Short Period Urban Traffic Volume Forecasts Using Speed Data

A short period traffic volume forecasting model ,which based on Kalman filtering algorithm and without assuming the state variables to be stationary ,is proposed with considering the characteristics of speed variation .On the basis of the spatial temporal evolution relationship between the traffic flow of upstream and downstream in urban road net‐work ,a time variant state transition matrix is developed from the average speed data collected in field .The new state transition matrix will replace the constant state transition matrix of the existing short period traffic volume forecasting model based on Kalman filtering algorithm .Traffic volume forecasts of 4 days on 2 real road sections were conducted ,the results show that the improved model has a better overall forecasting accuracy than the original model due to the enhance‐ment of dynamic performance .The mean absolute relative error (MARE) decreased from 7 .64% to 7 .25% and from 16 . 04% to 15 .75% ;equality coefficient (EC) increased from 0 .957 2 to 0 .960 2 and from 0 .925 0 to 0 .926 8 .For those time periods when the traffic volume changed rapidly ,the improvement is even more significant .In this case ,the mean absolute relative error reduced by 14 .8% .The results also verified the effectiveness of the proposed approach .