The invention relates to a self-learning based geometric auxiliary line adding method. The method comprises the following steps: creating an auxiliary line knowledge base; coding relationship state space, and comparing with an original state space to determine whether the relationship state space is similar with the original state or not; if so, selecting an auxiliary line, and solving; updating the state space after successful solving, and finishing the whole process; otherwise, sequencing the auxiliary lines generated in the auxiliary line knowledge base by utilizing Monte Carlo tree search and branch shearing integrated method as a value network; selecting a proper auxiliary line, and adding; then solving; updating the state space after successful solving; in case of failed solving, re-selecting the auxiliary line from the sequenced auxiliary line knowledge base; adding the auxiliary line; then solving. According to the method for automatically adding the auxiliary line, the success rate of automatic adding of the auxiliary line is greatly increased, and moreover, the efficiency is obviously increased.