Green Transportation Development Optimization Based on G²-HGNN Multiple Traffic States

In intelligent transportation systems (ITSs), the collaborative optimization of multimodal traffic data is the core support for improving the efficiency of urban green travel. However, due to the dual contradictions of surging travel demand and resource mismatch faced by the green transportation system, traditional methods are difficult to effectively analyze the heterogeneous associations of multimodal traffic data. Compared with existing methods, the distribution of shared single spatiotemporal data is dynamically unbalanced, and the interactive modeling of multi-modal traffic states lacks dynamic topological adaptive modeling capabilities. To solve this problem, this paper proposes a G<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>-HGNN framework (GAN-GBC-Hypergraph Neural Network), which integrates the OD matrix of shared bicycles, online car-hailing, and buses and subways into a multi-modal traffic state hypergraph, and constructs a dynamic hypergraph with dual-domain coupling of time and space for traffic state prediction and optimization. Firstly, the spatiotemporal distribution of orders from multiple transportation parties is analyzed, and the GAN network is used to solve the problem of dynamic imbalance of shared bicycles. Secondly, the granular ball computing classifier (GBC) is used to construct feature hyperedges, and a multivariate traffic state graph of traffic mode distribution is constructed. The dynamic G<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>-HGNN framework is constructed to model the high-order relationships of the hypergraph and dynamically embed nodes. The traffic state prediction optimization is performed through the joint optimization mechanism of hypergraph convolution and topology perception. Finally, it is verified in two travel mode datasets of bimodal (working days) and scattered (rest days) in Luohu District, Shenzhen. Comprehensive experiments show that compared with other benchmark methods, the prediction accuracy, F1-Score, recall rate and precision of the scheme in green transportation planning reach 88.71%, 92.51%, 97.32% and 88.16% respectively, which reflects the superior performance of G<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>-HGNN in actual traffic planning tasks and provides data and modeling support for future urban transportation development planning.