The multi-agent model for constructing process control intelligent systems is discussed in the paper. Agents of the model are based on three paradigms: pattern recognition, formal (string and graph) automata and rules. The efficient syntactic pattern recognition schemes are used for analysing string and graph structures that represent a structured knowledge. For string-like structures DPLL(k) quasi-context sensitive languages are applied. Graph structures are analysed with ETPL(k) graph parsers in a polynomial time. Grammatical inference algorithms can be used for both kinds of structures. It allows one to embed self-learning schemes in agents of the model.