In light of the continuous change of the state and the drift of performance of complex electromechanical systems in process industry, a dynamic state tagging method based on self-organizing feature map network is proposed. First, a multi-variable coupling relationship network is construct and the characteristics are extracted.Second, the dynamic tagging process is divided into two stages where states are updated passively and actively respectively. During the former process, the self-organizing feature map network is constantly trained to adapt to new situations and drift of performance. During the later process, the influence of the system's historical states on the network model is eliminated. Finally, the proposed method is verified effective byanalyzing the data of the real chemical production system. Experimental results show that this method is able to mark the constantly changing multiple states during the operation of complex electromechanical systems. It establishes a state marking knowledge base in agreement with the dynamic evolution process of the system, providing a basis for the identification and prediction of system state.