A self-tuning predictor

An adaptive predictor for discrete time stochastic processes with constant but unknown parameters is described. The predictor which in real time tunes its parameters using the method of least squares is called a self-tuning predictor. The predictor has attractive asymptotic properties. If the parameter estimation converges and if the predictor contains parameters enough, then it will converge to the minimum square error predictor that could be obtained if the parameters of the process were known. The computations to be carried out at each sampling interval are very moderate and the algorithm is well suited for real-time applications. The self-tuning predictor can be used to predict processes which contain trends or periodic disturbances. Further, the algorithm can easily be modified in order to make it possible to predict processes with slowly time-varying parameters.