Aiming at problem that it is difficult to predict caused by random and unstability of photovoltaic（ PV）power

a least square support vector machines（ LS-SVM） electricity generation power prediction model based on similar day and evolution particle swarm optimization（ EPSO） is proposed. Considering large difference of meteorology factor and electricity generation power in different weather types

a weighted gray relational projection method is proposed to select similar day. Based on generalization ability and small sample problem processing ability of LS-SVM algorithm

set up prediction model use penalty parameter and Kernel function parameter of EPSO model to improve the precision of prediction. A model is established and evaluated according to real data of PV power plant abroad. The results show that this model has higher prediction precision and better applicability.