数学物理学报, 2019, 39(5): 1228-1246 doi:

论文

具有p-进入规则和Min(N, D, V)-策略的M/G/1排队系统容量问题研究

罗乐1, 唐应辉,2,3

System Capacity Optimization Design and Optimal Control Policy (N*, D*) for M/G/1 Queue with p-Entering Discipline and Min(N, D, V)-Policy

Luo Le1, Tang Yinghui,2,3

通讯作者: 唐应辉, E-mail: tangyh@sicnu.edu.cn

收稿日期: 2018-10-23  

基金资助: 国家自然科学基金.  71571127

Received: 2018-10-23  

Fund supported: the NSFC.  71571127

摘要

该文研究具有p-进入规则和系统采取Min(NDV)-策略的M/G/1排队系统,其中在服务员多重休假期间到达的顾客以概率p(0 < p ≤ 1)进入系统.运用全概率分解技术和拉普拉斯变换工具讨论了系统从任意初始状态出发,在任意时刻t的瞬态队长分布,得到瞬态队长分布的拉普拉斯变换的表达式,进一步得到稳态队长分布的递推表达式.同时,结合稳态队长分布,通过数值计算实例讨论了系统容量的优化设计问题.最后,在建立系统费用结构模型的基础上,导出了系统长期单位时间内的期望费用的显示表达式,并通过数值实例确定了使得系统在长期单位时间内的期望费用最小的联合最优控制策略(N*D*).

关键词: 多重休假 ; p-进入规则 ; Min(N, D, V)-策略 ; 队长分布 ; 最优控制策略

Abstract

This paper considers a M/G/1 queueing system with p-entering discipline and Min(N, D, V)-policy, in which the customers who arrive during multiple vacations enter the system with probability p(0 < p ≤ 1). By using the total probability decomposition technique and the Laplace transform, we discuss the transient distribution of queue length at any time t which started from an arbitrary initial state, and obtain the expressions of the Laplace transform of transient queue-length distribution. Moreover, we obtain the recursion expressions of the steady-state queue length distribution. Meanwhile, we discuss the optimal capacity design by combining the steady-state queue length distribution and numerical example. Finally, the explicit expression of the long-run expected cost rate is derived under a given cost structure. And by through numerical calculation, we determine the optimal control policy (N*, D*) for minimizing the long-run expected cost per unit time.

Keywords: Multiple vacation ; p-Entering discipline ; Min(N, D, V)-policy ; Queue length distribution ; Optimal control policy

PDF (550KB) 元数据 多维度评价 相关文章 导出 EndNote| Ris| Bibtex  收藏本文

本文引用格式

罗乐, 唐应辉. 具有p-进入规则和Min(N, D, V)-策略的M/G/1排队系统容量问题研究. 数学物理学报[J], 2019, 39(5): 1228-1246 doi:

Luo Le, Tang Yinghui. System Capacity Optimization Design and Optimal Control Policy (N*, D*) for M/G/1 Queue with p-Entering Discipline and Min(N, D, V)-Policy. Acta Mathematica Scientia[J], 2019, 39(5): 1228-1246 doi:

1 引言

随着学者们对排队系统研究的深入和实际应用的需要,近几十年来在对有休假机制和控制策略的排队系统研究方面取得了较大的进展.早期的经典工作有Yadin和Naor的$N$ -策略[1], Balachandran的$D$ -策略[2], Yechiali和Heyman的$T$ -策略[3], Doshi的休假排队系统[4]等.从已有的研究情况看,各个经典控制策略和休假机制都是从不同的角度提出的,其目的就是在平衡顾客等待时间的同时降低系统的成本,增加系统的收入.在一些经典工作之后,一些推广研究文献陆续出现[5-15].考虑到单一策略有各自的优缺点,例如,当生产制造环境发生改变并且系统拥有者想转换成另一种控制策略时,在这种情况下抛弃现有的硬件系统是不现实的,从而联合控制策略就是一种可供选择的好方案,能更好地解决这一实际问题.因此,结合实际背景,一些作者提出了一些具有联合控制策略的排队系统模型[16-28].文献[24]建立了有${\rm Min}(N, D)$ -策略控制的$M/G/1$排队模型,文献[25]在文献[24]的基础上,把到达过程推广到马尔科夫过程,分析了有${\rm Min}(N, D)$ -策略控制的$MAP/G/1$排队系统,文献[26]进一步讨论了有${\rm Min}(N, D)$ -策略的$M/G/1$排队系统,文献[27]在文献[26]的基础上,把二维控制策略$(N, D)$与服务员的休假机制结合起来,构造了一个联合的${\rm Min}(N, D, V)$ -控制策略,使用全概率分解技术和利用拉普拉斯变换工具,研究了系统从任意初始状态出发的瞬态队长分布和稳态队长分布,同时,给出了稳态队长的随机分解结果和附加队长分布的显示表达式.最后在建立费用模型的基础上讨论了二维最优控制策略,并与单一的控制策略进行了比较分析.文献[28]讨论了基于${\rm Min}(N, D, V)$ -策略和单重休假的$M/G/1$排队系统.

在诸多研究排队系统的文献中,作者都假定顾客的输入率是不变的,但实际上,到达的顾客可能因为服务员不在岗而离去,或者系统管理者为了控制在下一个忙期开始时系统中的顾客总数以免形成拥挤而要采取限制进入措施.文献[29]首次讨论了休假期间顾客以概率$p(0 <p\leq 1)$进入系统的多重休假$M/G/1$排队系统.文献[30]讨论了假期顾客以概率$p(0\leq p\leq 1)$进入系统的单重休假$M/G/1$排队,之后一些具有可变到达率的排队系统得到了研究[31-32],同时,一些离散时间刻画的且有可变进入率的离散时间排队系统研究也取得了进展[33-37].本文在文献[27]的基础上,将"休假期间顾客以概率$p$进入系统"引入到多重休假的${\rm Min}(N, D, V)$ -策略$M/G/1$排队中,建立"具有$p$进入规则和多重休假的${\rm Min}(N, D, V)$ -策略$M/G/1$排队模型",运用全概率分解技术和拉普拉斯变换工具讨论了系统的瞬态队长和稳态队长分布.结合获得的稳态队长分布,并通过数值计算实例讨论了系统容量的优化设计问题.最后,建立费用模型,讨论了系统长期单位时间内的期望费用达到最小的最优二维控制策略.本文研究的系统模型描述如下:

1)系统是$M/G/1$排队系统,即顾客到达的的间隔时间$\left\{ {{\tau }_{n}}, n\ge 1 \right\}$相互独立同分布$F(t)=1-{{\rm e}^{-\lambda t}}, t\ge 0$.顾客所需的服务时间$\left\{ {{\chi }_{n}}, n \ge 1 \right\}$相互独立同一般分布$G(t)$,且设

2) $p$ -进入规则:在服务员休假期间到达的顾客以概率$p(0<p\leq 1)$进入系统.

3)系统采取${\rm Min}(N, D, V)$ -策略控制且服务员休假可中断,即当系统变空时,服务员马上开始一次休假.在服务员的休假期间,如果系统中到达的顾客数达到了$N$个,或者到达系统等待服务的顾客所需服务时间总量不小于$D(D\geq0)$,无论哪一个先发生,处于休假期的服务员立即中断休假回到系统为顾客服务;如果在服务员休假期间系统中有顾客到达,但到达数没有达到$N$个,且到达的顾客总的服务时间小于$D$,则等此次休假结束后服务员再回到系统开始为顾客服务;如果在服务员此次休假结束时系统中没有顾客到达,则服务员立刻开始另一次新的休假.设休假时间$V$服从任意分布$V(t)$.

4)随机变量$\tau, \chi, V$相互独立.

5)在$t=0$时刻,如果系统是空的,则系统不采取该控制策略,服务员留在系统中等待顾客的到达(这样的假设更符合实际情况),且到达的第一个顾客立即被服务.

注1.1  一些符号说明:$N(t)$表示时刻$t$系统中的顾客数; $g(s)$表示相应$G(t)$的拉普拉斯-斯蒂尔切斯(LS)变换, $g^*(s)$表示相应$G(t)$的拉普拉斯(L)变换; $G^{(k)}(t)$表示$G(t)$自身的$k(\geq 1)$重卷积,

$\Re(s)$为复变量$s$的实部.

2 系统的瞬态队长分布

又定义"服务员忙期"为从服务员开始为顾客服务的时刻起,直到系统再次变空为止的这一段时间,则本文定义的"服务员忙期"与文献[38,第4.4节]的"忙期"等价.令$b$表示该系统从一个顾客开始的"服务员忙期"长度,且令$B(t)=p\{b\leq t\}, \ b(s)=\int_{0}^{\infty}{\rm e}^{-st}{\rm d}B(t)$,则有如下引理:

引理2.1[38]  对$\Re(s)>0, b(s)$是方程$z=g(s+\lambda-\lambda z)$$|z| < 1$内的唯一根,且

其中$\rho=\frac{\lambda}{\mu}$表示系统的交通强度, $\omega(0 < \omega < 1)$是方程$z=g(\lambda-\lambda z)$在(0, 1)内的根.

$b^{\langle i\rangle }$表示从$i$个顾客开始的服务员忙期长度,因为到达过程是Poisson过程易知$b^{\langle i\rangle }$的分布

$Q_{j}(t)=p\{b>t\geq 0, N(t)=j\}$表示在服务员忙期$b$中队长为$j(\geq1)$且在$t=0$时服务员忙期$b$刚开始的瞬态概率,即$Q_{1}(0)=1, \ Q_{j}(0)=0, \ j>1$.

引理2.2[38]  令$q^{*}_{j}(s)=\int_{0}^{\infty}{\rm e}^{-st}Q_{j}(t){\rm d}t$$Q_{j}(t)$的L变换,对$\Re(s)>0$, $j\geq 1$,有

其中$\Delta_j(t)=\overline{G}(t)\frac{(\lambda t)^j}{j!}e^{-\lambda t}, j\geq0$;当$j < 0$时有$q^*_j(s)=0$且求和$\sum\limits_{k=0}^{j}=0$.

下面讨论系统队长的瞬态分布.令$p_{ij}(t)=p\{N(t)=j|N(0)=i\}$表示系统在时刻$t=0$,队长为$i(i\geq0)$的前提条件下,在任意时刻$t$系统队长为$j$的概率,且设

定理2.1  当$N\geq2$$D>0$时,对$\Re(s)>0$$i\geq1$,有

$p_{00}^{*}(s)=\frac{1-f(s)}{s}+ \frac{1-f_V(s)}{s} \frac{f(s)b(s)[1-v(s+\lambda p )]}{\Delta(s)-v(s+\lambda p)[1-f_V(s)b(s)]}, $

$p_{i0}^{*}(s)=\frac{1-f_V(s)}{s}\cdot\frac{b^{i}(s)[1-v(s+\lambda p)]} {\Delta(s)-v(s+\lambda p)[1-f_V(s)b(s)]}, $

其中

  用$\tau_{Vj}(j\geq 1)$表示在服务员休假期间,第$j$个到达并进入系统的顾客与第$j-1$个到达并进入系统的顾客的时间间隔.由模型假设可得$\{\tau_{Vj}, j\geq 1\}$相互独立、同分布$F_V(t)=1-{{\rm e}^{-\lambda p t}}$$ \widehat{\tau}_{Vj}$表示系统的第$j$个系统闲期长度,则其分布为

$S_{k}=\sum\limits_{i=1}^{k}V_{i}, l_{k}=\sum\limits_{i=1}^{k}\tau_{i}, H_{k}=\sum\limits_{i=1}^{k}\chi_{i}, $$S_{0}=l_{0}=H_{0}=0$.在时刻$t$队长为零当且仅当时刻$t$处在系统闲期中,依据模型的初始条件假设,运用全概率分解技术,有

$\begin{eqnarray} p_{i0}(t)&=&p\{b^{\langle i\rangle }\leq t <b^{\langle i\rangle }+ \widehat{\tau}_{V1}\}+p\{b^{\langle i\rangle }+\widehat{\tau}_{V1}\leq t, N(t)=0 \}\nonumber \\ &=&\int_0^t{\rm e}^{-\lambda p(t-x)}{\rm d}B^{(i)}(x) \nonumber \\ &&+\sum\limits_{k=1}^{\infty}\sum\limits_{m=1}^{N-1}p\{b^{\langle i\rangle }+S_{k}\leq t, S_{k-1}\leq\widehat{\tau}_{V1} <S_{k}, S_{k-1} <\widehat{\tau}_{V1}+l_{m-1}\leq S_{k}<\widehat{\tau}_{V1}+l_{m}, \nonumber \\ && H_{m}<D, N(t)=0\}\nonumber \\&&+\sum\limits_{k=1}^{\infty}p\{b^{\langle i\rangle }+\widehat{\tau}_{V1}+l_{N-1}\leq t, S_{k-1}\leq\widehat{\tau}_{V1}, \widehat{\tau}_{V1}+l_{N-1}\leq S_{k}, H_{N-1}<D, N(t)=0\}\nonumber \\&&+\sum\limits_{k=1}^{\infty}\sum\limits_{m=1}^{N-1}p\{b^{\langle i\rangle }+\widehat{\tau}_{V1}+l_{m-1}\leq t, S_{k-1}\leq\widehat{\tau}_{V1}, \widehat{\tau}_{V1}+l_{m-1}<S_{k}, \nonumber \\&& H_{m-1}<D\leq H_{m}, N(t)=0\} \nonumber \\&=&\int_0^t{\rm e}^{-\lambda p(t-x)}{\rm d}B^{(i)}(x) \nonumber \\&&+\sum\limits_{k=1}^{\infty}\sum\limits_{m=1}^{N-1}\int_{0}^{t}\int_{0}^{t-x}\int_{0}^{t-x-y}G^{(m)}(D)p_{m0}(t-x-y-z)\cdot\frac{(\lambda pz)^{m}}{m!}{\rm e}^{-\lambda p(y+z)}\nonumber \\&&{\rm d}V(z){\rm d}V^{(k-1)}(y){\rm d}B^{(i)}(x)\nonumber \\&&+\sum\limits_{k=1}^{\infty}\int_{0}^{t}\int_{0}^{t-x}\int_{0}^{t-x-y}G^{(N-1)}(D)p_{N0}(t-x-y-z)\overline{V}(z){\rm e}^{-\lambda py}\nonumber \\&&{\rm d}F_V^{(N)}(z){\rm d}V^{(k-1)}(y){\rm d}B^{(i)}(x)\nonumber \\&&+\sum\limits_{k=1}^{\infty}\sum\limits_{m=1}^{N-1}\int_{0}^{t}\int_{0}^{t-x}\int_{0}^{t-x-y}\overline{V}(z)[G^{(m-1)}(D)-G^{(m)}(D)]p_{m0}(t-x-y-z){\rm e}^{-\lambda py}\nonumber \\&&{\rm d}F_V^{(m)}(z){\rm d}V^{(k-1)}(y){\rm d}B^{(i)}(x).\end{eqnarray}$

同理有

$\begin{eqnarray}p_{00}(t)&=&\overline{F}(t)+\int_{0}^{t}\overline{F}_V(t-x){\rm d}[F(x)*B(x)]\nonumber \\&&+\sum\limits_{k=1}^{\infty}\sum\limits_{m=1}^{N-1}\int_{0}^{t}\int_{0}^{t-x}\int_{0}^{t-x-y}G^{(m)}(D)p_{m0}(t-x-y-z)\cdot\frac{(\lambda pz)^{m}}{m!}{\rm e}^{-\lambda p(y+z)}\nonumber \\&&{\rm d}V(z){\rm d}V^{(k-1)}(y){\rm d}[F(x)*B(x)]\nonumber \\&&+\sum\limits_{k=1}^{\infty}\int_{0}^{t}\int_{0}^{t-x}\int_{0}^{t-x-y}G^{(N-1)}(D)p_{N0}(t-x-y-z)\overline{V}(z){\rm e}^{-\lambda py}\\&&{\rm d}F_V^{(N)}(z){\rm d}V^{(k-1)}(y){\rm d}[F(x)*B(x)]\nonumber \\&&+\sum\limits_{k=1}^{\infty}\sum\limits_{m=1}^{N-1}\int_{0}^{t}\int_{0}^{t-x}\int_{0}^{t-x-y}\overline{V}(z)[G^{(m-1)}(D)-G^{(m)}(D)]p_{m0}(t-x-y-z){\rm e}^{-\lambda py}\nonumber \\&&{\rm d}F_V^{(m)}(z){\rm d}V^{(k-1)}(y){\rm d}[F(x)*B(x)].\end{eqnarray}$

对(2.3)式和(2.4)式作L变换,有

$\begin{eqnarray}p_{i0}^{*}(s)&=&\frac{[1-f_V(s)]b^{i}(s)}{s}\nonumber\\&&+\frac{b^{i}(s)}{1-v(s+\lambda p)}\bigg\{{\sum\limits_{m=1}^{N-1}[G^{(m-1)}(D)-G^{(m)}(D)]p^{*}_{m0}(s)\int_{0}^{\infty}{\rm e}^{-st}\overline{V}(t){\rm d}F_V^{(m)}(t)}\nonumber\\&&+G^{(N-1)}(D)p^{*}_{N0}(s)\int_{0}^{\infty}{\rm e}^{-st}\overline{V}(t){\rm d}F_V^{(N)}(t)\\&&+\sum\limits_{m=1}^{N-1}G^{(m)}(D)p_{m0}^{*}(s)\int_{0}^{\infty}{\rm e}^{-(s+\lambda p)t}\frac{(\lambda pt)^m}{m!}{\rm d}V(t)\bigg\}, \end{eqnarray}$

$\begin{eqnarray}p_{00}^{*}(s)&=&\frac{1-f(s)+f(s)b(s)[1-f_V(s)]}{s}\nonumber\\&&+\frac{f(s)b(s)}{1-v(s+\lambda p)}\bigg\{{\sum\limits_{m=1}^{N-1}[G^{(m-1)}(D)-G^{(m)}(D)]p^{*}_{m0}(s)\int_{0}^{\infty}{\rm e}^{-st}\overline{V}(t){\rm d}F_V^{(m)}(t)}\nonumber\\&&+G^{(N-1)}(D)p^{*}_{N0}(s)\int_{0}^{\infty}{\rm e}^{-st}\overline{V}(t){\rm d}F_V^{(N)}(t)\\ &&+\sum\limits_{m=1}^{N-1}G^{(m)}(D)p_{m0}^{*}(s)\int_{0}^{\infty}{\rm e}^{-(s+\lambda p)t}\frac{(\lambda pt)^m}{m!}{\rm d}V(t)\bigg\}.\end{eqnarray}$

由(2.5)式与(2.6)式可得$p^{*}_{00}(s)$$p^{*}_{i0}(s)$的关系如下

$\begin{eqnarray}p_{i0}^{*}(s)=\frac{b^{i-1}(s)}{f(s)}\left[p_{00}^{*}(s)-\frac{1-f(s)}{s}\right], \qquad i\geq1.\end{eqnarray}$

把(2.7)式代入(2.6)式,经整理可得(2.1)式,再把(2.1)式代入(2.7)式,可得(2.2)式,证毕.

注2.1  当$N=1$$D=0$时,本文研究的排队系统就是文献[29]讨论的在休假期内到达顾客以概率$p$进入系统的$M/G/1$多重休假排队系统,因此下面对$N=1$$D=0$的情况不再讨论.

定理2.2  当$N\geq2$$D>0$时,对$\Re(s)>0$$j=1, 2, 3, \cdots, N-1 $,有

$p_{0j}^{*}(s)=f(s)\bigg\{q_{j}^{*}(s)+\frac{{\sigma}_{j}(s)}{\Delta(s)-v(s+\lambda p)[1-f_V(s)b(s)]}\bigg\}, $

$p_{ij}^{*}(s)=\sum\limits_{k=1}^{i}q_{j-i+k}^{*}(s)b^{k-1}(s)+\frac{b^{i-1}(s)\sigma_{j}(s)}{\Delta(s)-v(s+\lambda p)[1-f_V(s)b(s)]}, \qquad i\geq1.$

其中

  当$j=1, 2, 3, \cdots, N-1$时,时刻$t$队长为$j$的充要条件是时刻$t$处于服务员忙期中且队长为$j$或者处于服务员假期中且队长为$j$,类似定理2.1的全概率分解,有

$\begin{eqnarray}p_{ij}(t)&=&p\{0\leq t<b^{\langle i\rangle }, N(t)=j\}\nonumber \\&&+\sum\limits_{k=1}^{\infty}p\{b^{\langle i\rangle }+\widehat{\tau}_{V1}\leq t<b^{\langle i\rangle }+S_{k}, S_{k-1}\leq \widehat{\tau}_{V1}, l_{j-1}\leq t-(\widehat{\tau}_{V1}+b^{\langle i\rangle })<l_{j}, H_{j}<D\}\nonumber \\&&+\sum\limits_{k=1}^{\infty}\sum\limits_{m=1}^{N-1}p\{b^{\langle i\rangle }+S_{k}\leq t, S_{k-1}\leq\widehat{\tau}_{V1}<S_{k}, S_{k-1}<\widehat{\tau}_{V1}+l_{m-1}\leq S_{k}<\widehat{\tau}_{V1}+l_{m}, \nonumber \\&& H_{m}<D, N(t)=j\}\nonumber \\&&+\sum\limits_{k=1}^{\infty}p\{b^{\langle i\rangle }+\widehat{\tau}_{V1}+l_{N-1}\leq t, S_{k-1}\leq\widehat{\tau}_{V1}, \widehat{\tau}_{V1}+l_{N-1}\leq S_{k}, H_{N-1}<D, N(t)=j\}\nonumber \\&&+\sum\limits_{k=1}^{\infty}\sum\limits_{m=1}^{N-1}p\{b^{\langle i\rangle }+\widehat{\tau}_{V1}+l_{m-1}\leq t, S_{k-1}\leq\widehat{\tau}_{V1}, \widehat{\tau}_{V1}+l_{m-1}<S_{k}, \nonumber \\&& H_{m-1}<D\leq H_{m}, N(t)=j\} \nonumber \\&=&\sum\limits_{k=1}^{i}\int_0^tQ_{j-i+k}(t-x){\rm d}B^{(k-1)}(x) \nonumber \\&&+\sum\limits_{k=1}^{\infty}\int_{0}^{t}\int_{0}^{t-x}G^{(j)}(D)\overline{V}(t-x-y){\rm e}^{-\lambda p(t-x)}\cdot\frac{[\lambda p(t-x-y)]^{j}}{j!}{\rm d}V^{(k-1)}(y){\rm d}B^{(i)}(x)\nonumber \\&&+\sum\limits_{k=1}^{\infty}\sum\limits_{m=1}^{N-1}\int_{0}^{t}\int_{0}^{t-x}\int_{0}^{t-x-y}G^{(m)}(D)p_{mj}(t-x-y-z)\cdot {\rm e}^{-\lambda p(y+z)}\frac{(\lambda pz)^{m}}{m!}\nonumber \\&&{\rm d}V(z){\rm d}V^{(k-1)}(y){\rm d}B^{(i)}(x)\nonumber \\&&+\sum\limits_{k=1}^{\infty}\int_{0}^{t}\int_{0}^{t-x}\int_{0}^{t-x-y}G^{(N-1)}(D)p_{Nj}(t-x-y-z)\overline{V}(z){\rm e}^{-\lambda py}\nonumber \\&&{\rm d}F_V^{(N)}(z){\rm d}V^{(k-1)}(y){\rm d}B^{(i)}(x)\nonumber \\&&+\sum\limits_{k=1}^{\infty}\sum\limits_{m=1}^{N-1}\int_{0}^{t}\int_{0}^{t-x}\int_{0}^{t-x-y}\overline{V}(z)[G^{(m-1)}(D)-G^{(m)}(D)]p_{mj}(t-x-y-z){\rm e}^{-\lambda py}\nonumber \\&&{\rm d}F_V^{(m)}(z){\rm d}V^{(k-1)}(y){\rm d}B^{(i)}(x).\end{eqnarray}$

同理可得

$\begin{eqnarray}p_{0j}(t)&=&\int_0^tQ_{j}(t-x){\rm d}F(x) \nonumber \\&&+\sum\limits_{k=1}^{\infty}\int_{0}^{t}\int_{0}^{t-x}G^{(j)}(D)\overline{V}(t-x-y){\rm e}^{-\lambda p(t-x)}\cdot\frac{[\lambda p(t-x-y)]^{j}}{j!}\nonumber \\&&{\rm d}V^{(k-1)}(y){\rm d}[F(x)*B(x)]\nonumber \\&&+\sum\limits_{k=1}^{\infty}\sum\limits_{m=1}^{N-1}\int_{0}^{t}\int_{0}^{t-x}\int_{0}^{t-x-y}G^{(m)}(D)p_{mj}(t-x-y-z)\cdot {\rm e}^{-\lambda p(y+z)}\frac{(\lambda pz)^{m}}{m!}\\ &&{\rm d}V(z){\rm d}V^{(k-1)}(y){\rm d}[F(x)*B(x)]\nonumber \\&&+\sum\limits_{k=1}^{\infty}\int_{0}^{t}\int_{0}^{t-x}\int_{0}^{t-x-y}G^{(N-1)}(D)p_{Nj}(t-x-y-z)\overline{V}(z){\rm e}^{-\lambda py}\nonumber \\&&{\rm d}F_V^{(N)}(z){\rm d}V^{(k-1)}(y){\rm d}[F(x)*B(x)]\nonumber \\&&+\sum\limits_{k=1}^{\infty}\sum\limits_{m=1}^{N-1}\int_{0}^{t}\int_{0}^{t-x}\int_{0}^{t-x-y}\overline{V}(z)[G^{(m-1)}(D)-G^{(m)}(D)]p_{mj}(t-x-y-z){\rm e}^{-\lambda py}\nonumber \\&&{\rm d}F_V^{(m)}(z){\rm d}V^{(k-1)}(y){\rm d}[F(x)*B(x)].\end{eqnarray}$

对(2.10)式和(2.11)式作L变换,得

$\begin{eqnarray}p_{ij}^{*}(s)&=&\sum\limits_{k=1}^{i}[b(s)]^{k-1}q_{j-i+k}^{*}(s)+\frac{b^{i}(s)}{1-v(s+\lambda p)}G^{(j)}(D)\int_{0}^{\infty}{\rm e}^{-(s+\lambda p)t}\overline{V}(t)\frac{(\lambda pt)^{j}}{j!}{\rm d}t\nonumber\\&&+\frac{b^{i}(s)}{1-v(s+\lambda p)}\bigg\{{\sum\limits_{m=1}^{N-1}[G^{(m-1)}(D)-G^{(m)}(D)]p^{*}_{mj}(s)\int_{0}^{\infty}{\rm e}^{-st}\overline{V}(t){\rm d}F_V^{(m)}(t)}\nonumber\\&&+G^{(N-1)}(D)p^{*}_{Nj}(s)\int_{0}^{\infty}{\rm e}^{-st}\overline{V}(t){\rm d}F_V^{(N)}(t)\\ &&+\sum\limits_{m=1}^{N-1}G^{(m)}(D)p_{mj}^{*}(s)\int_{0}^{\infty}{\rm e}^{-(s+\lambda p)t}\frac{(\lambda pt)^m}{m!}{\rm d}V(t)\bigg\}, \end{eqnarray}$

$\begin{eqnarray}p_{0j}^{*}(s)&=&f(s)q_{j}^{*}(s)+\frac{f(s)b(s)}{1-v(s+\lambda p)}G^{(j)}(D)\int_{0}^{\infty}{\rm e}^{-(s+\lambda p)t}\overline{V}(t)\frac{(\lambda p t)^{j}}{j!}{\rm d}t\nonumber\\&&+\frac{f(s)b(s)}{1-v(s+\lambda p)}\bigg\{{\sum\limits_{m=1}^{N-1}[G^{(m-1)}(D)-G^{(m)}(D)]p^{*}_{mj}(s)\int_{0}^{\infty}{\rm e}^{-st}\overline{V}(t){\rm d}F_V^{(m)}(t)}\nonumber\\&&+G^{(N-1)}(D)p^{*}_{Nj}(s)\int_{0}^{\infty}{\rm e}^{-st}\overline{V}(t){\rm d}F_V^{(N)}(t)\\ &&+\sum\limits_{m=1}^{N-1}G^{(m)}(D)p_{mj}^{*}(s)\int_{0}^{\infty}{\rm e}^{-(s+\lambda p)t}\frac{(\lambda pt)^m}{m!}{\rm d}V(t)\bigg\}.\end{eqnarray}$

由(2.12)式与(2.13)式可得$p_{0j}^{*}(s)$$p_{ij}^{*}(s)$的关系为

$\begin{eqnarray}p_{ij}^{*}(s)=\sum\limits_{k=1}^{i}q_{j-i+k}^{*}(s)b^{k-1}(s)+\frac{b^{i-1}(s)}{f(s)}[p_{0j}^{*}(s)-f(s)q_{j}^{*}(s)], \qquad i\geq1.\end{eqnarray}$

把(2.14)式代入(2.13)式,经整理可得(2.8)式,把(2.8)式代入(2.14)式可得(2.9)式,证毕.

定理2.3  当$N\geq2$$D>0$时,对$\Re(s)>0$$j\geq N $,有

$p_{0j}^{*}(s)=f(s)\left\{q_{j}^{*}(s)+\frac{\theta_{j}(s)}{\Delta(s)-v(s+\lambda p)[1-f_V(s)b(s)]}\right\}, $

$p_{ij}^{*}(s)=\sum\limits_{k=1}^{i}q_{j-i+k}^{*}(s)b^{k-1}(s)+\frac{b^{i-1}(s)\theta_{j}(s)}{\Delta(s)-v(s+\lambda p)[1-f_V(s)b(s)]}, i\geq1.$

  当$j\geq N$时,时刻$t$队长为$j$当且仅当时刻$t$在服务员忙期中且队长为$j$,类似定理2.2的证明,有

$\begin{eqnarray}p_{ij}(t)&=&p\{0\leq t<b^{\langle i\rangle }, N(t)=j\}\nonumber \\&&+\sum\limits_{k=1}^{\infty}\sum\limits_{m=1}^{N-1}p\{b^{\langle i\rangle }+S_{k}\leq t, S_{k-1}\leq\widehat{\tau}_{V1}<S_{k}, S_{k-1}<\widehat{\tau}_{V1}+l_{m-1}\leq S_{k}<\widehat{\tau}_{V1}+l_{m}, \nonumber \\&& H_{m}<D, N(t)=j\}\nonumber \\&&+\sum\limits_{k=1}^{\infty}p\{b^{\langle i\rangle }+\widehat{\tau}_{V1}+l_{N-1}\leq t, S_{k-1}\leq\widehat{\tau}_{V1}, \widehat{\tau}_{V1}+l_{N-1}\leq S_{k}, H_{N-1}<D, N(t)=j\}\nonumber \\&&+\sum\limits_{k=1}^{\infty}\sum\limits_{m=1}^{N-1}p\{b^{\langle i\rangle }+\widehat{\tau}_{V1}+l_{m-1}\leq t, S_{k-1}\leq\widehat{\tau}_{V1}, \widehat{\tau}_{V1}+l_{m-1}<S_{k}, \nonumber \\&& H_{m-1}<D\leq H_{m}, N(t)=j\} \nonumber \\&=&\sum\limits_{k=1}^{i}\int_0^tQ_{j-i+k}(t-x){\rm d}B^{(k-1)}(x) \nonumber \\ &&+\sum\limits_{k=1}^{\infty}\sum\limits_{m=1}^{N-1}\int_{0}^{t}\int_{0}^{t-x} \int_{0}^{t-x-y}G^{(m)}(D)p_{mj}(t-x-y-z)\cdot {\rm e}^{-\lambda p(y+z)}\frac{(\lambda pz)^{m}} {m!}\nonumber \\ && {\rm d}V(z){\rm d}V^{(k-1)}(y){\rm d}B^{(i)}(x)\nonumber \\ &&+\sum\limits_{k=1}^{\infty}\int_{0}^{t}\int_{0}^{t-x}\int_{0}^{t-x-y}G^{(N-1)}(D) p_{Nj}(t-x-y-z)\overline{V}(z){\rm e}^{-\lambda py}\nonumber \\ &&{\rm d}F_V^{(N)}(z){\rm d}V^{(k-1)}(y){\rm d}B^{(i)}(x)\nonumber \\ &&+\sum\limits_{k=1}^{\infty}\sum\limits_{m=1}^{N-1}\int_{0}^{t}\int_{0}^{t-x}\int_{0}^{t-x-y} \overline{V}(z)[G^{(m-1)}(D)-G^{(m)}(D)]p_{mj}(t-x-y-z){\rm e}^{-\lambda py}\nonumber \\ &&{\rm d}F_V^{(m)}(z){\rm d}V^{(k-1)}(y){\rm d}B^{(i)}(x). \end{eqnarray} $

同理可得

$\begin{eqnarray} p_{0j}(t)&=&\int_0^tQ_{j}(t-x){\rm d}F(x) \nonumber \\ &&+\sum\limits_{k=1}^{\infty}\sum\limits_{m=1}^{N-1}\int_{0}^{t}\int_{0}^{t-x} \int_{0}^{t-x-y}G^{(m)}(D)p_{mj}(t-x-y-z)\cdot {\rm e}^{-\lambda p(y+z)}\frac{(\lambda pz)^{m}} {m!}\nonumber \\ && {\rm d}V(z){\rm d}V^{(k-1)}(y){\rm d}[F(x)*B(x)] \nonumber \\ &&+\sum\limits_{k=1}^{\infty}\int_{0}^{t}\int_{0}^{t-x}\int_{0}^{t-x-y}G^{(N-1)}(D) p_{Nj}(t-x-y-z)\overline{V}(z){\rm e}^{-\lambda py} \\ && {\rm d}F_V^{(N)}(z){\rm d}V^{(k-1)}(y){\rm d}[F(x)*B(x)]\nonumber \\ &&+\sum\limits_{k=1}^{\infty}\sum\limits_{m=1}^{N-1}\int_{0}^{t}\int_{0}^{t-x}\int_{0}^{t-x-y} \overline{V}(z)[G^{(m-1)}(D)-G^{(m)}(D)]p_{mj}(t-x-y-z){\rm e}^{-\lambda py}\nonumber \\ && {\rm d}F_V^{(m)}(z){\rm d}V^{(k-1)}(y){\rm d}[F(x)*B(x)]. \end{eqnarray}$

对(2.17)式和(2.18)式作L变换,得

$ \begin{eqnarray} p_{ij}^{*}(s)&=&\sum\limits_{k=1}^{i}[b(s)]^{k-1}q_{j-i+k}^{*}(s)\nonumber\\ &&+\frac{b^{i}(s)}{1-v(s+\lambda p)} \bigg\{{\sum\limits_{m=1}^{N-1}[G^{(m-1)}(D)-G^{(m)}(D)]p^{*}_{mj}(s) \int_{0}^{\infty}{\rm e}^{-st}\overline{V}(t){\rm d}F_V^{(m)}(t)}\nonumber\\ &&+G^{(N-1)}(D)p^{*}_{Nj}(s)\int_{0}^{\infty}{\rm e}^{-st}\overline{V}(t){\rm d}F_V^{(N)}(t) \nonumber\\ &&+\sum\limits_{m=1}^{N-1}G^{(m)}(D)p_{mj}^{*}(s)\int_{0}^{\infty} {\rm e}^{-(s+\lambda p)t} \frac{(\lambda pt)^m}{m!}{\rm d}V(t)\bigg\}, \end{eqnarray} $

$ \begin{eqnarray} p_{0j}^{*}(s)&=&f(s)q_{j}^{*}(s)\nonumber\\ &&+\frac{f(s)b(s)}{1-v(s+\lambda p)} \bigg\{{\sum\limits_{m=1}^{N-1}[G^{(m-1)}(D)-G^{(m)}(D)]p^{*}_{mj}(s) \int_{0}^{\infty}{\rm e}^{-st}\overline{V}(t){\rm d}F_V^{(m)}(t)}\nonumber\\ &&+G^{(N-1)}(D)p^{*}_{Nj}(s)\int_{0}^{\infty}{\rm e}^{-st}\overline{V}(t){\rm d}F_V^{(N)}(t) \nonumber\\ &&+\sum\limits_{m=1}^{N-1}G^{(m)}(D)p_{mj}^{*}(s)\int_{0}^{\infty}{\rm e}^{-(s+\lambda p)t} \frac{(\lambda pt)^m}{m!}{\rm d}V(t)\bigg\}. \end{eqnarray}$

由(2.19)式与(2.20)式可得$p_{0j}^{*}(s)$$p_{ij}^{*}(s)$的关系为

$ \begin{eqnarray} p_{ij}^{*}(s)=\sum\limits_{k=1}^{i}q_{j-i+k}^{*}(s)b^{k-1}(s) +\frac{b^{i-1}(s)}{f(s)}[p_{0j}^{*}(s)-f(s)q_{j}^{*}(s)], \qquad i\geq1. \end{eqnarray} $

把(2.21)式代入(2.20)式,经整理可得(2.15)式,把(2.15)式代入(2.21)式可得(2.16)式,证毕.

3 系统的稳态队长分布

定理3.1  令$p_{j}=\lim\limits_{t\to \infty}p\{N(t)=j\}, $$j=0, 1, 2, \cdots, $则对任意初始状态,有

1)当$\rho\geq1$时, $\{p_{j}=0, j=0, 1, 2, \cdots\}$不构成概率分布;

2)当$\rho < 1$时,有

$p_{0}=\frac{1-\rho}{1-\rho(1-p)}\cdot\frac{1-v(\lambda p)}{\triangle_{N}}, $

$p_{j}=\frac{\lambda p(1-\rho)}{1-\rho(1-p)}\cdot\frac{\sigma_{j}}{\triangle_{N}}, \qquad j=1, 2, \cdots, N-1, $

$p_{j}=\frac{\lambda p(1-\rho)}{1-\rho(1-p)}\cdot\frac{\theta_{j}}{\triangle_{N}}, \qquad j=N, N+1, N+2, \cdots, $

其中

  当$\rho>1$$\rho=1$时,注意到此时有: $\lim\limits _{s\rightarrow0^{+}}b(s)=\omega(0 <\omega < 1)$$\lim\limits _{s\rightarrow0^{+}}b(s)=1$$E(b)=\infty$,以及

$\begin{eqnarray}\lim\limits_{s\rightarrow0^{+}}\Delta(s)&=&1-G^{(N-1)}(D)\int_{0}^{\infty}\overline{V}(t){\rm d}F_V^{(N)}(t)-\sum\limits_{m=1}^{N-1}G^{(m)}(D)\int_{0}^{\infty}{\rm e}^{-\lambda pt}\frac{(\lambda p t)^{m}}{m!}{\rm d}V(t)\nonumber\\\&&-\sum\limits_{m=1}^{N-1}[G^{(m-1)}(D)-G^{(m)}(D)]\int_{0}^{\infty}\overline{V}(t){\rm d}F_V^{(m)}(t)-\int_{0}^{\infty}V(t){\rm d}F(t)\nonumber\\\&=&-\sum\limits_{m=1}^{N}G^{(m-1)}(D)\int_{0}^{\infty}F_V^{(m)}(t){\rm d}V(t)\nonumber\\ &&+\sum\limits_{m=1}^{N-1}G^{(m)}(D)\int_{0}^{\infty}\bigg[1-{\rm e}^{-\lambda p t}\sum\limits_{i=0}^{m}\frac{(\lambda pt)^{i}}{i!}\bigg]{\rm d}V(t)+\int_{0}^{\infty}F(t){\rm d}V(t)\nonumber\\\&=&-\sum\limits_{m=0}^{N}G^{(m)}(D)\int_{0}^{\infty}F_V^{(m+1)}(t){\rm d}V(t)+\sum\limits_{m=1}^{N-1}G^{(m)}(D)\int_{0}^{\infty}F_V^{(m+1)}(t){\rm d}V(t)\nonumber\\ &&+\int_{0}^{\infty}F(t){\rm d}V(t)=0.\end{eqnarray}$

与当$E[b]=\infty$时,有

$\begin{eqnarray}\lim\limits_{s\rightarrow0^{+}}\Delta'(s)&=&G^{(N-1)}(D)\int_{0}^{\infty}[NE[b]+t]\overline{V}(t){\rm d}F_V^{(N)}(t)\nonumber\\ &&+\sum\limits_{m=1}^{N-1}G^{(m)}(D)\int_{0}^{\infty}[mE[b]+t]\frac{(\lambda pt)^m}{m!}{\rm e}^{-\lambda pt}{\rm d}V(t)\nonumber\\&&+\sum\limits_{m=1}^{N-1}[G^{(m-1)}(D)-G^{(m)}(D)]\int_{0}^{\infty}[mE[b]+t]\overline{V}(t){\rm d}F_V^{(m)}(t)\nonumber\\ &&+\int_{0}^{\infty}[E[b]+t]V(t){\rm d}F(t)\nonumber\\&=&\sum\limits_{m=1}^{N}G^{(m-1)}(D)E[b]\int_{0}^{\infty}F_V^{(m)}(t){\rm d}V(t)+\sum\limits_{m=0}^{N-1}G^{(m)}(D)\int_{0}^{\infty}t\frac{(\lambda pt)^m}{m!}{\rm e}^{-\lambda pt}{\rm d}V(t)\nonumber\\&&+\sum\limits_{m=1}^{N}G^{(m-1)}(D)\int_{0}^{\infty}t\overline{V}(t){\rm d}F_V^{(m)}(t)-\sum\limits_{m=1}^{N-1}G^{(m)}(D)\int_{0}^{\infty}t\overline{V}(t){\rm d}F_V^{(m)}(t)\nonumber\\&&+\frac{v(\lambda p)}{\lambda p}+v(\lambda p)E[b]\nonumber\\&=&\sum\limits_{m=1}^{N}G^{(m-1)}(D)E[b]\int_{0}^{\infty}F_V^{(m)}(t){\rm d}V(t)-\sum\limits_{m=1}^{N}G^{(m-1)}(D)\int_{0}^{\infty}\overline{V}(t)F_V^{(m)}(t){\rm d}t\nonumber\\&&+\frac{v(\lambda p)}{\lambda p}+\sum\limits_{m=0}^{N-1}G^{(m)}(D)\int_{0}^{\infty}\overline{V}(t)F_V^{(m)}(t){\rm d}t+v(\lambda p)E[b]=\infty.\end{eqnarray}$

使用洛必达法则,经计算可得$\lim\limits _{s\rightarrow0^{+}}sp_{ij}^{*}(s)=0$,从而当$\rho\geq1$时,有

$\rho < 1$时,注意到此时有$\lim\limits _{s\rightarrow0^{+}}b(s)=1$$E[b]=\frac{\rho}{\lambda(1-\rho)}$,并结合上面(3.4)式,有

$\begin{eqnarray}\lim\limits_{s\rightarrow0^{+}}\Delta'(s)&=&\sum\limits_{m=1}^{N}G^{(m-1)}(D)E[b]\int_{0}^{\infty}F_V^{(m)}(t){\rm d}V(t)-\sum\limits_{m=1}^{N}G^{(m-1)}(D)\int_{0}^{\infty}\overline{V}(t)F_V^{(m)}(t){\rm d}t\nonumber\\&&+\frac{v(\lambda p)}{\lambda p}+\sum\limits_{m=0}^{N-1}G^{(m)}(D)\int_{0}^{\infty}\overline{V}(t)F_V^{(m)}(t){\rm d}t+v(\lambda p)E[b]\nonumber\\&=&\sum\limits_{m=1}^{N}G^{(m-1)}(D)E[b]\int_{0}^{\infty}F_V^{(m)}(t){\rm d}V(t)\nonumber\\&&+\frac{1}{\lambda p}\sum\limits_{m=1}^{N}G^{(m-1)}(D)\int_{0}^{\infty}F_V^{(m)}(t){\rm d}V(t)+\frac{v(\lambda p)}{\lambda p(1-\rho)}\nonumber\\&=&\frac{v(\lambda p)+\Delta_{N}}{\lambda p(1-\rho)}.\end{eqnarray}$

再使用洛必达法则,经计算可得递推表达式(3.1)-(3.3)式.

定理3.2  当$\rho < 1$时,令$P(z)$表示该系统稳态队长分布$\{p_{j}, j=0, 1, 2, \cdots\}$的概率母函数,则

$\begin{eqnarray}P(z)&=&\frac{(1-\rho)}{[1-\rho(1-p)]\Delta_N}\cdot \frac{g(\lambda(1-z))(1-pz)-z(1-p)}{g(\lambda(1-z))-z}\nonumber\\&&\cdot \bigg[1-v(\lambda p)+\sum\limits_{m=1}^{N-1}z^{m}\int_{0}^{\infty}G^{(m)}(D)F_V^{(m+1)}(t){\rm d}V(t)\bigg], \qquad |z|<1, \end{eqnarray}$

且平均队长为

$\begin{equation}E[L]=\frac{2p\rho-\lambda^2(1-p)E[\chi^2]}{2[1-\rho(1-p)]}+\frac{\lambda^{2}E[\chi^{2}]}{2(1-\rho)}+\frac{\sum\limits_{m=1}^{N-1}m\int_{0}^{\infty}G^{(m)}(D)F_V^{(m+1)}(t){\rm d}V(t)}{\Delta_{N}}.\end{equation}$

  由$P(z)=\sum\limits_{j=0}^{\infty}z^{j}p_{j}$可得

$\begin{eqnarray}P(z)&=&\frac{(1-\rho)}{[1-\rho(1-p)]\Delta_N}\nonumber\\&&\cdot \bigg[1-v(\lambda p)+\lambda p\sum\limits_{j=1}^{\infty}z^{j}\theta_{j}+\lambda p \sum\limits_{j=1}^{N-1}z^{j}G^{(j)}(D)\int_{0}^{\infty}{\rm e}^{-\lambda pt}\overline{V}(t)\frac{(\lambda pt)^j}{j!}{\rm d}t\bigg].\end{eqnarray}$

$\begin{eqnarray}\sum\limits_{j=1}^{\infty}z^{j}\theta_{j}&=&\sum\limits_{j=1}^{\infty}z^{j}\sum\limits_{m=1}^{N-1}\sum\limits_{k=1}^{m}q_{j-m+k}\int_{0}^{\infty}\overline{V}(t)[G^{(m-1)}-G^{(m)}(D)]{\rm d}F_V^{(m)}(t)\nonumber\\&&+\sum\limits_{j=1}^{\infty}z^{j}G^{(N-1)}(D)\sum\limits_{k=1}^{N}q_{j-N+k}\int_{0}^{\infty}\overline{V}{\rm d}F_V^{(N)}(t)\\ &&+\sum\limits_{j=1}^{\infty}z^{j}\sum\limits_{m=1}^{N-1}G^{(m)}(D)\sum\limits_{k=1}^{m}q_{j-m+k}\int_{0}^{\infty}{\rm e}^{-\lambda pt}\frac{(\lambda pt)^{m}}{m!}{\rm d}V(t)\nonumber\\&=&\bigg(\sum\limits_{j=1}^{\infty}z^{j}q_{j}\bigg)\cdot\bigg[1-v(\lambda p)+\sum\limits_{m=1}^{N-1}z^{m}G^{(m)}(D)\int_{0}^{\infty}F_V^{(m+1)}(t){\rm d}V(t)\bigg].\end{eqnarray}$

$\begin{equation}\lambda p\sum\limits_{j=1}^{N-1}z^jG^{(j)}(D)\int_{0}^{\infty}{\rm e}^{-\lambda p t}\overline{V}(t)\frac{(\lambda pt)^j}{j!}{\rm d}t=\sum\limits_{j=1}^{N-1}z^jG^{(j)}(D)\int_{0}^{\infty}F_V^{(j+1)}(t){\rm d}V(t).\end{equation}$

$\begin{equation}\sum\limits_{j=1}^{\infty}z^{j}q_{j}=\frac{z[1-g(\lambda(1-z))]}{\lambda[g(\lambda(1-z))-z]}.\end{equation}$

将(3.10)-(3.12)式代入(3.9)式整理可得(3.7)式.再由

使用洛必达法则得(3.8)式.

注3.1  当$p=1$时,本文研究的排队系统等价于文献[27]研究的排队系统,在上述所得结论中,令$p=1$,可得与文献[27]完全一致的结果.

4 系统容量设计的优化

下面通过数值计算实例来讨论稳态队长分布$\{p_j, j=0, 1, 2, \cdots\}$在系统容量设计中的重要应用.

设服务时间服从参数$\mu(>0)$的负指数分布$G(t)=1-{\rm e}^{-\mu t}$与休假时间服从参数$\theta(>0)$的负指数分布$V(t)=1-{\rm e}^{-\theta t}$.

取参数$\lambda=0.8, \mu=2, D=4, N=10, \theta=0.5, p=0.9$,下面表 1给出了稳态队长$\{p_j, j=0, 1, 2, \cdots\}$的数值结果(小数点后保留六位).

表 1   $\lambda=0.8, \mu=2, D=4, N=10, \theta=0.5, p=0.9$时,稳态队长$\{p_j, j\geq0\}$的数值结果

$p_0$$p_1$$p_2$$p_3$$p_4$$p_5$$p_6$$p_7$$p_8$$p_9$$p_{10}$$p_{11}$$p_{12}$$p_{13}$$ p_{14}$$ p_{15}$$p_{16}$$p_{17}$$p_{18}$$p_{19}$$p_{20}$$p_{21}$$ p_{22}$$E[L]$
0.2622260.2340050.1734730.1160780.0732180.0441670.0255620.0141850.0075390.0038390.0015220.0006090.0002430.0000970.0000390.0000160.0000060.0000020.0000000.0000000.0000000.0000000.0000001.917604

新窗口打开| 下载CSV


表 1中数据可知,当$j$大于某一值,此时稳态队长分布$p_j$的取值已经十分接近于0,所以在系统容量设计中不需要把系统容量设计为无穷大.由表 1中数据可知:

也就是说,若按平均队长为标准来进行系统容量设计,那么到达的顾客由于容量不足而损失的概率为$0.503769$,即使容量增加一个单位,顾客损失的概率也达到了0.330296,损失的概率也很高.因此,在做系统容量设计时不能只依靠平均队长.

若要使到达的顾客因为系统容量不足而损失的概率不超过0.1,那系统容量至少为多大才合适?

$p\{L>M\}\leq0.1$,结合表 1中数据可得$M\geq5$,即设计系统容量至少为$M=5$才满足要求.若要使到达的顾客因为系统容量不足而损失的概率不超过0.05,则由$p\{L>M\}\leq0.05$,可得$M\geq8$,即至少取系统容量$M=8$.由此可见稳态队长分布$\{p_j, j=0, 1, 2, \cdots\}$在系统容量设计时有十分重要的作用.

5 费用模型下的最优控制策略

本文建立的费用结构模型如下:

1)系统(服务台)在一个周期内的固定消耗费用(启动费用)为$c$个单位;

2)一个顾客在系统中逗留(包括等待和服务)单位时间的成本费用为$h$个单位.

$F$为系统在${\rm Min}(N, D, V)$ -策略下,系统在长期运行单位时间内所产生的成本期望费用.因为服务员忙期的开始和结束时刻均为系统的更新时刻点,服务员忙期和“服务员非忙期”形成了系统的一个更新周期,其中这里的“服务员非忙期”是指服务员的实际休假长度(从系统刚变空的时刻起,直到服务员结束休假且立即开始为顾客服务的时刻为止的这段时间).因此,由更新报酬过程理论知:

系统的一个更新周期是由一个服务员忙期和一个"服务员非忙期"组成,其中这里的"服务员非忙期"是指服务员的实际休假长度(从系统刚变空的时刻起,直到服务员结束休假且立即开始为顾客服务的时刻为止的这段时间).

$A$表示服务员忙期开始时系统内的顾客数; $I$表示“服务员非忙期”长度; $B$表示服务员忙期的长度; $C$表示系统一个更新周期的长度,则在忙期开始时系统顾客数$A$有分布:

因此,服务员忙期开始时系统内的平均顾客数为

服务员忙期的平均长度为

由于服务员忙期开始时在系统内的顾客数就是在上一个“服务员非忙期”内到达的顾客数,而顾客到达过程是参数$\lambda$的Poisson过程,再结合$p$ -进入规则可得“服务员非忙期”的平均长度为

于是,系统一个更新周期的平均长度为

从而系统在长期运行单位时间内所产生的期望费用为

$\begin{eqnarray} F&=&hE[L]+\frac{c}{E[C]}\nonumber \\ &=&h\Bigg\{ \frac{2p\rho-\lambda^2(1-p)E[\chi^2]}{2[1-\rho(1-p)]} +\frac{{{\lambda }^{2}}E[{{\chi }^{2}}]}{2(1-\rho)} \\ &&+\frac{\sum\limits_{m=1}^{N-1}{m\int_{0}^{\infty }{{{G}^{(m)}}(D){{F}_V^{(m+1)}}}(x){\rm d}V(x)}}{\Big[ 1-v(\lambda p)+\sum\limits_{m=1}^{N-1}{\int_{0}^{\infty }{{{G}^{(m)}}(D){{F}_V^{(m+1)}}}(x){\rm d}V(x)} \Big]} \Bigg\}\nonumber \\ &&+\frac{c\lambda p(1-\rho )[1-v(\lambda p)]} {[1-\rho(1-p)]\Big[ 1-v(\lambda p)+\sum\limits_{m=1}^{N-1}{\int_{0}^{\infty }{{{G}^{(m)}}(D){{F}_V^{(m+1)}}}(x){\rm d}V(x)} \Big]}. \end{eqnarray} $

由(5.1)式看出, $F$是关于$N$$D$的非线性函数,理论上直接求出使其最小的最优解是非常困难的.

下面结合具体的数值计算实例来讨论最优控制策略$(N^*, D^*)$的寻求问题.令服务时间服从参数$\mu(>0)$的负指数分布$G(t)=1-{\rm e}^{-\mu t}$与休假时间服从参数$\theta(>0)$的负指数分布$V(t)=1-{\rm e}^{-\theta t}$时,可得$F$的表达式为

$\begin{eqnarray}F&=&h\left\{ \frac{\rho p }{(1-\rho)[1-\rho(1-p)]}+\frac{\sum\limits_{m=1}^{N-1 }{m{{\left( \frac{\lambda p}{\lambda p+\theta } \right)}^{m+1}}\cdot \left[ 1-{{{\rm e}}^{-\mu D}}\sum\limits_{i=0}^{m-1}{\frac{{{(\mu D)}^{i}}}{i!}} \right]}}{\frac{\lambda p}{\lambda p+\theta }+\sum\limits_{m=1}^{N-1 }{{{\left( \frac{\lambda p}{\lambda p+\theta } \right)}^{m+1}}\cdot \left[ 1-{{{\rm e}}^{-\mu D}}\sum\limits_{i=0}^{m-1}{\frac{{{(\mu D)}^{i}}}{i!}} \right]}} \right\}\nonumber \\&&+\frac{c\lambda p(1-\rho)\cdot \frac{\lambda p}{\lambda p+\theta }}{[1-\rho(1-p)]\cdot\left\{ \frac{\lambda p}{\lambda p+\theta}+ \sum\limits_{m=1}^{N-1} {{{\left( \frac{\lambda p}{\lambda p+\theta } \right)}^{m+1}}\cdot \left[ 1-{{{\rm e}}^{-\mu D}}\sum\limits_{i=0}^{m-1}{\frac{{{(\mu D)}^{i}}}{i!}} \right]} \right\}}.\end{eqnarray}$

取各参数值分别为

图 1表 2给出了$F$$N$$D$的变化情况(小数点后保留六位).

图 1

图 1   $p=0.9, \lambda=0.5, \rho=0.5, \theta=0.1, c=50, h=2$时, $F$$N$$D$的变化情况


表 2   $p=0.9, \lambda=0.5, \rho=0.5, \theta=0.1, c=50, h=2$时, $F$$N$$D$的变化情况

DN$1$$2$$3$$4$$5$$6$
113.1447379.9915369.5637659.5215309.5233489.525169
213.1447379.3121638.6666418.6067448.6418938.668439
313.1447379.0988308.3563638.3009008.3912888.474396
413.1447379.0245168.2352078.1849788.3257968.476696
513.1447378.9977098.1858168.1382598.3155108.527172
613.1447378.9879188.1654668.1189118.3194598.577624
713.1447378.9843258.1571018.1108288.3251548.615477
813.1447378.9830058.1536868.1075538.3294708.640445
913.1447378.9825198.1523048.1060518.3321708.655647
1013.1447378.9823048.1517488.1055718.3336998.664381
1113.1447378.9822758.1515268.1052348.3345108.669176
1213.1447378.9822518.1514388.1051378.3349218.671714
1313.1447378.9822428.1514048.1050978.3351238.673016
1413.1447378.9822398.1513908.1050828.3352988.673667
1513.1447378.9822378.1513858.1050758.3352658.673985
1613.1447378.9822378.1513838.1050738.3352858.674137
1713.1447378.9822378.1513828.1050728.3352948.674209
17.313.1447378.9822378.1513828.1050728.3352968.674222
17.413.1447378.9822378.151382$ \bf{\fbox{ 8.105071}}$8.3352968.674226
17.513.1447378.9822378.1513828.1050718.3352978.674229
1813.1447378.9822378.1513828.1050718.3352988.674243
1913.1447378.9822378.1513818.1050718.3353008.674258
2013.1447378.9822378.1513818.1050718.3353018.674265

新窗口打开| 下载CSV


图 1表 2中可以看出,当$(N^*, D^*)=(4, 17.4)$时, $F$取得最小值8.105071.因此,当系统中顾客数达到4个或者是到达顾客所需服务时间总量不小于17.4时,服务员立即终止休假开始为顾客服务.

验证:由于$N$是离散取值的,所以采用边际法验证$N^*$.表 2数据可知,当$(N^*, D^*)=(4, 17.4)$时,经计算有

同时

因此, $N^*=4$满足要求.下面验证$D^*$.

由于$F_{(N^*=4, D)}$关于$D$是可导的,且$D$是连续取值的.计算

于是在误差$\varepsilon=10^{-3}$前提下, $D^*=17.4$就可作为最优值点.

参考文献

Yadin M , Naor P .

Queueing system with a removable service station

Operation Research Quarterly, 1963, 14 (4): 393- 405

DOI:10.1057/jors.1963.63      [本文引用: 1]

Balachandran K .

Control policies for a single server system

Management Science, 1973, 19 (4): 1013- 1018

URL     [本文引用: 1]

Heyman D P .

$T$-policy for the $M/G/1$ queue

Management Science, 1977, 23 (7): 775- 778

DOI:10.1287/mnsc.23.7.775      [本文引用: 1]

Doshi B .

Queueing systems with vacations-A survey

Queueing Systems, 1986, 1: 29- 66

DOI:10.1007/BF01149327      [本文引用: 1]

Ke J C .

Modified $T$ vacation policy for an $M/G/1$ queueing system with an unreliable server and startup

Mathematical and Computer Modeling, 2005, 41: 1267- 1277

DOI:10.1016/j.mcm.2004.08.009      [本文引用: 1]

Lee H W , Beak J W , Jeon J .

Analysis of the $M^X/G/1$ queue under $D$-policy

Stochastic Analysis and Applications, 2005, 23 (4): 785- 808

DOI:10.1081/SAP-200064479     

Wang K H , Kuo C C , Ke J C .

Optical control of the $D$-policy $M/G/1$ queueing system with server breakdowns

American Journal of Applied Science, 2008, 5 (5): 565- 573

DOI:10.3844/ajassp.2008.565.573     

Lan S J , Tang Y H .

Analysis of $D$-policy discrete-time $Geo/G/1$ queue with second J-optional service and unreliable server

RAIRO-Operations Research, 2017, 51 (1): 101- 122

DOI:10.1051/ro/2016006     

Tang Y H , Tang X W .

The queue-length distribution for $M^X/G/1$ queue with single server vacation

Acta Mathematica Scientia, 2000, 20B (3): 397- 408

骆川义, 唐应辉, 刘仁彬.

多级适应性休假$M^X/G/1$排队系统的队长分布

系统科学与数学, 2007, 27 (6): 899- 907

DOI:10.3969/j.issn.1000-0577.2007.06.011     

Luo C Y , Tang Y H , Liu R B .

The queue length distribution of $M^X/G/1$ with adartive mulitistage vacation

Journal of Systems Science & Mathematical Sciences, 2007, 27 (6): 899- 907

DOI:10.3969/j.issn.1000-0577.2007.06.011     

Tang Y H .

The departure process of the $M/G/1$ queueing model with server vacation and exhaustive service discipline

J of Applied Probability, 1994, 31 (4): 1070- 1082

DOI:10.2307/3215330     

Tang Y H , Yun X , Huang S J .

Discrete-time queue with unreliable server and multiple adaptive delayed vacation

J of Computational and Applied Mathematics, 2008, 220 (3): 439- 455

URL    

Yu M M , Tang Y H , Fu Y H , Pan L M .

$GI/Geom/1/MWV$ queue with changeover time and searching for the optimum service rate in working vacation period

J of Computational and Applied Mathematics, 2011, 235 (8): 2170- 2184

DOI:10.1016/j.cam.2010.10.013     

Luo C Y , Tang Y H .

The recursive solution for $Geom/G/1(E, SV)$ queue with feedback and single server vacation

Acta Mathematicae Applicatae Sinica, 2011, 27 (1): 155- 166

DOI:10.1007/s10255-011-0049-y     

Luo C Y , Tang Y H , Chao B S , Xiang K L .

Performance analysis of a discrete-time $Geo/G/1$ queue with randomized vacations and at most J vacations

Applied Mathematics Modelling, 2013, 37 (9): 6489- 6504

DOI:10.1016/j.apm.2013.01.033      [本文引用: 1]

Lan S J , Tang Y H , Yu M M .

System capacity optimization design and optimal threshold $N^*$ for a $Geo/G/1$ discrete-time queue with single server vacation and under the control of ${\rm Min}(N, V)$-policy

Journal of Industrial & Management Optimization, 2016, 12 (4): 1435- 1464

[本文引用: 1]

Gu J X , Wei Y Y , Tang Y H , Yu M M .

Queue size distribution of $Geo/G/1$ queue under the ${\rm Min}(N, D)$-policy

J of Systems Science and Complexity, 2016, 29 (3): 752- 771

DOI:10.1007/s11424-016-4180-y     

井彩霞, 崔颖, 田乃硕.

${\rm Min}(N, V)$ -策略休假的$M/G/1$排队系统分析

运筹与管理, 2006, 15 (3): 53- 58

DOI:10.3969/j.issn.1007-3221.2006.03.011     

Jing C X , Cui Y , Tian N S .

Analysis of the $M/G/1$ queueing system with ${\rm Min}(N, V)$-policy

Operations Research & Management Science, 2006, 15 (3): 53- 58

DOI:10.3969/j.issn.1007-3221.2006.03.011     

唐应辉, 吴文青, 刘云颇, 刘晓云.

基于多重休假的${\rm Min}(N, V)$ -策略$M/G/1$排队系统的队长分布

系统工程理论与实践, 2014, 34 (6): 1533- 1546

Tang Y H , Wu W Q , Liu Y P , Liu X Y .

The queue length distribution of $M/G/1$ queueing system with ${\rm Min}(N, V)$-policy based on multiple server vacations

Systems Engineering-Theory & Practice, 2014, 34 (6): 1533- 1546

唐应辉, 吴文青, 刘云颇.

基于单重休假的${\rm Min}(N, V)$ -策略$M/G/1$排队系统分析

应用数学学报, 2014, 37 (6): 976- 996

Tang Y H , Wu W Q , Liu Y P .

Analysis of $M/G/1$ queueing system with ${\rm Min}(N, V)$-policy based on single server vacation

Acta Mathematicae Applicatae Sinica, 2014, 37 (6): 976- 996

蒋书丽, 唐应辉.

具有多级适应性休假和${\rm Min}(N, V)$ -策略控制的$M/G/1$排队系统

系统科学与数学, 2017, 37 (8): 1866- 1884

Jiang S L , Tang Y H .

$M/G/1$ queueing system with multiple adaptive vacations and ${\rm Min}(N, V)$-policy

Journal of Systems Science & Mathematical Sciences, 2017, 37 (8): 1866- 1884

高丽君, 唐应辉.

具有${\rm Min}(N, D)$ -策略控制的$M/G/1$可修排队系统及最优控制策略

数学物理学报, 2017, 37A (2): 352- 365

Gao L J , Tang Y H .

$M/G/1$ repairable queueing system and optimal control policy with ${\rm Min}(N, D)$-policy

Acta Mathematica Scientia, 2017, 37A (2): 352- 365

蔡晓丽, 唐应辉.

具有温储备失效特征和单重休假${\rm Min}(N, V)$ -控制策略的$M/G/1$可修排队系统

应用数学学报, 2017, 40 (5): 702- 726

Cai X L , Tang Y H .

$M/G/1$ repairable queueing system with warm standby failure and ${\rm Min}(N, V)$-policy based on single vacation

Acta Mathematicae Applicatae Sinica, 2017, 40 (5): 702- 726

Lee H W , Seo W J .

The performance of the $M/G/1$ queue under the dyadic ${\rm Min}(N, D)$-policy and its cost optimization

Performance Evaluation, 2008, 65 (10): 742- 758

DOI:10.1016/j.peva.2008.04.006      [本文引用: 2]

Lee H W , Seo W J , Lee S W , Jeon J .

Analysis of the $MAP/G/1$ queue under the ${\rm Min}(N, D)$-policy

Stochastic Models, 2010, 26 (1): 98- 123

DOI:10.1080/15326340903517121      [本文引用: 1]

魏瑛源, 唐应辉, .

基于${\rm Min}(N, D)$ -策略的$M/G/1$排队系统的队长分布及最优策略

系统科学与数学, 2015, 35 (6): 729- 744

[本文引用: 2]

Wei Y Y , Tang Y H , et al.

Queue length distribution and optimum policy for $M/G/1$ Queueing system under ${\rm Min}(N, D)$-policy

Journal of Systems Science & Mathematical Sciences, 2015, 35 (6): 729- 744

[本文引用: 2]

罗乐, 唐应辉.

具有${\rm Min}(N, D, V)$ -策略控制的$M/G/1$排队系统

运筹学学报, 2019, 23 (2): 1- 16

[本文引用: 4]

Luo L , Tang Y H .

$M/G/1$ queueing system with ${\rm Min}(N, D, V)$-policy control

Journal of Operations Research, 2019, 23 (2): 1- 16

[本文引用: 4]

王敏, 唐应辉.

基于${\rm Min}(N, D, V)$ -策略和单重休假的$M/G/1$排队系统的最优控制策略

系统科学与数学, 2018, 38 (9): 1067- 1084

[本文引用: 2]

Wang M , Tang Y H .

Optimal control policy of $M/G/1$ queueing system based on ${\rm Min}(N, D, V)$-policy and single server vacation

Journal of Systems Science & Mathematical Sciences, 2018, 38 (9): 1067- 1084

[本文引用: 2]

唐应辉, 毛勇.

服务员假期中以概率$p$进入的$M/G/1$排队系统的随机分解

数学物理学报, 2004, 24 (6): 683- 688

[本文引用: 2]

Tang Y H , Mao Y .

The stochastic decomposition for $M/G/1$ queue with $ p$-entering discipline during server vacations

Acta Mathematica Scientia, 2004, 24 (6): 683- 688

[本文引用: 2]

骆川义, 唐应辉.

假期中以概率$p$进入的单重休假$M/G/1$排队

应用数学, 2006, 19 (2): 246- 251

[本文引用: 1]

Luo C Y , Tang Y H .

The $M/G/1$ queue with $p$-entering discipline during single server vacation

Mathematic Application, 2006, 19 (2): 246- 251

[本文引用: 1]

李才良, 唐应辉, 牟永聪, .

在第二类故障期间以概率$p$进入的$M/G/1$可修排队系统

数学物理学报, 2012, 32A (6): 1149- 1157

[本文引用: 1]

Li C L , Tang Y H , Mu Y C , et al.

$M/G/1$ repairable queueing system with $p$-entering discipline during second type failure times

Acta Mathematica Scientia, 2012, 32A (6): 1149- 1157

[本文引用: 1]

刘云颇, 唐应辉.

多重假期中以概率$p$进入的$M/G/1$可修排队系统

系统工程学报, 2011, 26 (5): 718- 724

[本文引用: 1]

Liu Y P , Tang Y H .

$M/G/1$ repairable queueing system with $p$-entering discipline during server vacation

Journal of Systems Engineering, 2011, 26 (5): 718- 724

[本文引用: 1]

骆川义, 唐应辉.

具有可变到达率的多重休假$Geo^{\lambda_1, \lambda_2}/G/1$排队分析

数学学报, 2010, 53 (4): 805- 816

[本文引用: 1]

Luo C Y , Tang Y H .

Analysis of a multi-vacation $Geo^{\lambda_1, \lambda_2}/G/1$ queue with variable arrival rate

Acta Mathematic Sinica, 2010, 53 (4): 805- 816

[本文引用: 1]

Luo C Y , Xiang K L , Yu M M , Tang Y H .

Recursive solution of queue length distribution for $Geo/G/1$ queue with single server vacation and variable input rate

Computers and Mathematics with Applications, 2011, 61 (9): 2401- 2411

DOI:10.1016/j.camwa.2011.02.018     

Wei Y Y , Yu M M , Tang Y Y , Gu J X .

Queue size distribution and capacity optimum design for $N$-policy $Geo^{\lambda_1, \lambda_2, \lambda_3}/G/1$ queue with setup time and variable input rate

Mathematical and Computer Modelling, 2013, 57 (5/6): 1559- 1571

URL    

Luo C Y , Tang Y H , Yu K Z , Ding C .

Optimal $(r, N)$ -policy for discrete-time $Geo/G/1$ queue with different input rate and setup time

Applied Stochastic Models in Business & Industry, 2015, 31 (4): 405- 423

Lan S J , Tang Y H .

Analysis of a discrete-time $Geo^{\lambda_1, \lambda_2}/G/1$ queue with $N$-policy and $D$-policy

Journal of Applied Mathematics and Computing, 2017, 53 (1/2): 657- 681

[本文引用: 1]

唐应辉, 唐小我. 排队论-基础与分析技术. 北京: 科学出版社, 2006

[本文引用: 3]

Tang Y H , Tang X W . Queueing Theory-Foundations and Analysis Techniques. Beijing: Science Press, 2006

[本文引用: 3]

/