基于Res-GRU模型的共享单车骑行量预测和影响因素分析

Analysis of Bike Sharing Riding Amounts Prediction Based on Res-GRU Model and Factors

对城市共享单车的骑行影响因素进行深入研究,并对城市共享单车的未来骑行需求量进行科学预测,不仅为共享单车企业对单车调度运营提供参考,也为监管部门的精细化和智能化共享单车监管提供决策依据。为了对共享单车骑行影响因素进行量化分析,并构建高精度的骑行量预测模型,提出基于Res-GRU深度学习网络模型,将用于卷积神经网络的残差神经网络模块融入到门控循环神经网络模型GRU中,可以进一步提高GRU模型精度;同时,首次使用互信息模型对共享单车骑行使用的影响因素进行量化分析。以上海市骑行量数据为研究案例,结果显示,3 d在线量是对共享单车日骑行量影响最重要的因素之一。传统的机器学习模型对共享单车骑行量预测精度为80.1%,GRU模型预测精度为83.7%,而Res-GRU深度学习网络模型对共享单车骑行量预测精度为90.1%,取得较为明显的预测效果。

In-depth research on the influencing factors of urban shared bicycle riding,and accurate prediction of the total future riding volume of urban shared bicycles,not only provides reference for bicycle-sharing companies in bicycle scheduling operations,but also provides a basis for decision-making to bike sharing regulatory authorities.In order to quantitatively analyze the influencing factors of bike sharing riding and construct a high-precision forecasting model for the total amount of riding,a Res-GRU deep learning network model is proposed,which integrates the residual neural network module in the convolutional neural network into the gated recurrent unit model GRU.It can further improve theprediction accuracy of the GRU model.At the same time,the mutual information model is used for the first time to quantitatively analyze the factors affecting the use of bike sharing riding.Taking Shanghai's cycling data as a case study,the results show that three-day online bike volume is one of the key factors of bike sharing riding.the prediction accuracy of traditional machine learning model is 80.1%,and the GRU model prediction accuracy is 83.7%.The proposed Res-GRU deep learning network model prediction accuracy is 90.1%.