面向共享单车需求预测的多模型可视比较分析

Multi-model visual comparative analysis for demand forecasting of shared bicycles

针对不同的区域如何比较不同的模型并选择最佳模型进行预测以及是否存在普遍适用的最佳预测模型等问题,本文设计了一套面向共享单车需求预测的多模型可视比较分析系统。首先,使用隐含狄利克雷分布(LDA)模型,选择具有代表性的预测区域,并在预测区域之间设计相应的字形,通过字形比较多个模型的预测性能。其次,提出一种计算城市空间区域相似度的新方法area2vec,用于比较相似区域模型预测性能的差异;设计使用网格布局算法,有效地缓解了地图中字形遮挡的问题。最后,通过案例分析和用户调查验证了该系统的有效性和实用性。

Aiming at how to compare different models in different regions and choose the best model for prediction and whether there is a universally applicable best prediction model,a set of multi-model visual comparative analysis system for demand forecasting of shared bicycles is designed.First,using the latent Dirichlet allocation(LDA)model,representative prediction regions are selected,and corresponding glyphs are designed between the prediction regions,and the prediction performance of multiple models is compared by the glyphs.Second,a new method,area2vec,for calculating the similarity of urban spatial regions is proposed to compare the differences in prediction performance of models in similar regions.Then,a grid layout algorithm is designed to effectively alleviate the problem of glyph occlusion in the map.Finally,the effectiveness and usefulness of the system is confirmed by case studies and user surveys.