Predictive-comparative framework for construction cost control using long short-term memory and digital twin technologies

Abstract

Fragmented project records and delayed validation hinder proactive cost control in construction. Yet current research offers limited guidance on converting inconsistent, lagged evidence into timely and auditable control signals, which can delay or complicate corrective action in practice. This paper develops a predictive-comparative framework integrating long short-term memory (LSTM) cost forecasting with digital twin (DT) progress verification for early cost governance. A sequence-to-sequence LSTM generates probabilistic cumulative cost trajectories, while the DT workflow reconstructs verified progress from digital models and translates it into cost-weighted earned value. Forecast, verified, and baseline curves are aligned on a monthly scale, and tolerance-based deviation filtering suppresses transient noise and flags sustained divergence. Synthetic experiments and a 24-month case study show strong alignment with realised cumulative costs and timely identification of sustained divergence. The framework provides a practical basis for cost governance and is extensible to live DT data streams and adaptive thresholds.