Traffic efficiency and safety in work zone are major concerns for traffic engineers, highway agencies and road users. Microscopic traffic simulation is an efficient method to study work zone traffic and cellular automaton (CA) model is one of the most widely adopted microscopic traffic simulation methods. Existing CA-based simulations of work zone traffic face limitations of unrealistic deceleration behaviors. Firstly, an improved CA model for heterogeneous traffic flow in work zone is proposed based on the two-lane safety driving (SD) model. Secondly, the superiority of the proposed methodology is demonstrated through comparison of driving behaviors simulated by the proposed model and conventional CA model. With the proposed model, simulated vehicles can accelerate, decelerate, keep their velocity, and change lane in a more realistic way during the whole process of passing the work zone by avoiding the limitations of the existing models. Furthermore, parametric analyses of some key parameters are performed for work zone traffic simulation with the proposed model. An optimal work zone speed limit can be identified with the proposed model based on overall consideration of both traffic efficiency and safety performance through the performance indicators such as traffic flow rate, standard deviation of vehicle velocity, and lane-changing frequency.