第三章 Poisson过程(2)
1.Poisson过程的合成与分解
泊松过程的合成:对于两个泊松过程
N
1
,
N
2
\boldsymbol N_1,\boldsymbol N_2
N1,N2,参数分别为
λ
1
,
λ
2
\lambda_1,\lambda_2
λ1,λ2,令
N
(
t
)
=
N
1
(
t
)
+
N
2
(
t
)
N(t)=N_1(t)+N_2(t)
N(t)=N1(t)+N2(t),则
N
=
(
N
(
t
)
,
t
≥
0
)
\boldsymbol N=(N(t),t\ge0)
N=(N(t),t≥0)是参数为
λ
1
+
λ
2
\lambda_1+\lambda_2
λ1+λ2的泊松过程。由于
N
1
,
N
2
\boldsymbol N_1,\boldsymbol N_2
N1,N2各自具有独立平稳增量性,所以只要证明
N
(
t
)
N(t)
N(t)的分布即可。
P
(
N
(
t
)
=
k
)
=
P
(
N
1
(
t
)
+
N
2
(
t
)
=
k
)
=
∑
l
=
0
k
P
(
N
1
(
t
)
=
l
,
N
2
(
t
)
=
k
−
l
)
=
∑
l
=
0
k
(
λ
1
t
)
l
l
!
e
−
λ
1
t
(
λ
2
t
)
k
−
l
(
k
−
l
)
!
e
−
λ
2
t
=
(
λ
1
+
λ
2
)
k
t
k
k
!
e
−
(
λ
1
+
λ
2
)
t
\begin{aligned} &P(N(t)=k)\\ =&P(N_1(t)+N_2(t)=k)\\ =&\sum_{l=0}^k P(N_1(t)=l,N_2(t)=k-l)\\ =&\sum_{l=0}^k \frac{(\lambda_1t)^l}{l!}e^{-\lambda _1t}\frac{(\lambda_2t)^{k-l}}{(k-l)!}e^{-\lambda_2t}\\ =&\frac{(\lambda_1+\lambda_2)^kt^k}{k!}e^{-(\lambda_1+\lambda_2)t} \end{aligned}
====P(N(t)=k)P(N1(t)+N2(t)=k)l=0∑kP(N1(t)=l,N2(t)=k−l)l=0∑kl!(λ1t)le−λ1t(k−l)!(λ2t)k−le−λ2tk!(λ1+λ2)ktke−(λ1+λ2)t
泊松过程的分解:对于参数为
λ
\lambda
λ的Poisson过程
N
\boldsymbol N
N,发生的事件可以分为I型与II型,且每一个事件独立地以
p
p
p概率属于I型,以
1
−
p
1-p
1−p概率属于II型。则I型事件的发生服从参数为
λ
p
\lambda p
λp的Poisson过程
N
1
\boldsymbol N_1
N1,II型事件的发生服从参数为
λ
(
1
−
p
)
\lambda(1-p)
λ(1−p)的Poisson过程
N
2
\boldsymbol N_2
N2,且
N
1
,
N
2
\boldsymbol N_1,\boldsymbol N_2
N1,N2也相互独立。
设
ξ
i
=
1
\xi_i=1
ξi=1代表第
i
i
i个事件为I型,
ξ
i
=
0
\xi_i=0
ξi=0代表第
i
i
i个事件为II型,那么
ξ
i
\xi_i
ξi独立同分布于
b
(
1
,
p
)
b(1,p)
b(1,p)。这样就有
N
1
(
t
)
=
∑
i
=
1
N
(
t
)
ξ
i
,
N
2
(
t
)
=
∑
i
=
1
N
(
t
)
(
1
−
ξ
i
)
N_1(t)=\sum_{i=1}^{N(t)}\xi_i,\quad N_2(t)=\sum_{i=1}^{N(t)}(1-\xi_i)
N1(t)=i=1∑N(t)ξi,N2(t)=i=1∑N(t)(1−ξi)
证明
N
1
\boldsymbol N_1
N1是参数为
λ
p
\lambda p
λp的随机过程,有
P
(
N
1
(
t
)
=
k
)
=
∑
n
=
k
∞
P
(
N
1
(
t
)
=
k
∣
N
(
t
)
=
n
)
P
(
N
(
t
)
=
n
)
=
∑
n
=
k
∞
C
n
k
p
k
(
1
−
p
)
n
−
k
λ
n
t
n
n
!
e
−
λ
t
=
∑
n
=
k
∞
p
k
(
1
−
p
)
n
−
k
(
λ
t
)
n
k
!
(
n
−
k
)
!
e
−
λ
t
=
p
k
(
λ
t
)
k
k
!
e
−
λ
t
∑
n
=
k
∞
[
λ
t
(
1
−
p
)
]
n
−
k
(
n
−
k
)
!
=
(
λ
t
p
)
k
k
!
e
−
λ
t
p
\begin{aligned} &P(N_1(t)=k)\\ =&\sum_{n=k}^\infty P(N_1(t)=k|N(t)=n)P(N(t)=n)\\ =&\sum_{n=k}^\infty C_n^kp^k(1-p)^{n-k}\frac{\lambda^nt^n}{n!}e^{-\lambda t}\\ =&\sum_{n=k}^\infty \frac{p^k(1-p)^{n-k}(\lambda t)^n}{k!(n-k)!}e^{-\lambda t}\\ =&\frac{p^k(\lambda t)^k}{k!}e^{-\lambda t}\sum_{n=k}^\infty \frac{[\lambda t(1-p)]^{n-k}}{(n-k)!}\\ =&\frac{(\lambda tp)^k}{k!}e^{-\lambda tp} \end{aligned}
=====P(N1(t)=k)n=k∑∞P(N1(t)=k∣N(t)=n)P(N(t)=n)n=k∑∞Cnkpk(1−p)n−kn!λntne−λtn=k∑∞k!(n−k)!pk(1−p)n−k(λt)ne−λtk!pk(λt)ke−λtn=k∑∞(n−k)![λt(1−p)]n−kk!(λtp)ke−λtp
独立平稳增量性也可以证明。
2.到达时刻的条件分布
到达时刻的条件分布,指的是在给定
N
(
t
)
=
n
N(t)=n
N(t)=n的情况下,在
[
0
,
t
]
[0,t]
[0,t]区间内会发生
n
n
n个事件,它们的到达次序依次为
S
1
,
⋯
,
S
n
S_1,\cdots,S_n
S1,⋯,Sn,求
(
S
1
,
⋯
,
S
n
)
(S_1,\cdots,S_n)
(S1,⋯,Sn)的联合分布。事实上,有
(
S
1
,
⋯
,
S
n
∣
N
(
t
)
=
n
)
=
d
(
U
(
1
)
,
⋯
,
U
(
n
)
)
(S_1,\cdots,S_n|N(t)=n)\stackrel d=(U_{(1)},\cdots,U_{(n)})
(S1,⋯,Sn∣N(t)=n)=d(U(1),⋯,U(n))
这里
U
(
1
)
,
⋯
,
U
(
n
)
U_{(1)},\cdots,U_{(n)}
U(1),⋯,U(n)是
[
0
,
t
]
[0,t]
[0,t]上
n
n
n个独立同分布的均匀随机变量的次序统计量。
P
(
S
k
∈
(
x
k
−
ε
k
,
x
k
+
ε
k
)
∣
N
(
t
)
=
n
)
=
P
(
S
k
∈
(
x
k
−
ε
k
,
x
k
+
ε
k
)
,
N
(
t
)
=
n
)
P
(
N
(
t
)
=
n
)
=
[
∏
k
=
1
n
λ
(
2
ε
k
)
]
e
−
λ
t
(
λ
t
)
n
n
!
e
−
λ
t
=
n
!
t
n
∏
k
=
1
n
(
2
ε
k
)
\begin{aligned} &P(S_k\in (x_k-\varepsilon_k,x_k+\varepsilon_k)|N(t)=n)\\ =&\frac{P(S_k\in (x_k-\varepsilon_k,x_k+\varepsilon _k),N(t)=n)}{P(N(t)=n)}\\ =&\frac{[\prod_{k=1}^n \lambda(2\varepsilon_k) ]e^{-\lambda t}}{\frac{(\lambda t)^n}{n!}e^{-\lambda t}}=\frac{n!}{t^n}\prod_{k=1}^n (2\varepsilon_k) \end{aligned}
==P(Sk∈(xk−εk,xk+εk)∣N(t)=n)P(N(t)=n)P(Sk∈(xk−εk,xk+εk),N(t)=n)n!(λt)ne−λt[∏k=1nλ(2εk)]e−λt=tnn!k=1∏n(2εk)
注意,计算这个条件概率的分子时,实际上要将
[
0
,
t
]
[0,t]
[0,t]分为
2
k
+
1
2k+1
2k+1个区间,其中
k
k
k个区间是
(
x
k
−
ε
k
,
x
k
+
ε
k
)
(x_k-\varepsilon_k,x_k+\varepsilon_k)
(xk−εk,xk+εk),它们中各有一个事件发生;另外
k
+
1
k+1
k+1个是出去这
k
k
k个后的剩余区间,它们中都没有事件发生。由此计算条件概率
p
S
1
,
⋯
,
S
n
∣
N
(
t
)
(
x
1
,
⋯
,
x
n
∣
n
)
=
P
(
S
k
∈
(
x
k
−
ε
k
,
x
k
+
ε
k
)
∣
N
(
t
)
=
n
)
∏
k
=
1
n
(
2
ε
k
)
=
n
!
t
n
\begin{aligned} &p_{S_1,\cdots,S_n|N(t)}(x_1,\cdots,x_n|n)\\ =&\frac{P(S_k\in (x_k-\varepsilon_k,x_k+\varepsilon_k)|N(t)=n)}{\prod_{k=1}^n (2\varepsilon_k)}\\ =&\frac{n!}{t^n} \end{aligned}
==pS1,⋯,Sn∣N(t)(x1,⋯,xn∣n)∏k=1n(2εk)P(Sk∈(xk−εk,xk+εk)∣N(t)=n)tnn!
这恰好是
(
U
(
1
)
,
⋯
,
U
(
n
)
)
(U_{(1)},\cdots,U_{(n)})
(U(1),⋯,U(n))的联合密度函数。
3.复合Poisson过程
复合Poisson过程指的是,对于一系列独立同分布于
G
(
x
)
G(x)
G(x)的随机变量
ξ
i
\xi_i
ξi以及参数为
λ
\lambda
λ的泊松过程
N
(
t
)
N(t)
N(t),称随机过程
Z
=
(
Z
(
t
)
,
t
≥
0
)
\boldsymbol Z=(Z(t),t\ge 0)
Z=(Z(t),t≥0)为复合Poisson过程,其中
Z
(
t
)
=
∑
i
=
1
N
(
t
)
ξ
i
Z(t)=\sum_{i=1}^{N(t)}\xi_i
Z(t)=i=1∑N(t)ξi
如果
N
\boldsymbol N
N和
ξ
i
\xi_i
ξi相互独立且
E
ξ
i
=
μ
,
D
ξ
i
=
σ
2
E\xi_i=\mu,D\xi_i=\sigma^2
Eξi=μ,Dξi=σ2,那么
-
E ( Z ( t ) ) = μ λ t E(Z(t))=\mu\lambda t E(Z(t))=μλt;
E Z ( t ) = E ∑ i = 1 N ( t ) ξ i = ∑ n = 0 ∞ E [ ∑ i = 1 N ( t ) ξ i ∣ N ( t ) = n ] ⋅ P ( N ( t ) = n ) = ∑ n = 0 ∞ ( ∑ i = 1 n E ξ i ⋅ P ( N ( t ) = n ) ) = ∑ n = 0 ∞ ( n μ ⋅ ( λ t ) n n ! e − λ t ) = μ λ t \begin{aligned} EZ(t)=&E\sum_{i=1}^{N(t)}\xi_i\\ =&\sum_{n=0}^\infty E[\sum_{i=1}^{N(t)}\xi_i|N(t)=n]\cdot P(N(t)=n)\\ =&\sum_{n=0}^\infty \left(\sum_{i=1}^n E\xi_i \cdot P(N(t)=n)\right)\\ =&\sum_{n=0}^\infty \left(n\mu \cdot\frac{(\lambda t)^n}{n!}e^{-\lambda t}\right)\\ =&\mu\lambda t \end{aligned} EZ(t)=====Ei=1∑N(t)ξin=0∑∞E[i=1∑N(t)ξi∣N(t)=n]⋅P(N(t)=n)n=0∑∞(i=1∑nEξi⋅P(N(t)=n))n=0∑∞(nμ⋅n!(λt)ne−λt)μλt
这里用到全期望公式
E Y = ∑ x i E ( Y ∣ X = x i ) P ( X = x i ) E Y = ∫ − ∞ ∞ E ( Y ∣ X = x ) p ( x ) d x EY=\sum_{x_i}E(Y|X=x_i)P(X=x_i)\\ EY=\int_{-\infty }^\infty E(Y|X=x)p(x)dx EY=xi∑E(Y∣X=xi)P(X=xi)EY=∫−∞∞E(Y∣X=x)p(x)dx -
D ( Z ( t ) ) = ( μ 2 + σ 2 ) λ t D(Z(t))=(\mu^2+\sigma^2)\lambda t D(Z(t))=(μ2+σ2)λt;
E [ Z ( t ) ] 2 = E ( ∑ i = 1 N ( t ) ξ i ) 2 = ∑ n = 0 ∞ E ( ( ∑ i = 1 N ( t ) ξ i ) 2 ∣ N ( t ) = n ) P ( N ( t ) = n ) = ∑ n = 0 ∞ E ( ∑ i = 1 n ξ i 2 + 2 ∑ 1 ≤ i < j ≤ n ξ i ξ j ) P ( N ( t ) = n ) = ∑ n = 0 ∞ ( n ( σ 2 + μ 2 ) + n ( n − 1 ) μ 2 ) ( λ t ) n n ! e − λ t = [ ( σ 2 + μ 2 ) λ t ∑ n = 1 ∞ ( λ t ) n − 1 ( n − 1 ) ! + μ 2 ( λ t ) 2 ∑ n = 2 ∞ ( λ t ) n − 2 ( n − 2 ) ! ] e − λ t = λ t ( σ 2 + μ 2 ) + ( λ t μ ) 2 D [ Z ( t ) ] = E [ Z ( t ) ] 2 − [ E Z ( t ) ] 2 = λ t ( σ 2 + μ 2 ) \begin{aligned} &E[Z(t)]^2\\ =&E\left(\sum_{i=1}^{N(t)} \xi_i\right)^2\\ =&\sum_{n=0}^\infty E\left(\left(\sum_{i=1}^{N(t)}\xi_i\right)^2\Bigg|N(t)=n\right)P(N(t)=n)\\ =&\sum_{n=0}^\infty E\left(\sum_{i=1}^n \xi_i^2+2\sum_{1\le i<j\le n}\xi_i\xi_j\right)P(N(t)=n)\\ =&\sum_{n=0}^\infty (n(\sigma^2+\mu^2)+n(n-1)\mu^2)\frac{(\lambda t)^n}{n!}e^{-\lambda t}\\ =&\left[(\sigma^2+\mu^2)\lambda t\sum_{n=1}^\infty \frac{(\lambda t)^{n-1}}{(n-1)!}+\mu^2(\lambda t)^2\sum_{n=2}^\infty\frac{(\lambda t)^{n-2}}{(n-2)!}\right]e^{-\lambda t}\\ =&\lambda t(\sigma^2+\mu^2)+(\lambda t\mu )^2\\ \\ &D[Z(t)]=E[Z(t)]^2-[EZ(t)]^2=\lambda t(\sigma^2+\mu^2) \end{aligned} ======E[Z(t)]2E⎝⎛i=1∑N(t)ξi⎠⎞2n=0∑∞E⎝⎜⎛⎝⎛i=1∑N(t)ξi⎠⎞2∣∣∣∣∣N(t)=n⎠⎟⎞P(N(t)=n)n=0∑∞E(i=1∑nξi2+21≤i<j≤n∑ξiξj)P(N(t)=n)n=0∑∞(n(σ2+μ2)+n(n−1)μ2)n!(λt)ne−λt[(σ2+μ2)λtn=1∑∞(n−1)!(λt)n−1+μ2(λt)2n=2∑∞(n−2)!(λt)n−2]e−λtλt(σ2+μ2)+(λtμ)2D[Z(t)]=E[Z(t)]2−[EZ(t)]2=λt(σ2+μ2) -
Z ( t ) Z(t) Z(t)具有独立平稳增量性。令 ϕ ( ⋅ ) \phi(\cdot) ϕ(⋅)为 ξ 1 \xi_1 ξ1的特征函数,任给 s < t s<t s<t和 u , v ∈ R u,v\in \R u,v∈R,得
E e i u [ Z ( t ) − Z ( s ) ] + i v Z ( s ) = E N E ξ [ e i u ∑ j = N ( s ) N ( t ) ξ j + i v ∑ k = 1 N ( s ) ξ k ] = E N ( E ξ e i u ∑ j = N ( s ) + 1 N ( t ) ξ j ⋅ E ξ e i v ∑ k = 1 N ( s ) ξ k ) = E N ϕ ( u ) N ( t ) − N ( s ) ⋅ E N ϕ ( v ) N ( s ) = E e i u [ Z ( t ) − Z ( s ) ] E e i v Z ( s ) \begin{aligned} &Ee^{iu[Z(t)-Z(s)]+ivZ(s)}\\ =&E_{\boldsymbol N}E_{\xi}[e^{iu\sum\limits_{j=N(s)}^{N(t)}\xi_j+iv\sum\limits_{k=1}^{N(s)}\xi_k}]\\ =&E_{\boldsymbol N}\left(E_\xi e^{iu\sum_{j=N(s)+1}^{N(t)}\xi_j}\cdot E_\xi e^{iv\sum_{k=1}^{N(s)}\xi_k}\right)\\ =&E_{\boldsymbol N}\phi(u)^{N(t)-N(s)}\cdot E_{\boldsymbol N}\phi (v)^{N(s)}\\ =&Ee^{iu[Z(t)-Z(s)]}Ee^{ivZ(s)} \end{aligned} ====Eeiu[Z(t)−Z(s)]+ivZ(s)ENEξ[eiuj=N(s)∑N(t)ξj+ivk=1∑N(s)ξk]EN(Eξeiu∑j=N(s)+1N(t)ξj⋅Eξeiv∑k=1N(s)ξk)ENϕ(u)N(t)−N(s)⋅ENϕ(v)N(s)Eeiu[Z(t)−Z(s)]EeivZ(s)
这就证明了独立增量性,再证明 E e i u [ Z ( t ) − Z ( s ) ] = E e i u Z ( t − s ) Ee^{iu[Z(t)-Z(s)]}=Ee^{iuZ(t-s)} Eeiu[Z(t)−Z(s)]=EeiuZ(t−s)即证明平稳增量性。证明两个随机变量 X , Y X,Y X,Y独立,只需要证明以下等式成立(择一):
E e u X + v Y = E e u X E e v Y E e i u X + i v Y = E e i u X E e i v Y Ee^{uX+vY}=Ee^{uX}Ee^{vY}\\ Ee^{iuX+ivY}=Ee^{iuX}Ee^{ivY} EeuX+vY=EeuXEevYEeiuX+ivY=EeiuXEeivY
复合泊松过程在实际应用中,常常用到全期望公式和期望与概率的关联式。全期望公式即
E
Y
=
E
[
E
(
Y
∣
X
)
]
=
E
∫
−
∞
∞
E
(
Y
∣
X
=
x
)
d
F
X
(
x
)
EY=E[E(Y|X)]=E\int_{-\infty} ^\infty E(Y|X=x)dF_X(x)
EY=E[E(Y∣X)]=E∫−∞∞E(Y∣X=x)dFX(x)
当随机变量
X
X
X为非负随机变量,即
X
>
0
X>0
X>0时,有
E
X
=
∫
0
∞
x
d
F
(
x
)
=
∫
0
∞
∫
0
x
d
t
d
F
(
x
)
=
∫
0
∞
P
(
X
>
x
)
d
t
EX=\int_0^\infty xdF(x)=\int_0^\infty \int_0^x dtdF(x)=\int_0^\infty P(X>x)dt
EX=∫0∞xdF(x)=∫0∞∫0xdtdF(x)=∫0∞P(X>x)dt
当运用全期望公式时会经常需要计算条件概率,此时要联系 S i S_i Si与 U i U_i Ui的关系换元计算。