IEEE Power & Energy Magazine - March/April 2021 - 64

Normalized
RMSE (%)

decide where to devote more resources to improve energy
security, for instance, by upgrading the congestion forecast
tool or by investing in a DSS for the optimization of remedial actions.

11
10
9
8

Forecast Error for
System-Level Active Power Injection

3
5
7
9
Forecast Horizon (h)

Normalized
RMSE (%)

(a)

11
10
9
8

Forecast Error for
Node-Level Power Injection: 25% Quantile

3
5
7
9
Forecast Horizon (h)

Normalized
RMSE (%)

(b)

35
34
33
32

Forecast Error for
Node-Level Active Power Injection: Median

3
5
7
9
Forecast Horizon (h)

(c)

figure 7. The dependence of prediction quality of net active power load on the forecast horizon.

Currently, Swissgrid is working on such ex-post analytics, and here we present some preliminary results from the
assessment of the net load forecast. As part of the intraday
congestion forecast process, a tool generates predictions
of the aggregate net load in the Swiss system as well as
of the net load at each transmission bus. Specifically, the
bus-level forecasts are obtained based on the aggregate
forecast with the usage of predetermined load allocation
factors. The predictions are updated every hour and have
a forecast horizon in the range of 8-24 h, depending on the
time of the day.
Figure 7 displays the normalized root-mean-square
error (RMSE) of the predicted net active power load for six
increasing forecast horizons. The errors are computed for
the period from January to April 2020. In Figure 7, (a) is
for the aggregate net load, whereas (b) and (c) correspond
to bus-level predictions. Specifically, Figure 7(b) shows the
RMSE's 25% quantile, and Figure 7(c) plots the median
value, both computed across all buses. Although the error
generally increases with the forecast horizon, the differences are rather small, and this trend is less pronounced
for bus-level predictions. In contrast to the relatively low
errors for the aggregate net load, the median of the bus-level
RMSEs is unexpectedly high, mostly due to shifts and/or
biases in the predicted time series for some of the buses.
This may be partly because of a significant amount of production resources in and/or a meshed topology of the underlying lower-voltage grids.
Furthermore, Figure 8 illustrates the ratio of RMSE for
forecasts obtained with a 1-h lead time over those with a 7-h
lead time. The smaller the ratio, the better the relative performance of 1-h-ahead forecasts. While, for most transmission

Ratio Between RMSE of 1-h-Ahead
Forecast With That of 7-h-Ahead Forecast (Active Power)
1.017

Latitude

47° 30′ N

0.906
Horizon 1 h: Higher Error
Horizon 1 h: Lower Error

47° N

46° 30′ N

46° N

50 km
20 mi
6° E

7° E

8° E
Longitude

9° E

10° E

figure 8. The geographical distribution and classification of transmission buses with regards to the relative accuracy of
1-h-ahead and 7-h-ahead forecasts.
64

ieee power & energy magazine

march/april 2021

IEEE Power & Energy Magazine - March/April 2021

Table of Contents for the Digital Edition of IEEE Power & Energy Magazine - March/April 2021