Abstract

In data-based modelling, the number and quality of measurements are significant. There may be a great number of variables with different kinds of interactions and delays compared to each other. The aim of this study was to evaluate time delays using genetic algorithms (GA) and principal component analysis (PCA). Also possibilities to group variables using GA were studied. Variable selection and grouping were aimed to find the significant variables and to form groups of similar variables to decrease the number of variables used in modelling. This study is a part of Production 2010 -project financed by the Technology Development Centre of Finland.

The main idea in grouping was to identify and group strongly coupled variables automatically by using genetic algorithms. The objective function was based on correlations between variables. Time delay estimation utilized principal component analysis and genetic algorithms. Optimization of delays was performed with genetic algorithm, which utilized the results from the PCA as objective functions.

The algorithms were tested with two simulator data sets. The other data set was from the simulator of a chemical process with rather small number of variables while the other data set was from the paper machine simulator with over 50 variables. The grouping algorithm was tested with the paper machine simulator data set with good results. The results given by genetic algorithm was similar to those obtained with cross-correlation and graphical analysis. Time delay estimation algorithm was tested with both data sets. Also, different objective functions were tested. Testing showed good results.

Some work was also done with actual paper machine data. The importance of variable grouping was noticed when PCA was tested. This paper machine data will be used in practical testing of the reported method in the future.