Recursive data analysis in large scale complex systems

Abstract

<div lang="en" class="abs"><p class="abs_header">Abstract</p><p>Advanced data analysis is needed in practical applications in large scale complex systems. Variable specific datadriven solutions provide consistent levels, which can be used in compact model structures. In changing operating conditions, the recursive analysis extends the applicability of these structures in building and tuning dynamic and case-based models for complex systems since the meanings change more frequently than the interactions. The methodology provides information about uncertainty, fluctuations and confidence in results. The scaling approach brings temporal analysis to all measurements and features: trend indices are calculated by comparing the averages in the long and short time windows, a weighted sum of the trend index and its derivative detects the trend episodes and severity of the trend is estimated by including also the variable level in the sum. The trend episodes and temporal adaptation of the scaling functions with time are used in the early detection of changes in the operating conditions. The levels are understood as fuzzy labels and the decision making is based on fuzzy calculus. The solution is highly compact: all variables, features and indices are transformed to the range [-2, 2] and represented in natural language which is important in integrating datadriven solutions with domain expertise.</p></div>