In northern regions, monitoring ice loads on high voltage transmission lines is highly desirable in order to foresee failures. Artificial intelligence technique such as neural networks can be used to design a virtual instrument to estimate ice accretion loads on transmission line structures from meteorological variables and icing rate measurements. Hydro-Qu6bec has instrumented a high voltage transmission line at the Mt. B61air test site located near Qu6bec City, Canada, and the recorded data has been used to train a neural network, based on measurements of temperature, wind speed, wind direction, precipitation and icing rate, measured on a transmission line. Using back propagation, the neural network was trained to relate the meteorological instruments readings to the transmission line ice loads. Measurements recorded between the spring of 1994 and 1997 were used to optimize the network. The neural network was then validated by comparing estimated and measured ice loads on the same site for new icing events. Results were satisfactory on a few trial events using new Mt. B61air data, but the neural network needed to be verified also on data collected at different icing sites. Hence, the trained neural network is used to estimate transmission line icing using data recorded at the Mt. Valin icing site, located near Chicoutimi approximately 270 km north of Mt. Bélair. Out of five icing events recorded in November 1995, three in-cloud icing events gave a comparable accuracy, but for two freezing rain events the neural network gave unacceptable results. By looking at similar types of events at Mt. B61air, the cause of the inaccurate response is established and improvements for the neural network model are suggested
Atmospheric icing of structures is a relatively rare meteorological phenomenon often occurring in remote northern regions.
