Research methods for digitalization of transport systems using artificial intelligence

Background. There is a need to move from isolated “point” solutions to comprehensive digitalization of transport systems that integrates infrastructure-level pavement monitoring, operational traffic management, and strategic planning. To this end, it is reasonable to combine machine learning (for forecasting), genetic algorithms (for optimization), and multi-agent simulation (for robustness checking).

Purpose. To assess the effect of such integration using a set of metrics (delays, costs, risk, profit, service) and an integral objective function F.

Materials and methods. Infrastructure level: computer vision (YOLO), mAP ≈ 0.84; defect-generation forecasting (XGBoost), error ≤ 12%. Operational level: short-term traffic-intensity forecasts (LSTM/XGBoost, RMSE 8–10%) and traffic-signal phase optimization with a genetic algorithm. Strategic level: demand and tariff forecasting, optimization scenarios. The robustness of solutions was verified via multi-agent simulation; comparisons were made against baseline (“as-is”) scenarios.

Results. Total delays were reduced by 37%, overall logistics costs by 12%, and profitability increased by 10–11%; with a 20% demand increase, >90% of deliveries were completed within SLA. The integral function F improved by 22–24%. The plans demonstrated robustness and sensitivity to criterion weights.