电动自行车违规行为对信号交叉口通行效率和安全具有显著的影响，是重要的安全管控对象。文中基于视频录像调查法获得的信号交叉口9 726辆非机动车数据，对比分析了3类非机动车在骑行者个人属性和时空场景影响下的违规行为特征。以骑行者个人属性、信号交叉口特征和交通流特性3方面共14个因素作为协变量，构建基于多元Logistic回归的电动自行车多类别违规行为模型，以揭示信号交叉口电动自行车闯红灯骑行、占用机动车道骑行、越线等待和逆向骑行行为机理。研究结果表明：电动自行车在信号交叉口总体违规率为44.01%，是传统自行车的1.21倍；中老年电动自行车骑行者发生4种违规行为的可能性高于青年；女性骑行者更易发生越线等待和逆向骑行违规行为，男性骑行者更易发生占用机动车道骑行；增设协管员可有效降低电动自行车在信号交叉口的闯红灯、越线等待和逆向骑行行为，但同时会增加电动自行车占用机动车道骑行的可能性；设置左转专用相位能有效降低电动自行车占用机动车道骑行、越线等待和逆向骑行行为。

The e-bike violations have significant impacts on traffic efficiency and safety at signalized intersections, and they are important safety control objects. Based on the data of 9 726 non-motor vehicles at signalized intersections obtained by video survey method, the research compared and analyzed the characteristics of violation behaviors of three types of non-motor vehicles under the influence of riders’ personal attributes and time-space scenes. Considering 14 factors as covariables, including rider’s personal attributes, signalized intersection characteristics and traffic flow characteristics, a multi-category violation model of e-bikes based on multiple Logistic regression was developed to reveal the mechanism of e-bikes running red lights, occupying motor lanes, waiting for crossing lines and reverse riding at signalized intersections. The results show that: the overall violation rate of e-bikes at signalized intersections is 44.01%, which is 1.21 times that of traditional bicycles; the model middle-aged and elderly e-bike riders are more likely to commit four kinds of violations than young ones; female cyclists are more likely to cross the line waiting and reverse cycling, while male cyclists are more likely to occupy the motorway. The addition of coordinators can effectively reduce the red light running, line crossing and reverse riding behaviors of e-bikes at signalized intersections, but might increase the possibility of e-bikes occupying the motorway. Setting the exclusive phase of left turn can effectively reduce the occupation of motor vehicle lane, line crossing waiting and reverse riding behavior of e-bikes.