A nonlinear filter switch method based on normalized innovation square

In the nonlinear system, a simple filter is usually adopted to estimate the state. However, the traditional kind based on the framework of Kalman filter has a large estimation error, even leads to divergence, when handling with a stronger nonlinear system in spite of owning a better real-time. The other kind about particle filter and its derived algorithms has a large computational complexity, although it can deal with the nonlinearity with a better performance. Therefore, a novel switching approach based on normalized innovation square (NIS) is proposed to estimate the state of nonlinear systems in this paper. By constructing a statistic of the NIS and testing their validities in the time window, whether the algorithm at current time is valid or not is decided while the nonlinearity of this system is varying. Then, the filter is switched between two kinds of above methods. Compared with the kind of particle filter, the simulation result shows that the computing load of this approach is obviously less without affecting the performance.