高斯过程
正态分布的基本概念
- n维正态分布 X ⃗ ∼ N ( μ ⃗ , Σ ) \vec{X}\sim N(\vec{\mu},\Sigma) X∼N(μ,Σ)
f ( x ⃗ ) = 1 ( 2 π ) n / 2 ( det Σ ) 1 / 2 e − 1 2 ( x ⃗ − μ ⃗ ) T Σ − 1 ( x ⃗ − μ ⃗ ) f(\vec{x})=\frac{1}{(2\pi)^{n/2}(\det\Sigma)^{1/2}}e^{-\frac{1}{2}(\vec{x}-\vec{\mu})^{T}\Sigma^{-1}(\vec{x}-\vec{\mu})} f(x)=(2π)n/2(detΣ)1/21e−21(x−μ)TΣ−1(x−μ)
- 对正态来说:独立 ⟺ \Longleftrightarrow ⟺不相关。所以可以从协方差矩阵来判断正态分布的随机变量是不是独立。
- 正态随机变量线性变换后依旧是正态。
Y ⃗ = C X ⃗ ⟹ Y ∼ N ( C μ ⃗ , C Σ C T ) \vec{Y}=C\vec{X}\Longrightarrow Y\sim N(C\vec{\mu},C\Sigma C^{T}) Y=CX⟹Y∼N(Cμ,CΣCT)
- n维正态 ⟹ \Longrightarrow ⟹每个分量都是正态的
- 每个分量都是正态的且相互独立 ⟹ \Longrightarrow ⟹n维正态
- 对于联合高斯分布,这个计算四阶矩的公式非常有用:
E ( X 1 X 2 X 3 X 4 ) = E ( X 1 X 2 ) E ( X 3 X 4 ) + E ( X 1 X 3 ) E ( X 2 X 4 ) + E ( X 1 X 4 ) E ( X 2 X 3 ) − 2 E ( X 1 ) E ( X 2 ) E ( X 3 ) E ( X 4 ) E(X_1X_2X_3X_4)=E(X_1X_2)E(X_3X_4)+E(X_1X_3)E(X_2X_4)+E(X_1X_4)E(X_2X_3)-2E(X_1)E(X_2)E(X_3)E(X_4) E(X1X2X3X4)=E(X1X2)E(X3X4)+E(X1X3)E(X2X4)+E(X1X4)E(X2X3)−2E(X1)E(X2)E(X3)E(X4)
- 有时二维联合正态分布还会这样表示: N ( μ 1 , μ 2 , ρ , σ 1 2 , σ 2 2 ) N(\mu_1,\mu_2,\rho,\sigma_1^2,\sigma_2^2) N(μ1,μ2,ρ,σ12,σ22)
也就是 μ ⃗ = ( μ 1 μ 2 ) , Σ = ( σ 1 2 ρ σ 1 σ 2 ρ σ 1 σ 2 σ 2 2 ) \vec{\mu}=\begin{pmatrix}\mu_1\\\mu_2\end{pmatrix},\Sigma=\begin{pmatrix}\sigma_1^2&\rho\sigma_1\sigma_2\\\rho\sigma_1\sigma_2&\sigma_2^2\end{pmatrix} μ=(μ1μ2),Σ=(σ12ρσ1σ2ρσ1σ2σ22)
高斯过程的基本概念
- 定义:随机过程 ξ ( t ) \xi(t) ξ(t)的任意有限维分布都是正态分布。
- 正态过程是二阶矩过程。
- 对于正态过程来说,宽平稳和严平稳是等价的。
- 如果正态过程是均方可导的,那它的导数过程也是正态过程。
- 如果正态过程是均方可积的,那它的积分过程也是正态过程。
- 均值为零的实正态过程具有马氏性 ⟺ \Longleftrightarrow ⟺对于任意三个时刻 0 ≤ t 1 < t 2 < t 3 0\leq t_1<t_2<t_3 0≤t1<t2<t3有, ρ ( t 1 , t 3 ) = ρ ( t 1 , t 2 ) ρ ( t 2 , t 3 ) \rho(t_1,t_3)=\rho(t_1,t_2)\rho(t_2,t_3) ρ(t1,t3)=ρ(t1,t2)ρ(t2,t3)。
其中 ρ ( s , t ) = R ( s , t ) R ( s , s ) R ( t , t ) \rho(s,t)=\frac{R(s,t)}{\sqrt{R(s,s)R(t,t)}} ρ(s,t)=R(s,s)R(t,t)R(s,t),也就是s和t时刻的相关系数。
例题
- X ( t ) X(t) X(t)正态过程,写出 ( X ( t ) X ( t + τ ) X ′ ( t ) X ′ ( t + τ ) ) (X(t)\quad X(t+\tau)\quad X^{'}(t)\quad X^{'}(t+\tau)) (X(t)X(t+τ)X′(t)X′(t+τ))的协方差矩阵。
( R X X ( 0 ) R X X ( − τ ) R X X ′ ( 0 ) R X X ′ ( − τ ) R X X ( τ ) R X X ( 0 ) R X X ′ ( τ ) R X X ′ ( 0 ) R X ′ X ( 0 ) R X ′ X ( − τ ) R X ′ X ′ ( 0 ) R X ′ X ′ ( − τ ) R X ′ X ( τ ) R X ′ X ( 0 ) R X ′ X ′ ( τ ) R X ′ X ′ ( 0 ) ) \begin{pmatrix} R_{XX}(0)&R_{XX}(-\tau)&R_{XX'}(0)&R_{XX'}(-\tau)\\ R_{XX}(\tau)&R_{XX}(0)&R_{XX'}(\tau)&R_{XX'}(0)\\ R_{X'X}(0)&R_{X'X}(-\tau)&R_{X'X'}(0)&R_{X'X'}(-\tau)\\ R_{X'X}(\tau)&R_{X'X}(0)&R_{X'X'}(\tau)&R_{X'X'}(0)\\ \end{pmatrix} ⎝⎜⎜⎛RXX(0)RXX(τ)RX′X(0)RX′X(τ)RXX(−τ)RXX(0)RX′X(−τ)RX′X(0)RXX′(0)RXX′(τ)RX′X′(0)RX′X′(τ)RXX′(−τ)RXX′(0)RX′X′(−τ)RX′X′(0)⎠⎟⎟⎞
又因为
R X ( n ) X ( m ) ( t , s ) = ∂ ( n + m ) ∂ t n ∂ s m R X X ( t , s ) R_{X^{(n)}X^{(m)}}(t,s)=\frac{\partial^{(n+m)}}{\partial t^n\partial s^m}R_{XX}(t,s) RX(n)X(m)(t,s)=∂tn∂sm∂(n+m)RXX(t,s)
所以
R X X ′ ( t , s ) = ∂ ∂ s R X X ( t , s ) , R X X ′ ( τ ) = − d d τ R X X ( τ ) R_{XX'}(t,s)=\frac{\partial}{\partial s}R_{XX}(t,s),R_{XX'}(\tau)=-\frac{d}{d \tau}R_{XX}(\tau) RXX′(t,s)=∂s∂RXX(t,s),RXX′(τ)=−dτdRXX(τ)
R X ′ X ( t , s ) = ∂ ∂ t R X X ( t , s ) , R X ′ X ( τ ) = d d τ R X X ( τ ) R_{X'X}(t,s)=\frac{\partial}{\partial t}R_{XX}(t,s),R_{X'X}(\tau)=\frac{d}{d \tau}R_{XX}(\tau) RX′X(t,s)=∂t∂RXX(t,s),RX′X(τ)=dτdRXX(τ)
R X ′ X ′ ( t , s ) = ∂ 2 ∂ t ∂ s R X X ( t , s ) , R X ′ X ′ ( τ ) = − d 2 d τ 2 R X X ( τ ) R_{X'X'}(t,s)=\frac{\partial^2}{\partial t \partial s}R_{XX}(t,s),R_{X'X'}(\tau)=-\frac{d^2}{d \tau^2}R_{XX}(\tau) RX′X′(t,s)=∂t∂s∂2RXX(t,s),RX′X′(τ)=−dτ2d2RXX(τ)
又实高斯过程的自相关函数对称,即
R
X
X
(
τ
)
=
R
X
X
(
−
τ
)
R_{XX}(\tau)=R_{XX}(-\tau)
RXX(τ)=RXX(−τ),所以得到:
(
R
X
X
(
0
)
R
X
X
(
τ
)
R
X
X
′
(
0
)
=
0
R
X
X
′
(
τ
)
R
X
X
(
τ
)
R
X
X
(
0
)
−
R
X
X
′
(
τ
)
R
X
X
′
(
0
)
=
0
R
X
X
′
(
0
)
=
0
−
R
X
X
′
(
τ
)
−
R
X
X
′
′
(
0
)
−
R
X
X
′
′
(
τ
)
R
X
X
′
(
τ
)
R
X
X
′
(
0
)
=
0
−
R
X
X
′
′
(
τ
)
−
R
X
X
′
′
(
0
)
)
\begin{pmatrix} R_{XX}(0)&R_{XX}(\tau)&R'_{XX}(0)=0&R'_{XX}(\tau)\\ R_{XX}(\tau)&R_{XX}(0)&-R'_{XX}(\tau)&R'_{XX}(0)=0\\ R'_{XX}(0)=0&-R'_{XX}(\tau)&-R''_{XX}(0)&-R''_{XX}(\tau)\\ R'_{XX}(\tau)&R'_{XX}(0)=0&-R''_{XX}(\tau)&-R''_{XX}(0)\\ \end{pmatrix}
⎝⎜⎜⎛RXX(0)RXX(τ)RXX′(0)=0RXX′(τ)RXX(τ)RXX(0)−RXX′(τ)RXX′(0)=0RXX′(0)=0−RXX′(τ)−RXX′′(0)−RXX′′(τ)RXX′(τ)RXX′(0)=0−RXX′′(τ)−RXX′′(0)⎠⎟⎟⎞
因为是实矩阵,所以
R
X
Y
(
τ
)
=
E
(
X
(
t
)
Y
(
t
+
τ
)
)
=
R
Y
X
(
−
τ
)
R_{XY}(\tau)=E(X(t)Y(t+\tau))=R_{YX}(-\tau)
RXY(τ)=E(X(t)Y(t+τ))=RYX(−τ)
所以为了直观,改下标就行,时间直接用正的 τ \tau τ,这样可以避免后面再考虑函数的奇偶性。
- 证明是不是正态过程的题:
只要证明任意维分布服从高斯分布即可:任取 ( ξ ( t 1 ) , ξ ( t 2 ) , ⋯ , ξ ( t n ) ) T (\xi(t_1),\xi(t_2),\cdots,\xi(t_n))^T (ξ(t1),ξ(t2),⋯,ξ(tn))T,证明它是一组相互独立的正态随机变量的线性组合。