Back to search

BEMATA-Beregningsorientert matematikk i anv.

Finding features from dependent data

Awarded: NOK 1.3 mill.

When analyzing time series it is important to have methods for detecting whether a time series is nonstationary and when nonstationarities occur. Significant zero crossing of derivatives (siZer) is a newly developed method to explore features in independe nt data. Fourier transformations can be used to get indeoendent data so taht SiZer can be used to analyze the time series. Splitting the time series in parts SiZer can be run on each part to detect whether the part seems to be stationary or whether there has beein a disturbance. By running this on different scales it can be explored when the change has happened. Markov Chain Monte Carlo (MCMC) simulation can be used to obtain samples from distributions with a huge number of variables. A 2D version of Si Zer: Significance in Scale Space (SSS) is developed. There are few methods to explore when a stationary state of the simulated data is reached. SSS will be used to develop a method for this since when stationarity is reached there will b e no more signifi cant changes in the time series. By looking at different parts of the time series, information about the correlation structure in the Markov Chain can also be obtained.

Funding scheme:

BEMATA-Beregningsorientert matematikk i anv.

Thematic Areas and Topics

No thematic area or topic related to the project