Validity estimation for multi-model identification using constrained kalman filter

Multi-model approach is an effective way of modeling and identification of complex nonlinear systems that relies on problem decomposition strategy by identifying several models, which are combined in a way that each model contributes to the system output according to a certain degree of validity. Despite the simplicity of the approach and performance, the implementation does still face some challenges. Validity computation is one of these challenges as it plays a crucial role in correct identification of the underlying system and represents a key decision making tool in multi-model fault detection and isolation. In this study constrained Kalman Filter is formulated for validity computation by minimizing the global learning objective of a multi-model output. Simulation example illustrates the effectiveness of the proposed validity computation compared to other commonly used methods.