Author: Ostanaqulov, Hikmatillo Akmaljon o‘g‘li; Nazarova, Gulchehra Nurmuxanbetovna

Annotation: The expansion of a metro network complicates passenger-flow forecasting because newly opened stations have no historical observations, while demand at existing stations changes over time. This article examines the operating principles of the Meta-LSTM, Metro-MGAT, and IMGSN models developed to address these problems. Annual ridership on the Tashkent Metro increased from 53.5 million passengers in 2016 to 287 million in 2025, while the network is planned to reach 79 stations and 103 kilometres by 2030. These changes require an updated approach to passenger-flow forecasting. For the Tashkent Metro, the most suitable directions include combining turnstile, electronic payment, timetable, and station-level data; estimating initial demand at new stations using data from similar stations; and regularly updating forecasts as new observations become available.

Keywords: passenger flow, Tashkent Metro, spatiotemporal analysis, graph neural network, transfer learning, incremental learning, new station, information and analytical management.

Pages in journal: 219 - 229