Application of graph neural networks for transport route analysis

Abstract

the purpose of the article is to substantiate and practically verify the approach to the analysis of transport routes based on graph neural networks in the tasks of short-term forecasting of the state of the urban network. The focus of the research is to increase the accuracy of route travel time prediction, early detection of potentially congested areas and support for operational decisions of dispatching management. An additional goal is to assess the suitability of modern spatiotemporal architectures for operation in conditions of non-stationary demand, partial incompleteness of data and the influence of external factors (weather, incidents, calendar peaks). In the work, the transport system is presented as an oriented weighted graph, in which nodes correspond to key elements of the infrastructure, and edges to possible directions of movement. The forecasting problem statement for time horizons of 15–60 minutes was formulated, input features and target quality metrics (MAE, RMSE, MAPE) were determined. A comparative analysis of modern approaches (DCRNN, T-GCN, Graph WaveNet, etc.) was performed and a model with an adaptive adjacency matrix and a temporal attention mechanism was selected. A computational experiment was conducted on a synthetically realistic dataset of the urban network, including the “morning peak + incident” scenario. The results obtained demonstrated higher accuracy of the selected approach compared to the baseline models, as well as better resistance to noise and gaps in telemetry. The results of the study show that graph neural networks are an effective tool for practical tasks of transport analytics and predictive route management. The proposed approach provides better consistency of forecasts on longer horizons, allows for timely identification of critical network sections and formation of justified management actions. The practical value of the results lies in the possibility of their integration into decision support subsystems to reduce delays, increase transportation reliability and improve the quality of transport service..