Prediction of Induced Voltage on Metallic Pipeline due to AC Power of OHTLs Via a Hybrid Feed-Forward and Radial Basic Function Neural Network

Electromagnetic interference (EMI) from high-voltage power systems can significantly jeopardize adjacent conductive structures, such as trains, communication lines, and pipelines, potentially undermining system integrity and operational safety. Accurately predicting the degree of induced voltage is crucial for designing effective mitigation systems for metallic pipelines. Researchers can estimate electromagnetic fields (EMF) with remarkable accuracy in a brief period of time using Artificial Intelligence (AI) techniques. Three Artificial Neural Network (ANN) models are introduced in this paper that were created to estimate the voltage induced by AC power from Overhead Transmission Lines (OHTLs) in pipelines. The models were trained on a dataset of pipeline-induced voltage measurements and OHTLs parameters to predict the induced voltage based on these features. The models include a FeedForward Neural Network (FFNN), a Radial Basis Function Neural Network (RBFNN), and a Hybrid Neural Network (HNN). A sensitivity analysis was performed on the hyper-parameters of these models to identify the ideal configuration for enhanced accuracy and response time. The HNN model significantly outperformed FFNN and RBFNN in predicting pipeline-induced voltage, demonstrating an impressive decrease of 90.92848% and 56.59041% within Root-Mean-Squared Errors (RMSE), respectively. This makes HNN model a promising choice for pipeline-induced voltage prediction, outperforming methods proposed in other recent studies. After training, the model is tested with a separate dataset, and its accuracy and speed for new data points are evaluated. The model can predict induced voltage with nearly 97.15% accuracy within 15 milliseconds. These results show that the hybrid method outperforms existing AI-based techniques, with sensitivity analysis revealing that HNN models with hidden layers of triple and double are the most effective for pipeline-induced voltage prediction.