Method of Stepwise Smoothing Experimental Dependences for Problems of Short-Term Forecasting

Abstract

The article considers the correspondence of the step-by-step smoothing method as one of the possible algorithms for short-term forecasting of statistics of equal-current measurements of monotone functions, which represent the values of the determining parameters that evaluate the dynamics of the states of complex technical systems based on the operating time. The true value of the monitored parameter is considered unknown, and the processed measurement values are distributed normally. The measurements are processed by step-by-step smoothing. As a result of processing, a new statistic is formed, which is a forecast statistic, each value of which is a half-sum of the measurement itself and the so-called private forecast. First, the forecasts obtained in this way prove to have the same distribution law as the distribution law of a sample of equally accurate measurements. Second, the forecast trend should be the same as the measurement trend and correspond to the theoretical trend, that is, the true values of the monotone function. Third, the variance of the obtained statistics should not exceed the variance of the original sample. It is inferred that the method of step-by-step smoothing method can be proposed for short-term forecasting