Abstract: An accurate and reliable model is the basis and premise for achieving fermentation process optimization. Series hybrid modeling is a relatively more effective method for fermentation process with complex reaction mechanisms, but it needs data obtained from interpolation to develop the black-box models of intermediate variables, which considerably influences the generalization performance of the final hybrid model. Therefore, in this paper, we present a simultaneous series hybrid modeling method, which can transform the black-box model development problem into a dynamic model parameter identification problem, and thus overcome the shortage that existing methods need interpolation data for the development of black-box models. By introducing multi-elite learning and adaptive inertia weight adjustment strategies, an improved particle swarm optimization (PSO) called adaptive multi-elite learning PSO (AMLPSO) is constructed to determine the parameters of black-box models. Uniform design method is used to select the structure of black-box models. Experimental study is carried out based on the practical production data from nosiheptide batch fermentation process, and the results show the effectiveness of the proposed method.