基于集成学习的城市空间停停车难度预测

Urban Space Parking Difficulties Prediction Based on Ensemble Learning

停车位预测技术是解决城市停车难问题的一种可行方案。针对神经网络等预测模型难以应对诸如路边占道停车等复杂情况,提出了一个基于支持向量机和决策树集成的模型训练方法,不再着重预测停车位的个数,而是预测某一位置的停车难度。在每轮训练过程中拟合一个支持向量机模型,同时收集预测出错的样本,最后在误分类样本集合上训练决策树模型来提高整个模型的预测准确性。采用该方法训练了一个城市空间停车难度预测模型,并利用该模型预测了近一周时间的停车难度。实验结果显示,该方法的预测效果优于单独使用支持向量机、决策树和全连接神经网络模型,可以较好地捕捉到停车难度随时间变化的基本情况。

Parking space prediction technology is a feasible solution to solve the problem of urban parking difficulties.In view of the difficulty of predictive models such as neural networks in dealing with complex situations like roadside parking,an integrated model training method based on support vector machine and decision tree is proposed,which no longer focuses on predicting the number of parking spaces,but on predicting the parking difficulty of a certain location.In each training cycle,a support vector machine model is fitted,and the prediction error samples are collected.Finally,the decision tree model is trained on the set of misclassification samples to improve the prediction accuracy of the whole model.This method is used to train a prediction model of parking difficulty in urban space,and the model is used to predict parking difficulty in nearly a week.The experimental results show that the prediction effect of this method is better than that of using support vector machine,decision tree and fully connected neural network model alone,and can capture the basic situation of parking difficulty changing with time.