Resilient Adaptive Filtering Based on Model Modification and Underwater Application

  Objectives  Multi-sensor information integration provides important support for positioning, navigation and timing services in extreme environments. High-precision dynamic vehicle navigation and localization requires observation models and dynamic models with the capability of quasi-real-time or real-time correction. In the process of multi-source navigation information fusion, the observed model and noise of navigation sensors vary in time and space, so it is of great theoretical importance and practical value to construct accurate functional and stochastic models.

  Methods  A resilient adaptive filtering algorithm based on information filtering is proposed for filtering systems containing both observational and dynamic model errors. In the proposed algorithm, the standard information filter without model errors is used as the primary filter, and error compensation filters for the observed and dynamic function models are constructed to compensate the errors of the two types of models, respectively.

  Results  The proposed approach emphasizes the elastic adaptive estimation of the model compensation term and the elastic combination of the state parameters, which improves the stability of the error estimates for time-varying models. Semi physical simulation experimental results show that the elasticity of the adaptive filter algorithm based on function model compensation can effectively estimate the observation model error term and vehicle dynamics model, with the error of three-dimensional position of the underwater towed body deviation less than 0.2 m.

  Conclusions  It can fundamentally eliminate the influence of two kinds of model errors and clearly improve the estimation accuracy of vehicle dynamic parameters.