To overcome the difficulty of choosing appropriate external structure parameters for support vector machine(SVM)models, the genetic algorithm(GA)is introduced and a GA-SVM optimal prediction model of underflow concentration is built. The change laws of thickener underflow concentration are discussed under different parameters, and the structure parameters of deep cone thickener are optimized. Sijiaying iron mine is taken as an example, and the results show that with the optimal underflow concentration of 72%, the optimized external structure parameters of deep cone thickener are 10 m high and 30 degree cone. The optimized deep cone thickener in Sijiaying runs steady with continuous underflow concentration flowing. Compared with other similar thickeners, its energy load and fault probability are reduced by 15% and 80% respectively.