Integrating a Safety Factor into Ampacity Forecasting Models using Quantile Losses

Abstract

The safety and the life span of the electrical network have the highest priority. A safety factor for ampacity predictions is defined by several system operators to avoid overestimations. Applying this, they still have time to react without producing long-term damages of the line in case the forecast is higher than the actual current capacity. An example of a safety factor is the 2% overestimation rate. This article presents and compares two possible solutions to include a safety factor into ampacity forecasting systems. One is a simple bias, which is statistically calculated and then added as a constant to all predictions. The second approach can be applied to machine learning models, since it uses a quantile loss function for training, which has a predefined threshold of overestimations as a quantile. Both solutions are explained using a standard database for ampacity forecasting studies. At the end, the benefits of training based on a quantile loss function get clear, which opens a new experimentation field to get nearer to the requirements of system operators.