序问题在现代工业生产、管理科学、服务业等方面有着广泛的应用[1-9].基于带有配送问题的供应链排序问题是目前的研究热点.本文考虑截断学习效应、与时间相关的加工时间以及带有配送时间的供应链排序模型.配送时间就是工件加工完成后配送到客户处的时间.考虑的目标函数为最大完工时间、总(权)完工时间和最大延迟.

1.   问题描述
  • 本文假定:n个工件J1J2,…,Jn在同一台机器上进行加工,每次只能加工一个工件且在加工过程中不能中断;每个工件Jj有正常加工时间pj、权重wj、工期dj和配送时间qj;工件Jj的实际加工时间为pjr(t)=pjα(t)max{rab},其中t为工件Jj的开工时间,r为工件Jj在加工序列中的位置,α(t)为开工时间t的非增凸函数且满足0 < α(t)≤1,α(0)=1,α′(t) < 0,a < 0为学习效应,b为给定的参数且0 < b < 1.工件配送时间qj采用Koulamas[6]中的模型,即第r位工件的配送时间为qr=$c\sum\limits_{l = 1}^{r{\rm{ - }}1} {p[l]} $p[l]为排在第l位工件的正常加工时间且c≥0,简记为qpsd.符号CjLj表示工件Jj的完工时间和延迟,Cmax=max{Cj|j=1,2,…,n}表示最大完工时间,Lmax=max{Lj=Cj-dj|j=1,2,…,n}表示最大延迟.本文所研究的模型如下:

    利用简单的数学计算可以得到如下引理1:

    引理1  设$\lambda {\rm{ = }}\frac{{\max \left\{ {{{\left( {r + 1} \right)}^a}, b} \right\}}}{{\max \left\{ {{r^a}, b} \right\}}}$a < 0,0 < b < 1,则0 < λ≤1.

    引理2[10]  设α(t)是非增可微凸函数且α(t)在[0,+∞)上是非减函数,则当0 < k≤1,0 < λ≤1,m > 0,t0 > 0时,(k-1)α(t0)+λα(t0+mk)-λkα(t0+m)≤0成立.

    引理3[10]  设α(t)是非增可微凸函数且α(t)在[0,+∞)上是非减函数,则当0 < k≤1,0 < λ≤1,m > 0,0 < λ1λ2 < 1,t0 > 0时,(k-1)α(t0)+λλ1α(t0+mk)-λλ2kα(t0+m)≤0成立.

    定义1[10]  反一致关系:工件Jj的正常加工时间和权重满足pipjwiwj.

    定义2[10]  一致关系:工件Jj的正常加工时间和工期满足pipjdidj.

2.   单机排序问题
  • 定理1  对于模型(1),工件按照pj非减的顺序(SPT)排列即为最优序列.

      假设S1π1JiJjπ2是满足pipj的序列,交换工件JiJj得到S2π1JjJiπ2π1π2表示部分序列且π1中有r-1个工件,且第r个工件开工时间为t0.则有

    故有

    则0 < k≤1,m > 0.由引理1,2知0 < λ≤1,

    所以Cj(S1)≤Ci(S2).工件JiJj前面的工件在其交换位置后完工时间没有改变,工件JiJj后面的工件在其交换位置后开工时间没有变小,所以工件按照pj非减的顺序(SPT)排列即为最优序.

  • 定理2  对于模型(2),工件按照pj非减的顺序(SPT)排列即为最优序.

      类似于定理1,假设S1π1JiJjπ2S2π1JjJiπ2.在pipj的条件下只需证Cj(S1)≤Ci(S2)和Ci(S1)+Cj(S1)≤Ci(S2)+Cj(S2)即可.由定理1可知Cj(S1)≤Ci(S2),又pipj,则

    成立,故

    即工件按照pj非减顺序(SPT)排列为最优序.

  • 注意到经典的WSPT序并不能产生模型(3)的最优序列.接下来利用WSPT序作为启发式算法,给出了最坏竞争比分析如下:

    定理3  对于模型(3),将WSPT序排列所得序列和最优序列分别表示为ππ*,且在零时刻开工,则模型(3)的最坏竞争比分析为

    且这个界是紧的.

      记ππ(1)→π(2)→…→π(n)和π*π*(1)→π*(2)→…→π*(n),设p[i]pπ(i)表示π中第i个工件的正常加工时间和实际加工时间,p[i]*pπ*(i)表示π*中第i个工件的正常加工时间和实际加工时间,则

    进一步得

    其中Cπ*为定理1中的最大完工时间.

    由于0 < max{rab}≤1,0 < α(t)≤1,下面证明这个界是紧的.如果max{rab}=1(b→1)和α(t)=1,WSPT序即为模型(3)的最优序列,因此这个界是紧的.

    尽管经典的WSPT序规则不能得到模型(3)的最优解,但可以证明加工时间和权重满足反一致关系时,模型(3)仍然可以得到多项式时间算法.

    定理4  对于问题模型(3),当工件的正常加工时间和权重存在反一致关系即pipjwiwj时,按照$\frac{{{p_j}}}{{{w_j}}}$非减的顺序WSPT排列即为最优序.

      若S1π1JiJjπ2是满足$\frac{{{p_i}}}{{{w_j}}} \le \frac{{{p_j}}}{{{w_j}}}$的序列,交换工件JiJj得一新序S2π1JjJiπ2π1中有r-1个工件,且第r个工件的开工时间为t0.则有

    由引理1和引理3可知

    故当满足反一致关系时按照$\frac{{{p_j}}}{{{w_j}}}$非减的(WSPT)排列即为最优序.

    由定理4易知下面推论1成立.

    推论1  对于问题模型(3),当工件的权重和加工时间成比例时,pj非减的排列顺序(SPT)即为最优序.

  • 类似可知经典的EDD序规则不能得到模型(4)的最优序列.为了得到当前问题的最坏竞争比分析,需要重新定义最坏竞争比函数为

    其中dmax=max{dj|j=1,2,…,n},ππ*表示EDD序列和最优序列.

    定理5  对于模型(4),将按照EDD序排列所得序列和最优序列分别表示为ππ*,在零时刻开工,则

    且这个界是紧的.

      记ππ(1)→π(2)→…→π(n)和π*π*(1)→π*(2)→…→π*(n),设p[i]pπ(i)分别表示π中第i个工件的正常加工时间和实际加工时间,p[i]*pπ*(i)分别表示π*中第i个工件的正常加工时间和实际加工时间.

    其中Cmax*为定理1中的最优时间表长.

    由于0 < max{rab}≤1,0 < α(t)≤1,当max{rab}=1(b→1)和α(t)=1时EDD序即为模型(4)的最优序列,因此这个界是紧的.

    尽管经典的EDD序规则不能得到模型(4)的最优解,但可以证明加工时间和工期满足一致关系时,当前模型仍然是多项式时间可解的.

    定理6  对于模型(4),当正常加工时间和工期存在一致关系时,按照dj非减的顺序(EDD)排列即为最优序.

      若S1π1JiJjπ2是满足pipjdidj的序列,交换工件JiJj得一新序S2π1JjJiπ2π1中有r-1个工件,且第r个工件的开工时间为t0.则有

    由定理1,2可知max{Li(S1),Lj(S1)}≤max{Li(S2),Lj(S2)},Lmax(S1)≤Lmax(S2)成立,交换工件JiJj最大工期不会变小,所以当满足一致关系时按照dj非减的顺序(EDD)排列即为最优序.

    由定理6易知下面推论2和推论3成立.

    推论2  对于模型(4),如果工件的加工时间都相同,工件按照dj非减的顺序(EDD)排列即为最优序.

    推论3  对于模型(4),如果工件的工期都相同,工件按照pj非减的顺序(SPT)排列即为最优序.

3.   实例验证
  • 下面给出3个算例来验证上面所给出的结论的合理性和正确性.

    假设pjr(t)=pjα(t)max{rab}里α(t)=1-0.02ta=-0.5,b=0.75,c=0.1,在零时刻开始加工.

    例1  n=5,p1=4,p2=2,p3=6,p4=8,p5=10,对Cmax${\sum C _j}$,按照SPT序排列5个工件得到S1J2J1J3J4J5,计算可得C2=2,C1=5.08,C3=9.54,C4=15.07,C5=21.29,则有Cmax(S1)=21.29,${\sum C _j}$ (S1)=52.98.交换一下J1J2,得一新序S2J1J2J3J4J5,计算可得C1=4,C2=5.78,C3=10,C4=15.47,C5=21.63,则有Cmax(S2)=21.63,${\sum C _j}$ (S2)=56.88.由上面结果可知Cmax(S1) < Cmax(S2),${\sum C _j}$ (S1) < ${\sum C _j}$ (S2).

    例2  n=5,p1=4,p2=2,p3=6,p4=8,p5=10,w1=4,w2=5,w3=3,w4=2,w5=1,对${\sum {{w_j}C} _j}$,按照WSPT序排列5个工件得到S1J2J1J3J4J5,计算可得C2=2,C1=5.08,C3=9.54,C4=15.07,C5=21.29,则有${\sum {{w_j}C} _j}$ (S1)=110.37.交换一下J1J2,得一新序S2J1J2J3J4J5,计算可得C1=4,C2=5.78,C3=10,C4=15.47,C5=21.63,则有${\sum {{w_j}C} _j}$ (S2)=127.47.由上面结果可知${\sum {{w_j}C} _j}$ (S1) < ${\sum {{w_j}C} _j}$ (S2).

    例3  n=5,p1=6,p2=4,p3=9,p4=12,p5=15,d1=4,d2=3,d3=9,d4=10,d5=14,对于Lmax,按照EDD序排列5个工件得到序列S1J2J1J3J4J5,计算可得C2=4,L2=1,C1=8.54,L1=4.54,C3=14.79,L3=5.79,C4=22.21,L4=12.21,C5=30.09,L5=16.09,则有Lmax(S1)=16.09.交换一下J1J2,得一新序S2J1J2J3J4J5,计算得C1=6,L1=2,C2=9.24,L2=6.24,C3=15.22,L3=6.22,C4=22.56,L4=12.56,C5=30.36,L5=16.36,则有Lmax(S2)=16.36.从上面结果可知Lmax(S1) < Lmax(S2).

4.   结论
  • 本文研究了基于截断学习效应和非增函数的时间相关的供应链单机排序模型.工件的实际加工时间是截断学习效应及位置有关凸函数.首先给出了多项式时间算法.对于一般情形,利用经典的WSPT序,EDD序作为启发式算法给出了最坏竞争比分析.

Reference (10)

Catalog

    /

    DownLoad:  Full-Size Img  PowerPoint
    Return
    Return