Two new algorithms for sequential parameter estimation with unknown but bounded noise

Two algorithms for sequential parameter identification when the measurement errors are not statistically described are introduced. These are the projection estimate algorithm and the central estimate algorithm. Their convergence properties are illustrated, and a comparison with existing algorithms is performed. In practical applications it seems that the best choice for sequential parameter identification is the algorithm that gives the projection estimate. It is computationally lighter than the algorithm that computes the central estimate, and requires only the knowledge of the relative weights of the measurement errors and not their actual values