An increasing number of high voltage transmission lines are exposed to atmospheric icing in remote northern regions. Instrumentation able to operate unattended in this remote environment is increasingly being used in conjunction with an appropriate model to estimate transmission line icing. Hydro-Québec has implemented such a system (SYGIVRE) and rely on the Mt. Bélair icing site to collect icing data required to improve the system. Mt. Bélair, located near Quebec City, Canada offers both an instrumented high voltage transmission line and appropriate atmospheric icing instrumentation. An icing rate meter (IRM) estimates the severity of the atmospheric icing on the icing site. The hourly icing load increase is measured on a 35 mm cable of a 315 kV transmission line with load cells installed on insulator strings. The number of IRM signals are recorded for one hour periods while measurements of average temperature, wind speed and direction, and precipitation rate are also recorded on the site with the same frequency. Icing data have been collected from the spring of 1994 to 1997 and constitute the data base on which a model can be eventually developed and used at other sites.

A neural network is the first technique being used to estimate ice accretion loads from the instrumentation data. The experimental data are divided in two parts: the first part is used to train the network, while the rest is used to verify the results. The same data were also analysed using multiple regression to yield a relationship between instrumentation readings and icing loads. To allow for a comparison of the methods, the same set of data is used for the training of the neural network and for the evaluation of the multiple linear coefficients with the multiple regression.

Results show that the neural network is a more flexible system offering multiple network configurations and a choice of non-linear activation functions. Hence the design of an appropriate neural network model may require more time initially in comparison with the multiple regression approach, but the final result is a more accurate model.

#Hydro-Québec, Montréal, Québec, Canada

@Université du Québec à Chicoutimi, Chicoutimi, Québec, Canada.