At present

high-speed train positioning system generally has problems of low positioning precision and poor stability during acceleration

deceleration and turning of trains（ hereinafter referred to as maneuvering）. The interacting multiple model algorithm is a feasible way to solve this problem. However

traditional interacting multiple model algorithm uses a fixed parameter Markov probabilistic transition matrix

this leads to the increase of the positioning error caused by the model lag in the process of model switching. In this regard

adopt the model error rate to define the ratio of error compression rate between three models of constant velocity（ CV）

constant accelerated（ CA） and constant turn（ CT）. And use this information to realize the adaptive adjustment of Markov probability transfer matrix parameters. It can be used as a filtering algorithm of the combined positioning system of global positioning system（ GPS）/inertial navigation system（ INS）/dead reckoning（ DR）. Simulation results show that the localization errors of the proposed algorithm is less than that of the traditional interactive multi-model algorithm.