Machine Learning for Predictive Cost and Schedule Risk Mitigation in Inter-State Transmission Line Buildouts

Abstract

Inter-state transmission projects in the United States routinely exceed their approved schedules and budgets, and the point of maximum financial exposure, sanction, when capital is irreversibly committed, is typically reached before the largest sources of delay are quantified. This paper proposes a machine-learning framework for forecasting cost and schedule risk on inter-state transmission line buildouts at the pre-sanction stage, using only information available before financial commitment. The framework draws on publicly reported data layers, including transmission plans filed under FERC Form 715, reliability and equipment-addition records from NERC and the U.S. Energy Information Administration, environmental permitting timelines, interconnection-queue position from the Lawrence Berkeley National Laboratory Queued Up series, and published equipment lead-time indices, and it combines a gradient-boosted ensemble for overrun magnitude with a survival model for milestone timing. We specify the data architecture, feature design, target definitions, and an evaluation protocol appropriate for a domain with few projects and long horizons, and we discuss governance and interpretability requirements for adoption by utilities, transmission developers, and their engineering, procurement, and construction partners. The paper is a framework and methodology contribution: illustrative figures denote expected behaviour and are not empirical results. We conclude with an implementation pathway and the data-access steps required for empirical validation.