交通运输研究2025，Vol.11Issue(2)：93-104,12.DOI:10.16503/j.cnki.2095-9931.2025.02.008
基于GRU-KAN模型的地铁站周边共享单车需求预测
Demand Forecasting of Shared Bicycles Around Subway Stations Based on GRU-KAN Modeling
摘要
Abstract
In order to accurately predict the demand for shared bicycles at different subway stations and overcome the problems of long-term sequential dependence and the deficiencies in the assignment of feature weights in traditional models,a GRU-KAN prediction model based on the improvement of the Gated Recurrent Unit(GRU)with the Kolmogorov-Arnold Network(KAN)was proposed.The GRU module was utilized to extract the features of time series,and the KAN module realized adaptive feature weighting and learning of nonlinear relationships.Taking the subway stations in Shenzhen as the research objects,first of all,the spatiotemporal features and meteorological features of the factors influencing the demand for shared bicycles were visualized,and the Spearman correlation analysis method was used to analyze the correlations of influencing factors.Subsequently,the order data of shared bicycles,meteorological features,and travel features were input into the GRU-KAN model to construct a demand prediction model.To verify the effectiveness of the model,four representative sta-tions were selected for experiments,and compared and analyzed with models such as LSTM,BiL-STM,GRU,KAN,LSTM-KAN.The experimental results show that the prediction accuracy of GRU-KAN model is better than that of single time series prediction models and other combined models,with MAE and RMSE decreasing by an average of 19.71％and 15.20％respectively,and R2 increas-ing by 12.38％.Furthermore,the demands of all subway stations in Shenzhen at different time periods were selected for verification.The results show that the GRU-KAN model has small prediction errors at most subway stations,which verifies the prediction performance of the GRU-KAN model,provides a new method for the prediction of shared bicycle demand.关键词
智能交通/需求量预测/GRU-KAN模型/神经网络/时空特征Key words
intelligent transportation/demand forecasting/GRU-KAN model/neural network/spatio-temporal characterization分类
交通运输引用本文复制引用
马飞虎,肖婷,王雪杰,孙翠羽,李明..基于GRU-KAN模型的地铁站周边共享单车需求预测[J].交通运输研究,2025,11(2):93-104,12.基金项目
江西省03专项及5G项目(20212ABC03W04) （20212ABC03W04）
