Multi-step Identification Algorithm for the Multi-sensor System with Unknown Parameters

For the multi-sensor system with unknown observation noises and parameters, based on ARMA model, a multi-step algorithm is presented. The first step is to use recursive instrumental variable method and the method of average local model parameter estimations to get the initial fusion estimations. The second step is to get the estimations of the observation noises variance by system identification method. The third step is to take the reciprocal of the estimations of the sensor's observation noise variance obtained in the second step as the weight, use weighted recursive instrumental variable method to get the final local estimations of the parameters, use the method of average local model parameter estimations to get the final estimations. The final fusion estimations have better convergence than the initial fusion estimations. The convergence of the estimations of the parameters and noise variance are proved. This algorithm could be used to get a kind of self-tuning Kalman filter. An example shows the effectiveness of the algorithm.