An Adaptive Variable Structured-Based Filter Using Multiple Models

Abstract

—The Kalman filter (KF) is the most well-known estimation strategy which yields the optimal solution in terms of error to the linear quadratic estimation problem for linear, known systems in the presence of Gaussian noise. While the KF is effective under the stated conditions, it lacks robustness to disturbances which are prevalent in real-world applications. Since its inception 60 years ago, there have been numerous variants of the KF developed to accommodate nonlinear systems, non-Gaussian noise, and modeling uncertainties. The smooth variable structure filter (SVSF) is as an alternative to the KF with improved robustness, especially in the case of external disturbances. It is based on sliding mode techniques that offer robustness at the cost of optimality. The static multiple models estimator incorporates several possible operating modes and generates an estimation that is weighted based on the likelihood of each mode. This paper introduces an adaptive formulation of the SVSF based on static multiple models, and applies the developed strategy on an electrohydrostatic actuator.