The South African rail sector is a key contributor to the national economy, boosting gross domestic product (GDP) and creating jobs. However, serious malfunctions often jeopardize the reliability of locomotives such as the Class 8E locomotive, leading to lost output and longer lead times. Accurate forecasting of the catenary line voltage is essential to ensure timely activation of protective mechanisms and maintain the safe operation of electric traction systems under undervoltage conditions. To reduce unscheduled downtime in the 8E locomotives, this study proposes a framework that analyzes the impact of clustering methods and hyperparameter settings on artificial neural network (ANN) and adaptive neuro‐fuzzy inference system (ANFIS) models. Real‐time operational data, including line current, ambient temperature, oil temperature, and line voltage, were gathered on the 8E locomotive at Impala Platinum Mine in Rustenburg, South Africa (SA), between August and October 2024. Three distinct clustering methods, namely, subtractive clustering (SC), grid partitioning (GP), and fuzzy c‐means (FCM), along with other key hyperparameters, resulting in a total of 24 developed submodels, were examined and analyzed. The performance of the developed models was analyzed using 7 renowned statistical metrics. With a clustering radius of 0.3, the ANFIS‐SC model delivered improvements of 28.45% (MAPE), 28.64% (MAE), 20.80% (SD), 27.53% (CVRMSE), 28.11% (RMSE), and 27.50% (Theil’s U) compared to its ANN counterparts. In addition, better performance was obtained compared to the PSO‐based ANFIS model. The study demonstrates the potential of the proposed model as a reliable tool for catenary line voltage in the Class 8E locomotive rail sector in SA.
Mufamadi et al. (Thu,) studied this question.