Distributed particle filter for state estimation of hybrid systems based on a learning vector quantization algorithm

Traditional centralized state estimation algorithms pose stringent scaling restrictions for modern distributed hybrid plants due to their enormous communication overhead requirements. This paper presents a novel distributed estimation approach for hybrid systems composed of a proposed distributed particle filter based on a learning vector quantization algorithm. The proposed approach makes use of a particle filter estimation engine to estimate locally the mode and continuous state of hybrid system in each sensor location or node. The distributed nature of the algorithm is handled by quantizing the modes with a number of generative probabilistic models and transferring the associated parameters in the network of sensors. The sharing data in the network can provide the essential information needed to enhance the overall state estimation and at the same time, the low amount of shared data helps to achieve substantial saving in the communication load.