The application discloses a vehicle risk prediction method,
system, device and storage medium, relates to the technical field of intelligent transportation and Internet of Vehicles, and comprises the following steps: collecting real-time signals of vehicles, predicting a road
icing probability through a Stacking model trained by updating soft labels based on Bayes, mapping the
icing probability into a multi-level early warning, and outputting early warning information; analyzing vehicle state data by using a clustering
algorithm, identifying a high-risk
icing area and defining a spatial range, and finally directing early warning information to target vehicles in the area. The scheme combines real-time signals of vehicles and soft labels optimized by updating Bayes, trains a Stacking model, improves the prediction accuracy of the road icing probability, mines the clustering characteristics of vehicle states by using a clustering
algorithm, accurately identifies and delimits a high-risk icing area, provides support for cross-area risk pushing, and grades the output of early warning content by matching a multi-level early warning mapping mechanism, so that the
data source can be enriched and the early warning accuracy can be improved.