Abstract: Aiming at the features of urban traffic state having many influencing factors and difficulty to distinguish，by analyzing K-means clustering algorithm and Probability Neural Network（PNN），and using complementary of multiple sources detection information，this paper proposes probability neural network ensemble model based on rapid global clustering technique to improve diversity of the neural networks ensemble.While using Principal Component Analysis（PCA） to optimize the PNN structure.The simulation results show that the method can use more simple network structure，quickly and efficiently distinguish the urban traffic state than traditional integration methods Bagging，be able to provide data reference for future urban traffic warning and induction.

Key words: global K-mean clustering, Probability Neural Network（PNN）, Principal Component Analysis（PCA）, neural network ensemble

摘要： 针对城市道路交通状态影响因素多、判别难的特点，在分析K-均值聚类算法和概率神经网络（PNN）的基础上，利用多源检测信息的互补性，提出一种基于快速全局聚类分析的概率神经网络集成模型，通过聚类提高集成网络间的差异度，同时利用主成分分析（PCA）优化概率神经网络结构，仿真实验表明该模型与传统的集成方法Bagging相比，能够利用更简单的网络结构，快速有效地识别出城市道路交通状态，为交通预警和诱导策略的制定提供数据依据。

关键词: 全局K-均值聚类, 概率神经网络, 主成分分析, 神经网络集成