The invention discloses a method for carrying out soft-sensing on the lysine fermenting process on the basis of chaos particle swarm optimization (CPSO). The method comprises the following steps of: firstly, establishing a historical database of primary variables and auxiliary variables of the lysine fermenting process and selecting the historical auxiliary variables with consistent degrees of association; combining the primary variables and the auxiliary variables of the fermenting process of a current batch to be predicted with the primary variables and the historical auxiliary variables ofthe historical fermenting process to form a least squares (LS) support vector machine soft-sensing training sample database and constructing a soft-sensing model corresponding to the LS support vector machine soft-sensing training sample database; optimizing normalized parameters and kernel parameters in the soft-sensing model by the CPSO and establishing a support vector machine soft-sensing model on the basis of the CPSO; and finally obtaining corresponding key biochemical parameter predicted values. In the method, a chaos sequence is adopted to initialize the positions of particle swarms, so that particles are more uniform and are distributed in a solution space in a non-repeated mode, the global optimization capacity of the soft-sensing model is improved, the predicting and measuring accuracy is high, the predicting and measuring speed is high, the predicting and measuring capacity is high, and the generalization ability is good.