中文摘要：

正确的判别信控交叉口黑点，可以显著提高安全管理效率，改善交通安全状况。传统黑点判别方法（如事故数法、事故率法）忽略了事故随机波动性的影响，没有考虑相似对象的事故均值，容易导致判别结果不准确。提出了经验贝叶斯方法，并考虑了安全可提高空间，克服了传统黑点判别方法的不足。研究选取上海市195个信控交叉口进行分析，利用广义估计方程建立交叉口事故预测模型，结合经验贝叶斯法估计事故发生的期望数，并计算交叉口可以降低的事故数。在此基础上，引入交叉口安全指数，作为安全排序的依据，并与传统的事故数法和事故率法判定结果进行比较。结果表明，以安全指数排序的结果与传统方法相比有较大的区别，传统方法会造成黑点判别结果的偏差。

英文摘要：

Appropriate hotspot identification methods can improve the effectiveness of safety management and safety level. Traditional hotspot identification methods （crash frequency method, crash rate method） ignore the influence of traffic characteristic and fluctuation of crash frequencies, which may bring forth biased results. In this study, empirical Bayes method considering potential accident reduction which could resolve the problems of traditional methods was proposed. Based on the dataset of 195 signalized intersections in Shanghai, crash prediction model was developed, using generalized estimating equation （GEE）. Safety index of intersections was calculated combined with empirical Bayes and potential accident reduction methods. It was found that, the result of ranking by safety index was significantly different from classical crash frequency method and crash rate method and classical method would result in biased estimations of hotspot.