Neural-network assisted modelling of cable ampacity for dynamic load profiles

This paper evaluates the use of neural networks (NNs) to model the surroundings of a buried cable. Two types of models are implemented in this work – one “conventional” model and one that utilizes a NN to determine the thermophysical properties of the soil. The models are evaluated based on their performance on day-ahead prediction of conductor temperature on a cable buried in an air-filled pipe under a road crossing. To investigate performance in three different dynamic load scenarios, ranging from step response to rapidly changing load were imposed on a full-scale cable installation dedicated to research. The models were evaluated for the three load regimes, separately, with the NN outperforming the conventional model in two of them. Nevertheless, the results were very similar, with overlapping standard deviations in two of them. The results were therefore deemed to be inconclusive, and further refining of the method is needed.