With rising demands in industry for reliable electrical cable distribution networks comes an inherent need for utility providers to know well the condition of the assets in their network. Heightened expectations from regulators and consumers require methods of reliability assessment to improve the processes adhering to performance-based ratemaking models.

The present work is a comparative assessment of survival modelling techniques in a real-world application. This work investigates the incorporation of class balancing methods in survival analysis models, and how to assess model performance in the context of time-to-event failure probability. Numerous modelling strategies are assessed in the context of calibration and discrimination performance, leading to the choice of a model that most accurately describes individualized asset survival and hazard functions.

A method is developed for a multi-stage approach to risk scoring and failure prediction for underground, medium-voltage, power distribution cable. The method applies data class balancing techniques and survival analysis models to accurately describe the survival probability of individual cables in a distribution network. Class balancing is applied to the failure state of the cables in the dataset using both under-sampling and over-sampling methods. The balanced data forms the input data for various machine learning survival analysis models and the performance of the models are compared against models trained using the original, highly unbalanced, dataset. The resultant survival models are evaluated for their efficiency, discrimination power and overall goodness-of-fit to determine the best suited model given the data. Additionally, examination of cable parameters leads to the determination of the importance of cable properties and operating conditions that most significantly influence the model and the failure likelihood of cables. The information is used to generate individualized hazard and survival functions for specific cable instances with their unique covariate conditions.

The work concludes with a discussion of the implementation of the comparative assessment in the utility providers’ data analytics and reliability processes, suggestions for the supplementary capabilities for the annual cable testing program, and recommends further work and strategies to create a more all-encompassing strategy for cable reliability analytics.