R语言-ARIMA模型

#下载和安装包
install.packages("forecast")
install.packages("tseries")
install.packages("readxl")
install.packages("aTSA")
library(forecast)
library(tseries)
library(readxl)
library(aTSA)

#导入数据集
Nile
#1.绘制时序图
plot(Nile)
#划分训练集与测试集
train<-window(Nile,start=1871,end=1960)
test<-window(Nile,start=1961,end=1970)

#自相关图#偏自相关图
acf(train)
pacf(train)

#一阶差分
ndiffs(train)  #判断需要几阶差分才能转化为平稳序列。
dtrain <- diff(train)    #一阶差分 

#单位根平稳性检验
ADF<-adf.test(dtrain)   #P值小于显著性水平，认为该序列显著平稳

#差分后序列纯随机性检验

for(k in 1:2)print(Box.test(dtrain,lag=6*k,type="Ljung-Box"))

#差分后序列自相关和偏自相关

acf(dtrain)
pacf(dtrain)

#手动定阶拟合

fit011 <- Arima(train, order = c(0,1,1))
fit011
summary(fit011)

fit110 <- Arima(train, order = c(1,1,0))
fit110

fit010 <- Arima(train, order = c(0,1,0))
fit010

fit111 <- Arima(train, order = c(1,1,1))
fit111
summary(fit111)

fit210<- Arima(train, order = c(2,1,0))
fit210

fit212 <- Arima(train, order = c(2,1,2))
fit212

fit012<- Arima(train, order = c(0,1,2))
fit012

#.模型自动定阶及拟合,基于最小AIC准测
fit <- auto.arima(train)
fit
accuracy(fit) 

#3.模型显著性检验
tsdiag(fit111)
Box.test(fit111$residuals, type="Ljung-Box") #残差检验，不显著：残差平稳

#4.用ARIMA(1,1,1)模型做预测
fore<-forecast::forecast(fit111, 10)#预测10期

#预测效果图
plot(fore, xlab="Year", ylab="Annual Flow")
lines(fore$fitted,col=2,lty=2)