Potential of randomization in kalman-type prediction algorithms at arbitrary external noise in observations

The problem of predicting the values of a random process is considered. The uncertainties generating the process studied are assumed to be of a statistical nature, and observations are carried out with unknown, but bounded, disturbances. A randomized algorithm, which filters out arbitrary external noise in observations, is proposed. The operability of the new algorithm at irregular noises in observations is illustrated by simulation as compared to traditional approaches.