The invention discloses an intelligent fusion navigation filtering method and
system for complex environment transition. Navigation data and environment data are collected by a sensor arranged on ocean equipment, preliminary navigation state
estimation is obtained after the navigation data passes through a strapdown inertial navigation module, and local
error state estimation and filter health degree are obtained after the navigation data and the navigation data are input into a parallel sub-filter; combining the navigation data and the environment data with the health degree of the filter to serve as fusion data, inputting the fusion data into a
reinforcement learning agent to obtain a dynamic fusion weight, performing weighted fusion on the plurality of local
error state estimations by using the dynamic fusion weight, and superposing the fusion data on the initial navigation state
estimation to generate a final navigation result. According to the method, the optimization mapping relation between the environment characteristics and the filtering weight is directly learned through the
reinforcement learning agent, integrated self-adaptive adjustment from environment
perception to fusion decision is achieved, and the estimation precision and robustness of a
navigation system in dynamic complex environments such as a bay-ocean transition section are effectively improved.