Week 13

Here we tried to remove two rivers out of the model and look at the performance. The two rivers are kukule ganga and niriella ganga.
Considering the kukule ganga, there is a power station controlled by the CEB. so the water level of the kukule ganga is actually not related to the kalu ganga river level in a natural way because of the human interactions. It is shown in the graph below by markings of red.

Considering the niriella ganga, it doesnt have a lag compairing with the kalu ganga water level. So we decided to leave these two rivers aside and make the model.

The model weights are shown below which used to predict one hour ahead values.

The one hour lag has the highest impact.

The predicted curve (in red) and the actual curve is shown in the below figure

The prediction is following the actual curve to a good extend. The root mean square error is 13.63 cm.
In the previous case whre we used all the rivers had an error of 13.64 cm. So this model is not affected much because of the reduction of two rivers.

The boxplot and the error variation is shown below

So the other predictions also have the same characteristics

Then we tried another model by using two lags of kalu ganga for the model. As the autocorrelation plot of the kalu ganga showed in week 10, lags up to two hours are very much correlated

The model weights are shown below which used to predict one hour ahead values

The two lags have the most impact on the prediction

The predicted curve (in red) and the actual curve is shown in the below figure

The prediction is following the actual curve to a good extend. The root mean square error is 8.31 cm.

The boxplot and the error variation is shown below

Page updated

Google Sites

Report abuse