Provided is a system and a method for emergency supervision of tunnel traffic in smart cities based on an Internet of Things (IoT) large model. The method comprises: constructing a fusion feature vector of a vehicle in a target tunnel based on multi-dimensional data of the target tunnel; determining an abnormal event based on the fusion feature vector and congestion data of the target tunnel; predicting an upstream congestion condition of the target tunnel based on the abnormal event, the congestion data, and peripheral traffic flow data of the target tunnel; and generating a signal regulation instruction including a passage duration according to the upstream congestion condition, the signal regulation instruction being configured to control a target signal light to maintain a target state for the passage duration, and the target signal light being a signal light on a key road section in a direction toward the target tunnel.