The off-ramps of urban expressways can assume various forms. Compared with other countries, many off-ramps in China have special Chinese characteristics; for example, the off-ramps are extremely short and connected to urban roads and intersections within a short distance. Thus, many foreign control methods may not be suitable for use in China. This paper provides a dual-phase signal timing optimization model for an intersection connected to the off-ramp of an urban expressway that has special Chinese characteristics. This model is solved by an improved genetic algorithm. To better evaluate traffic conditions, this paper also establishes methods by which stranded vehicles can be identified and the traffic flow at intersections can be predicted. The performance of the optimized signal timing is tested with PTV Vissim software, and the timing plan is calculated by MATLAB using an actual scenario in Xi’an, China. The optimized effect of the proposed model is tested under different flow conditions by changing the input traffic flow of the intersection. The results show that at the peak time, the stops, delays and queue lengths of the whole intersection are reduced by 29.0%, 27.8% and 23.7%, respectively, compared with the original plan and by 7.7%, 14.5% and 11.9%, respectively, compared with the plan from the single-phase model. This dual-phase model can prevent congestion from spillback onto urban expressways as well as reduce delays and stop times at intersections, especially in a supersaturated state. In general, this model can be used to address signal timing in ordinary urban intersections.