Power load prediction can provide reliable decisionmaking basis for power system planning and operation. With the development of smart grid, the amount of data collected by supervisory control and data acquisition increases largely, and the structure of data becomes more complex. Frequent changes of load and regional meteorological factors have influence on the accuracy of load forecasting. A shortterm load forecasting method is proposed in this study, which is based on elastic network (EN) for large data dimension reduction and flower pollination algorithm (FPA) for BP neural network optimization. By adding norms and norms to penalty items, the elastic network has the advantages of least absolute shrinkage and selection operator (LASSO) and ridge regression. It can solve the problem of dimension reduction effect, which is affected by collinearity and group effect in LASSO dimension reduction. Then, FPA is introduced to optimize BP neural network, in which the weights and thresholds are easily affected by initial values, slow convergence speed and easy to fall into local optimum. Compared with particle swarm optimization method, the optimization speed of flower pollination algorithm is faster and the effect is better. The proposed method has been applied for predicting power load. Experimental results show that the prediction accuracy can be effectively improved.