A Velocity-Free and Memoryless PSO for Extreme Learning Machine-Based Deflection Basin Area Estimation in Industrial Cyber–Physical Pavement Monitoring

Abstract

Industrial cyber–physical systems (ICPS) for transportation infrastructure require accurate, data-driven condition assessment for intelligent monitoring and maintenance. Yet pavement responses under dynamic loading are highly nonlinear, causing conventional optimizers and learning models to stagnate in local optima. This study proposes two PSO variants—Inverse PSO (INV-PSO) and Chaotic Inverse PSO (CINV-PSO)—for robust global optimization. INV-PSO is supported by a convergence theorem and improves exploration via inverse updates, while CINV-PSO further introduces chaotic maps to enhance escape from local minima and maintain diversity. Both optimizers are used to train a lightweight Extreme Learning Machine (ELM) for predicting asphalt pavement deflection basin area from Falling Weight Deflectometer (FWD) measurements and related operational variables. Experiments show that INV-PSO-ELM and CINV-PSO-ELM outperform classical machine learning baselines and PSO-ELM in accuracy and stability. The proposed framework provides an efficient and practical cyber–physical solution that reduces the need for frequent on-site deflection testing while preserving high-fidelity pavement condition estimation.