In order to improve the prediction precision, a novel electricity consumption prediction model is proposed based on combination optimization theory. Firstly, the learning samples is obtained by phase space reconstruction. Then the learning samples are input into least square support vector machine and train, which the parameters of model are optimized by artificial fish swarm algorithm, and electricity consumption prediction model is established. Finally, the performance of model is test by simulation experiment. The results show that the proposed model can describe electricity consumption change rule, and improve the prediction precision.