model_order = c(1,1,3)
model_seasonal = c(1,1,1)
model_period = 24
model_xreg = temp[169:816]
#kun aikasarja on stationaarinen, luodaan malli differoidulle aikasarjalle valitsemalla kertaluvut korrelaatiofunktioiden avulla
model = arima(dele168, order=model_order, seasonal=list(order=model_seasonal, period=24), method="CSS", xreg=model_xreg)


dele_t = 168
pre_ele = ele[1:(816-dele_t)]
pre_dele168 = diff(pre_ele, lag = 168, difference = 1)
pre_model = arima(pre_dele168, order=model_order, seasonal=list(order=model_seasonal, period=model_period), method="CSS", xreg=model_xreg[1:(length(model_xreg)-dele_t)])

pre_predict = predict(model, newxreg=temp[(817-dele_t):816], n.ahead=dele_t)


pre_values = append(ele[1:(816-dele_t)], (ele[(816-167-dele_t):((816-168-dele_t)+dele_t)] + pre_predict$pred[1:dele_t]))

pre_enne = ts(pre_values, frequency = 24)
plot(t[(816-dele_t):816], pre_values[(816-dele_t):816], col="red", "l")
lines(t[(816-dele_t):816], ele[(816-dele_t):816], col="blue", "l")