Reinforced Lattice Kalman Filters: A Robust Nonlinear Estimation Strategy

Abstract

This article introduces the Sliding Innovation Lattice Filter (SILF), a robust extension of the Lattice Kalman Filter (LKF) that leverages sliding mode theory. SILF incorporates a sliding boundary layer in the measurement update formulation, enabling the filter innovation to slide within predefined upper and lower bounds. This enhances the robustness of SILF, making it resilient to model uncertainties and noise. Additionally, a derivative-free formulation of SILF is developed using statistical linear regression, eliminating the need for Jacobian calculations. To further improve accuracy, robustness, and convergence behavior in the presence of abrupt changes in system model/parameters, SILF is reinforced with the Iterated Sigma Point Filtering and Strong Tracking Filtering strategies, resulting in the Reinforced Lattice Kalman Filter (RLKF). The experimental findings for the estimation of distorted power waveforms illustrate the superior performance of SILF and RLKF over competing methods, especially when operating in scenarios characterized by model uncertainties and noisy environments.