Multiscale Analysis of False Neighbors for state space reconstruction of complicated systems

摘要

This paper introduces Multiscale False Neighbors Analysis (MSFNA) as a supporting tool for state space reconstruction for real-data based modeling techniques. Common false neighbors analysis evaluates uncertainty in mapping of input data to output data for a single setup of radii that define neighborhood and whose correct definition is usually unknown. Contrary to common false neighbors analysis, MSFNA evaluates uncertainty of input-output mapping data by evaluation of false neighbors along the whole intervals of radii that results in overall characterization of uncertainty in input-output data. The power-law concept is applied to the MSFNA as a supportive technique for characterization of uncertainty in data. The proposed MSFNA is demonstrated on comparison of various estimations of state vectors of an artificial plant as well as a real power plant coal burning furnace.