An Approach to Building Intelligent Transport Decision Support Systems

this paper presents a multilevel hierarchical architecture of an intelligent decision support system designed for transportation applications. The proposed architecture integrates the levels of heterogeneous data perception, know ledge and memory formation, feature engineering, forecasting, decision-making, and safety and trust assurance within a unified closedloop framework. For each architectural level, a consistent system of mathematical notation and formalized models is introduced, ensuring methodological compatibility between the levels and enabling model transferability across different application domains. The endto end implementation of all six architectural levels is demonstrated using the example of an intelligent dynamic pricing system for passenger transportation. The proposed approach covers the complete decision-making pipeline, including transactional sales data processing, threelevel train clustering, and generation of tariff policy recommendations using the Single-Player Monte Carlo Tree Search method with a multicriteria reward function. The reward function incorporates target performance indicators, including the planned average yield rate, passenger car capacity utilization coefficient, and the number of transported passengers. It is demonstrated that the mathematical representations of rewards, states, and actions within the intelligent dynamic pricing system are consistent with the unified notation framework of the proposed intelligent decision support system architecture, confirming the applicability and transferability of the developed approach to a wide range of transportation management tasks.