Short-Time Traffic Flow Forecasting Based on the K-Nearest Neighbor Model

Abstract

In order to accurately forecast the short-term traffic flow, a K-nearest neighbor model was set up, study the influence of key factors in model on predicted result. Using combination of different state vectors and distance metrics to form four kinds of K-nearest neighbor model, combined with Beijing real road traffic data apply four kinds of models to carry out example verification, pick up relative error and equalization coefficient evaluate forecasting result. The results show that: with time-space parameters of a higher prediction accuracy, which has minimum prediction error, average 7.8%. take index weights into distance metric can be more precise in neighbor selection, which has minimum prediction error, average 7.34%. Visibly, compared to traditional K-nearest neighbor short-term traffic flow forecasting model which only consider the time dimension, K-nearest neighbor model with time-space parameters and index weights more accurately reflect the state of the traffic flow change condition, can be used as effective road traffic flow forecasting means.