Evaluation of Temperature Sensor Placements for Machine Learning Applications in the Thermal Rating of a 400-kV-Cable System

Abstract

In cooperation with Vienna's grid operator, a 400 kV test setup was installed under real conditions on which different steady-state and dynamic load tests were carried out over a period of three years. The setup is equipped with a sophisticated measuring setup containing more than 90 sensors to capture temperatures, soil characteristics and environmental parameters. Among other things, an evaluation of the temperature sensor positions and their influence on machine learning applications in thermal rating was carried out. These have shown that the considerable dependence of the distance to the cable system and its influence on the resulting thermal rating, as is the case with conventional methods, can be overcome to a reasonable extent by the application of machine learning. The investigations have shown that all positions described in this paper can be used within the cable trench and that even external sensor position can be used in machine learning applications to derive reasonable thermal models. These findings could thus contribute to a potential improvement of the robustness of DTS system installations for power cable systems.