Abstract
M.Sc.
The need for a fire-induced flashover (power line arcing to the ground) probability index
for Eskom transmission (high voltage power) lines became evident soon after the
installation the Advanced Fire Information System (AFIS) in 2004. AFIS is a satellite
based fire detection system that utilizes polar and geostationary satellite sensors to detect
fires as small as 50 m x 50 m in size.
As soon as a fire is detected by either, the Terra, and Aqua Moderate Resolution Imaging
Spectro-radiometer (MODIS) or Meteosat Second Generation (MSG) geostationary
satellites close to any of the 28 000 km of Eskom transmission lines, a cell phone and
email text warning is sent out to line managers responsible for the management of the
particular section of line affected.
Between 3000 - 6000 fires are recorded annually close to Eskom transmission lines with a
fire-induced flashover rate of 100 - 150 transmission line trips per year. Fire-induced
flashovers occur when the air around high voltage transmission lines are ionised due to a
hot flame (> 500° C). As the air becomes conductive, electricity can move from the line to
the ground in the form of a lightning flash. Studies have shown that one flashover can
cause an average of three voltage depressions (dips) on the electrical transmission system,
and each voltage depression can cause damage to a customer’s production ranging between
R5000 and R150000 per dip. The aim of this study was to develop a prediction model with
the ability to accurately predict fire-induced flashover occurrences on Eskom transmission
lines in order to reduce the large amount of false alarms (SMS and email messages)
produced annually by AFIS.
The prediction model in the form of a probability index was derived from a combination of
remote sensing satellite products as well as weather forecast variables. With the MODIS
active fire product as base layer, weather forecast variables in the form of air temperature,
relative humidity, wind speed and wind direction, as well as topographical elevation and a
satellite derived vegetation condition product served as input to the predictor data set of the
model, while flashover statistics for 2007 provided the target data set within a
Classification and Regression Tree (CART) analysis.
iii
The prediction capabilities for each of the variables were evaluated based on their
prediction accuracy and Receiver Operation Characteristic (ROC) value in terms of the
validation data set. Wind speed, relative humidity, wind direction and air temperature were
shown to have the highest predictor importance and were used to develop the probability
index calculated from a logistic regression analysis. The Fire-induced Flashover
Probability Index (FIFPI) was tested through simulations of predictor variables and was
also compared to existing Fire Danger Indices (Willis et al. 2001). The FIFPI was able to
outperform most of the standard Fire Danger Indices (FDI’s) with only the McArthur
Grassland Index (MK 4) which demonstrated some prediction capability.
The importance of wind direction as an environmental component in the prediction of
flashovers became clear as it tended to decrease the misclassification rate from 4.45%
when only wind speed, relative humidity and temperature were used to 3.87% when wind
direction was added. The research has shown that wind speed, wind direction, relative
humidity and temperature can be used as an indicator of possible fire-induced flashovers
underneath Eskom transmission lines. However, additional research is needed to verify the
results from 2007. Ideally at least 3 years of data should be used.