用函数“arima”预测在R中产生NaN值

Question

我正在使用arima函数进行日内预测。一天中每15分钟可获得一次数据。在下面的代码中，除了for循环中的一些i值之外，它工作正常。输出显示为i = 223。预测函数产生NaN值。我检查了数据集，那里没有NaN值。请建议我如何摆脱这个问题。

predict1=c()

predict2=c()

for(i in 1:456){

  y1n=ts(nbl$nbp[(12385+(i-1)*4):(15356+(i-1)*4)],frequency=96)

  y1n[which(is.na(y1n))]=y1n[which(is.na(y1n))-96]

  arima1=arima(y1n,order=c(1,1,0),seasonal=c(2,1,1),method="CSS")

  predict1=predict(arima1,8)

  predict2[(((i-1)*4)+1):(((i-1)*4)+4)]=predict1$pred[5:8]

}

arima1

Call:

arima(x = y1n, order = c(1, 1, 0), seasonal = c(2, 1, 1), method = "CSS")

Coefficients:

         ar1     sar1     sar2     sma1

         0.0241  -0.1122  -0.1068  -0.7830

    s.e.    0.0188   0.0240   0.0224   0.0189


sigma^2 estimated as 1490:  part log likelihood = -14583.05

> predict1=predict(arima1,8)

> predict1

$pred

Time Series:

Start = c(31, 93) 

End = c(32, 4) 

Frequency = 96 

[1] NaN NaN NaN NaN NaN NaN NaN NaN


$se

Time Series:

Start = c(31, 93) 

End = c(32, 4) 

Frequency = 96 

[1] NaN NaN NaN NaN NaN NaN NaN NaN