The mean absolute relative error of the prediction is reduced by 3.31% compared to that of network trained with raw data without filter.

• 对实测交通量采用奇异值分解的滤波方法进行除噪预处理，并通过训练径向基函数网络进行预测。