基于KNN回归的短时交通流预测

Short-term Traffic Flow Prediction based on KNN Regression

短时交通流精确预测对于提供、诱导用户选择最适合自己的路径,提高行驶效率和交通安全、减轻交通环境负荷势,最大限度地利用道路资源具有重要意义。针对此问题,证明了KNN回归模型的普遍相合性和收敛性,在此基础上建立了大样本空间下单点短时交通流预测的KNN回归模型。实验采用平均绝对百分比误差MAPE、平均预测误差MFE和平均绝对偏差MAD 3个指标衡量了预测的有效性,结果表明K取6时,平均MAPE、平均MFE和平均MAD指标均取得最优值,预测效果较好。

Accurate short-term traffic flow predictions of great significance for reasonable road navigation and improvement of travel efficiency and traffic safety.To tackle this problem,the theoretical foundation of universal consistency and convergence for modeling short-term traffic flow forecasting as a k-nearest neighborhood nonparametric regression is verified and the model is experimented on large-scale datasets.MAPE(mean absolute percentage error),MFE(mean forecast error) and MAD(mean absolute deviation) are used to appreciate the effectiveness of the model.The results show that all the evaluation indicators of average MAPE,MFE and MAD achieve good performance providing neighbors equal to 6.