Power Grid Path Prediction via Regression: A Transformer-Based Deep Learning Approach

Effective design and maintenance of power grids depend on the accurate geospatial projection of the paths of electrical networks comprising conductors and cables. The accurate prediction of unknown geographical trajectories of conductors and cables is crucial for enhancing grid resilience against power outages and integrating renewable energy sources by ensuring optimal energy flow. Conventional research studies have primarily addressed the path prediction issue as a link prediction classification task that compromises the precision of geographic trajectory-based mapping. Moreover, instead of concentrating just on the binary link between the initial and final positions, a multi-line string including many consecutive latitude and longitude values is crucial for accurately mapping the conductors. Thus, we develop a novel approach for overhead line route prediction by considering the problem as a regression task. We preserve the exact geographic trajectories of conductors by using a transformer-based deep learning model to forecast the continuous spatial coordinates of grid routes from multi-line string data. Numerical investigations indicate that our model performs efficiently in terms of regression performance indicators. Our research underscores the advantages of multi-output regression in precisely predicting the interrelated outputs concurrently and establishes a new benchmark for grid design by improving the accuracy of infrastructure mapping.