Portal de Periódicos da CAPES

Analysis of the Structural Inhomogeneity of a Short Time Series as a Factor in Predicting a Emergency

2021; Springer Science+Business Media; Linguagem: Inglês

10.1007/978-3-030-71119-1_21

1876-1119

Oleg V. Prokofiev, Aleksandr Savochkin, M. A. Linkova,

Forecasting Techniques and Applications

When operating technically complex objects, the accidents and emergencies occurring at them should be forecasted. Such a forecast can be made by analyzing the technical state of an object. The variety of technical states of technically complex objects forms a set of emergency variables indicating the onset of a change in the TCO’s technical state – an emergency. The technical state of an object can often be characterized based on the physical process parameters recorded by sensors mounted on the object. The time series obtained by converting sensor data on a technically complex object is often short due to the nature of the impact on this TCO. Forecasting by studying a short time series becomes possible through analyzing information on the series structure patterns. There is an issue of detecting a change in the technically complex object state, which is expressed as a change in the variance of the residual component of a short time series formed by the sensor output values read with a uniform pitch (residuals, disturbances, noises). This issue cannot always be solved by visual monitoring of the sensor readings for technical (high speed of the processes) and psychological (visual effects, different interpretations of the same plots by experts) reasons. One of the tools for analyzing TSs of this size are techniques used to detect structural heterogeneity – the heteroscedasticity tests.