Study On Interacting Multiple Model Algorithm Based On Fuzzy Neural Network

Target tracking is to track and locate target by computers and other apparatus by some certain algorithms, then take measures according to the position and pulse of target. Maneuvering means that target changes its former track in order to execute certain tactics intention and/or some reasons un-forethought, for instance, turning, diving, gliding, climbing, snaking, accelerating and decelerating. Maneuvering target tracking mainly resolves the problem of stable and accurate tracking when target maneuvers.The key to maneuvering target tracking is how to distill useful information of target state from measurement values. A good maneuvering target tracking algorithm is propitious to distill this information. Most of the maneuvering target tracking algorithms are based on models systematically. They can be divided into Single Model (SM) algorithm and Multiple Model (MM) algorithm.The classical Single Model algorithms include white noise acceleration (CV) model, Wiener acceleration (CA) model, singer acceleration model, and the "current" statistical model and constant turn (CT) model and so on. In this paper, the modeling method of each model is analyzed, and tracking precision is also compared with each other by MATLAB simulations.MM algorithms can deal with the situation which structures and parameters are uncertain and/or alterable. This algorithm can make the problems easy. So it is more popular nowadays. MM algorithms can be divided into three generations, as static multiple model (SMM) algorithms, interacting multiple model (IMM) algorithms and variable structure multiple model (VSMM) algorithms.At present, in many tracking filtering algorithms, IMM algorithm is regarded as one of the most effective multiple model algorithms. So it is widely used in the area of maneuvering target tracking. But when the maneuvering occurs, the precision of tracking will reduce or exhale when it mismatches the model sets of IMM algorithm with the move modes of current target. Therefore, how to predict model parameters betimes according to maneuvering of target is important to improve the tracking performance of IMM algorithm. As fuzzy neural networks have the abilities of self-learning, association and optimized structure of neural network and advantages of easy-understanding of fuzzy logic. It plays an important role in maneuvering target tracking when introducing them into IMM algorithms.There are two improvements of IMM algorithm in this paper. First, a new IMM algorithm is proposed, which only uses two models (a CS model and an augmented CV model) for interaction. Second, it is designed a fuzzy neural network which accords with the characteristics of target maneuvering in maneuvering target tracking. The characteristic quantums of movement model are used to be inputs of fuzzy neural network, and target movement model parameters can be amended by its outputs. So the system can self-adapt to the complex and variable movement models of target. This fuzzy neural network connects each node according to fuzzy regulation. The fuzzy regulations (as the weighted-values of network) can adjust and amend themselves through neural network. The mismatching problem between maneuvering target movement mode and target tracking model can be effectively resolved by this algorithm. It reduces computing and improves tracking precision. MATLAB simulation experiments show this algorithm is valid.