基于机器学习算法的网约车需求量预测分析.PDFVIP

基于机器学习算法的网约车需求量预测分析 陈笑微，郑宏煜 （浙江大学建筑工程学院，浙江 杭州 310058） 摘 要：近几年来，网约车平台蓬勃发展。网约车平台相比于传统出租车最大的优势在于其具有强大的资源 调配能力，而与此息息相关的是对特定区域网约车需求量的预测。但是由于详细的网约车数据无法获得等 原因，少有论文对网约车的需求量进行分析。 机器学习算法现在已经广泛应用到交通领域中，特别是对短 时交通流的预测，但基于此于交通需求量预测的论文还较少。本论文将在滴滴网约车数据的基础上，提取 出时间、天气、空气质量、费用等属性，利用长短期记忆模型、随机森林、支持向量机、递归受限玻尔兹 曼机神经网络对杭州各区域的网约车需求量进行预测。同时将各种预测结果进行对比，并给出一些建议。 关键词：机器学习算法；需求量；网约车；长短期记忆模型；随机森林 中图分类号：U491.111 文献标识码：A Prediction and analysis of the on-demand service demand based on machine learning algorithms CHEN Xiao-wei,ZHENG Hong-yu (College of Civil Engineering and Architecture, Zhejiang University, Hangzhou 310058, China) Abstract: In recent years, on-demand service platform flourish quickly. Compared with traditional taxi, the greatest advantage of on-demand platform is its powerful provisioning capabilities. And the car demand forecasting in specific area is closely related to the deployment of the on-demand service. For the lack of details about vehicle data and other reasons, few papers analyze about the demand for on-demand platform. Machine learning algorithms have been widely utilized in the traffic field, particularly for short-time traffic flow prediction, but seldom for the traffic demand estimation based on machine learning. This paper will extract time, weather, air quality, and fees information based on vehicle data from on-demand platform. By using models of LSTM(Long-Short Term Memory), random forests, support vector machine, RNN-RBM （Recurrent Neural Network-Restricted Boltzmann Machine）,this paper estimate the car demand of several regions in Hangzhou, and propose some advice while comparing the estimations. Key words: machine learning algorithms; demand; on-demand platform: LSTM; random forest 本论文使用现有4种较为成熟的机器学 1. 引言