Avaliação estatística de algoritmos de exceção e compressão de dados aplicados em sistema de gestão de dados industriais.

Abstract

The definition of exception parameters and data compression for the PI System is intended to optimize the compression and storage of industrial data. In order to adjust the settings of both algorithms, they were replicated and validated in a data analysis environment and used as a simulation environment. In this environment, numerical experiments were performed for the data compression process applied in the PI System. Sensitivity analyzes for both exception and data compression algorithms were performed, adopting statistical metrics to assess the reconstruction of data from compressed data. A flash tank simulation for a mixture of hydrocarbons was built in SimCentral process simulator. A control loop for the level fraction was defined and data collected from this variable underwent compression process analysis. Statistical control charts for mean and amplitude were constructed for tank level under two operating conditions. These charts were used to verify the impacts of the exception and data compression on conclusions obtained from the control charts. The analysis of the performance of both algorithms in a simulation and analysis environment allowed the definition of adjusted parameters based on the intervals defined by the indicators used. The result obtained for the tank level data compression was satisfactory, the pairs ExcDev and ExcMax for exception and CompDev and CompMax for compression were defined, reaching mean square error of 0.000638% and reducing 80% of the original points. From the use of 20% of the points, the reconstructed data were able to maintain the statistical conclusions shown from the raw data. It was found that the statistical indicators were crucial in defining the parameters, as they delimited the acceptable limits of data exclusion, keeping the necessary information to maintain the statistical conclusions