| 摘 要: 针对传统单车需求预测研究在处理不同地区无桩共享单车需求预测中存在的特征提取不精细和地区需求预测匹配度不精准等问题,在对南京地区两个自然年度、三类典型地区共享单车的使用需求数量、天气状态、季节周期等多类数据进行采集和特征工程的基础上,提出了一种兼顾时空序列的基于ConvLSTM(Convolutional Long Short Term Memory)深度学习预测模型,该模型通过卷积操作后能够提取数据中隐含的更多空间信息,将其应用于测试集中并与经典时序LSTM(长短期记忆网络)和CNN(卷积神经网络)进行对比发现,RMSE分别提升0.05和0.04,最大误差分别提升约0.86和0.3。 |
| 关键词: 共享单车;需求预测;深度学习;ConvLSTM;交通 |
|
中图分类号: TP183
文献标识码: A
|
| 基金项目: 江苏省科研计划(产业前瞻与关键核心技术)(BE2021086);南京信息职业技术学院高层次人才科研启动基金项目(YB20221502);工信行指委重点项目(GXHZWZ13058 |
|
| Research on the Prediction of Shared Bicycle Demand in Nanjing Based on ConvLSTM |
|
WANG Jun1, YU Airong2
|
(1.School of Artif icial Intelligence, Nanjing Vocational College of Inf ormation Technology, Nanjing 210044, China;
2.College of Command Control Engineer, Army Engineering University, Nanjing 211117, China)
intraweb@163.com; yu_alice@163.com
|
| Abstract: In traditional shared bicycle demand prediction research, there are issues of imprecise feature extraction and inaccurate matching of regional demand prediction for predicting free floating shared bicycles in different regions. Aiming at these issues, this paper proposes a deep learning prediction model that takes into account the spatiotemporal sequences based on ConvLSTM (Convolutional Long Short Term Memory). The model is proposed on the basis of the collection and feature engineering of multiple types of data, such as the usage demand quantity, weather conditions, and seasonal cycles of shared bicycles in two natural years and three typical regions in Nanjing. This model can extract more spatial information hidden in the data after convolutional operations. Compared with the classical time-series LSTM network and CNN (Convolutional Neural Network), the RMSE of the test set improves by 0. 05 and 0. 04, respectively, and the maximum error improves by 0.86 and 0.3, respectively. |
| Keywords: shared bicycles; demand prediction; deep learning; ConvLSTM; traffic |
