Two improved models of control rules were proposed. A model in a form of AND/OR graph; in contrast to the known graphical model of general rules, is based on dividing the rules into groups based on the controlled object state. The graph contains special markup that allows to convert the graph paths corresponding to the rules into Boolean expressions including formulas for both and inverse rule sets. The basic formulas of the rules model in a form of Boolean expressions cannot be constructed for general rules and based on these formulas the three methods for verification of the rules were developed: ‒ the method for verifying the control rules premises for inconsistency based on the SAT problem for Boolean formulas; ‒ the method for verifying the control rules for completeness based on visualization of both direct and inverse rules with conclusions in “inverse” rules opposite to the conclusions of the original rules; ‒ the method for verifying reachability of the object state vertices from the control rules. The main advantage of these methods is that they allow to find errors in the rules at early stages when specialists in the field for which the knowledge-based system is used (experts and decision makers work with them). The specificity of the control tasks makes it possible to do this effectively from the point of view of analysis and verification of the rule quality. The developed procedure of the control rules verification and correction assists in to bringing together and placing in a correct order various types of verification and correct errors in an automated mode. Main components were proposed for knowledge-based control systems: the rule editor for knowledge engineers and experts and the control system itself which includes extraction of the controlled object parameters essential for analysis as well as analysis of these parameters and their transfer to a DM for making a decision. A rule editor has been developed and control systems for two domains: safe operation with electric installations and control of computer networks. The presented experimental results on the management of the training process using the developed systems have shown that the number of errors in the created rules was reduced. When verifying for reachability of the object states, errors in an average of 5.4 % control rules were found and removed. When verifying for inconsistency of the rule premises, errors were found and corrected on average in 11.5 % of rules. When verifying for completeness, the rules base was expanded by on average of 12.3 %. In addition, due to consulting, verification and correction of the rules, the time spent by trainees on execution of their work was reduced by an average of 8 %.