A traffic congestion judgement algorithm based on neural networks is proposed for urban freeway normal traffic congestion and occasional traffic congestion. The neural networks parameters are optimized by adaptive Gradient algorithm, which can ensure that the neural networks parameters converge the global optimum value and the neural networks have fast converging speed. This technology improves the judging accuracy for traffic congestion judgement. The traffic simulation software (PARAMICS) is applied to simulate urban freeway traffic congestion and produce the neural network studying sample which includes various traffic congestion data, which increases algorithm robustness. The trained neural network is applied to detect some actual traffic data. The detection results indicated that the given algorithm has better traffic congestion judgement accuracy for urban freeway, compared with some classical detection algorithms.%针对城市快速路的常发性拥堵和偶发性交通拥堵,提出了一种基于神经网络的自动判别算法.该方法利用改进的自适应梯度算法优化神经网络的权值参数,既能保证神经网络参数收敛到全局最优值,又具有快的学习速度,提高了神经网络的检测效果.利用微观交通仿真软件PARAMICS建立了城市快速路网,通过多次仿真获得了包含各种交通拥堵的学习样本,增强了算法的鲁棒性.将训练好的神经网络对多种实际的交通数据进行了仿真试验.实验结果表明,该算法在城市快速路交通拥堵判别中具有较高的检测率和较低的误报率.
展开▼
