It is one of the difficult problems in the data processing of plant systems, the automatic seismometry systems or in the speech recognition systems to detect the change of the characteristics of a stochastic process.
In this paper, a new method to detect the change and the change time in the stochastic process is presented based on the log-likelihood ratio function. If the quantity of data is very large and a rapid analysis is required as in the automatic seismometry systems, the algorithm should be simple. In this situation, the most suitable mathematical model of a stochastic process may be an autoregressive (AR) model.
In this method, both the amplitude and the spectral information are effectively utilized and a backtracking process is included which simulates the way taken by a human operator. The suitable order of the AR model and the numbers of the data in the calculation of AR model are discussed from the practical point of view.
The experimental results for seismic waves and speech waves show this method is highly reliable and practical.