Biological fermentation process is a complex nonlinear dynamic coupling process. As it is very difficult to measure the key biological parameters on line, the process control is unavailable to industrial production in time. In this respect, however, soft sensing can solve the above problem. To overcome some drawbacks of PSO and FNN, such as falling into local minimum occasionally and slow convergence speed, the extremum disturbed particle swarm optimization (tPSO) algorithm is proposed and then combined with fuzzy neural network (FNN) to optimize the network parameters. Furthermore, the tPSO-FNN is applied in the soft sensor modeling of lysine biological fermentation. Experiment results show that the model proposed could measure the key parameters. And the soft sensor model based on tPSO-FNN has higher precision and better performance than the model based on FNN.