基于(SAGA-FCM)-PNN的交通状态判别方法研究

A Method of Discrimination for Traffic State Based on (SAGA-FCM)-PNN

为了提高城市道路交通状态判别的正确性与稳定性,研究了一种基于遗传模拟退火算法改进的FCM算法与概率神经网络(PNN)结合的短时交通流状态判别方法。针对传统FCM算法会收敛到局部最优解的问题,利用遗传模拟退火算法对其进行改进,优化算法初始聚类中心;将已分类的数据分为训练集与测试集对概率神经网络(PNN)模型进行训练与测试,通过对径向基函数的扩展速度的优化提高PNN算法的准确性;并利用厦门市城市道路地磁检测数据对模型进行实例验证及性能分析。结果表明,文中方法能够有效的实现交通状态的判别,且能够得到全局最优解;同竞争神经网络模型、GRNN模型、SVM模型相比,文中模型的交通状态判别正确率分别提高2.1%,4.5%,2.7%,且具有更好的稳定性。

In order to improve accuracy and stability of discrimination for traffic state,a method to discriminate short time traffic flow is studied based on SAGA FCM and probabilistic neural network(PNN).SAGA can solve problems that traditional FCM converges to the locally optimal solution,and improves the initial clustering center.Classified data is divided into training set and test set to trained and tested PNN.The accuracy of PNN is improved by optimizing expansion speed of a radial basis function.A case study is carried out by using geomagnetic detection data of Xiamen urban roads to verify and analyze the model.The result shows that the proposed method can effectively discriminate the traffic state,and obtain the global optimal solution.Compared with competitive neural network model,GRNN,and SVM,the accuracy of this model is improved by 2.1%,4.5%,and 2.7%,respectively,with better stability.