Algebraic Bayesian networks belong to the class of probabilistic graphical models. They are represented as non-directional graphs with models of knowledge patterns in the nodes (KP). Each knowledge pattern contains closely related information about the subject domain, formalized in the form of an ideal of conjuncts or a set of quanta with truth probability estimates. In order to optimize the complexity of KPs through the use of scalar estimates, an approach to finding the canonical representation of KPs has previously been proposed. In this paper, the process of obtaining a canonical representation of the entire algebraic Bayesian network is proposed and studied for the first time. As a result, methods have been described that create a canonical representation based on the comprehensive KP and using chain generation. The results of this paper allow to reduce the time for calculating probability estimates in a priori inference by obtaining scalar estimates instead of interval estimates, which can be used to compute prior probability estimates or in systems where obtaining scalar probability estimates is preferable.