In this paper, a new trial of the time series prediction method by use of the adaptive functions for the environmental noise and vibration system is proposed. In general, though this kind of stochastic system can not be modeled by a simple equation, some kinds of correlations can be usually found between the observed past data of the fluctuation and their future values. The present prediction method is established based on not only the information of linear correlation but also that of higher order nonlinear correlations being latent among arbitrarily chosen samples of for the time series data. More concretely, the past and future fluctuation pattern of state variable is expressed in the universal function form of series expansion type with the linear combination of the newly introduced set of the adaptive functions. Consequently, the state prediction problem is converted into the problem of predicting each expansion coefficient in the above universal function form. Especially, by using systematically the orthogonal or nonorthogonal expansion type as the adaptive functions, this method enables the effective data-processing in the present prediction algorithm. Furthermore, the various or unified types of prediction algorithm with discrete and/or continuous time series types of adaptive functions can be found to suit the various fluctuation types of the observed data. Finally, the effectiveness of our methodology is confirmed experimentally by applying it not only to the simulation models of arbitrary system but also to the actually observed road traffic noise
抄録全体を表示