Multivariate time-series prediction of pavement performance evolution using extended long short-term memory with uncertainty quantification (Retraction Notice)

In transportation engineering, road infrastructure serves as the primary carrier of traffic flow and plays a vital role in ensuring the safety, efficiency, and sustainability of the transport systems. During long-term service, pavements are continuously subjected to repeated vehicle loads and environmental influences such as temperature variations, moisture, and freeze–thaw cycles. The coupled effects of these factors lead to material aging, structural deterioration, and the progressive evolution of critical performance indicators, notably rutting and deflection, which directly affect traffic safety and ride quality. To address the nonlinear and long-term degradation behavior of pavement structures, this study developed a multivariate time-series prediction framework based on the Extended Long Short-Term Memory (xLSTM) model. The xLSTM enhances temporal representation through improved gating mechanisms and multilayer memory structures, effectively capturing the coupled effects of traffic loading and environmental factors on pavement performance evolution. Additionally, an Uncertainty Quantification (UQ) method was incorporated to assess confidence intervals and prediction reliability, transforming deterministic forecasts into probabilistic performance assessments. This framework provides a data-driven approach for modeling the service performance evolution of pavements and offers quantitative tools for condition evaluation, early warning, and maintenance decision-making. These results contribute to the advancement of intelligent monitoring and lifecycle management of transportation infrastructure.