Condition recognition of complex systems based on multi-fractal analysis

Multifractal analysis is applied to extract nonlinear features from complex systems for condition recognition. Abnormal condition is hazardous for process industry complex system which may lead to accidents. Comparing with traditional techniques of condition recognition without concerning nonlinearity of complex system, multifractal spectrum elaborately reveals scale-invariance or self-similarity pro perties of observed data, which is one of the intrinsic characteristics of complex system. By using Multifractal Detrended Fluctuation Analysis (MF-DFA) algorithm, multifractal spectrum is calculated directly from monitoring time series data. The shape of multifractal spectrum is used to distinguish abnormal conditions from normal ones of complex system. After multi-source information fusion based on Dempster-Shafer evidence theory, the proposed approach can be used for abnormal condition recognition in process industry complex system where continuous multi-channel data are monitored. The effectiveness of the approach is illustrated using data from a simulated dataset and a chemical plant model where potential abnormal conditions are detected effectively, thus avoid severe system safety problems.