Anotation:
A set of experiments was conducted on the industrial decision support system to compare various combinations of settings. The system works with Bayesian probability mixtures. Mixtures representing the controlled system, user target and resulting advisory mixture can be of several types. Possible combinations of their types were successively tested on the real plant. Generated recommendations were not offered to operators - recommendations were just compared with operators' actions instead. Experiments helped to select appropriate and safe settings for the advisory system.