The invention relates to the field of tourism risk prediction, and particularly to a high land tourism safety risk warning method based on reinforcement learning, comprising: step S1, storing a tourism data set in a historical time period of a tourist site in a database, and based on the database, initializing a tourism risk warning indication function Q and a risk target function T; step S2, based on a tourism risk assessment factor sequence and combined with the selected tourism data set in the historical time period, training the tourism risk warning indication function Q and the risk target function T; and step 3, combined with an input time parameter and the trained tourism risk warning indication function Q, obtaining a warning action under the time parameter, and obtaining a risk assessment rank of the tourist site.