1 of 12
In this problem we are trying to understand and model the factors that influence the total daily electricity consumption for a large area in the UK. The data consists of 8 years of continuous data, the obvious annual cycles being seen in the screenshot. 2 of 12
Initially we model the load with 'day of the week' as the only model input. By looking at a running average of the model error it becomes clear that there is a strong annual cycle in the data. 3 of 12
The weather has an annual cycle and would also strongly influence electricity consumption for heating requirements.
We add average daily temperature to our model and it can be seen that this has reduced the cyclical nature of the error.
There is still a clear annual cycle though. 4 of 12
Another factor is added with a repeating annual cycle, which is a 'time of year' influence.
The actual model errors now clearly show days around Christmas and several easily identifiable 'Bank Holidays' as standing out from the 'norm'. 5 of 12
Inputs are added to 'flag' known national holidays. The average error now shows that there is a slow chaging 'base load' over time, due to growth. We can add a factor that will extract slow changing base levels. 6 of 12
The error now shows days around Christmas as anomalies. 7 of 12
Zooming in on a plot of the actual and modelled data we can investigate what is happening. 8 of 12
The error shows clearly a holday influence for several days around the official Christmas holiday. 9 of 12
An investigation helped in understanding the nature of this Christmas effect, for which model inputs were created.
The model error is now approaching a flat line, or 'white noise', which is what we are seeking. 10 of 12
As well as running errors, Tiberius can plot lagged errors. It can be seen that there is something happening in the middle of summer, which is infact annual school summer holidays. 11 of 12
As influences are discovered and accounted for the model becomes better. 12 of 12
xy plots can be be created to discover further outliers.
The 16th Oct 1987, which is clearly seen, saw the worst storm in the UK for 250 years, causing power blackouts.