论文部分内容阅读
交通流不同状态在实际中出现的频率存在很大差异,且不同交通状态之间的误判所造成的影响程度是不同的。因此,可以认为交通状态判别是一个类不平衡及代价敏感的分类问题。本文通过分析交通状态的类不平衡特性,结合了少数类样本合成的过采样技术和阈值移动方法,在给定的代价敏感矩阵引导下,对训练样本集进行过采样和对神经网络输出值进行阈值移动,并得到最终的判别结果。通过对广深高速公路上采集的数据进行测试,实验结果表明,所提出算法可以有效降低拥堵和缓慢状态的误判率。
There are great differences in the frequency of different states of traffic flow in practice, and the degree of influence due to miscarriage of justice between different traffic states is different. Therefore, it can be considered that the traffic state discrimination is a class imbalance and cost-sensitive classification problem. In this paper, by analyzing the unbalanced characteristics of the traffic state, combined with the over-sampling technique and the threshold shift method of the synthesis of minority samples, under the guidance of a given cost-sensitive matrix, the training sample set is oversampled and the output value of the neural network The threshold moves and the final result of the discrimination is obtained. By testing the data collected on Guangshen Expressway, the experimental results show that the proposed algorithm can effectively reduce the false positive rate of congestion and slow state.