The correlations among the building energy consumption factors can corrupt the prediction model’s performance, and get undesirable results. A prediction model based on KPCA-WLSSVM is proposed to forecast building energy consumption. The kernel principal component analysis (KPCA) method could not only solve the linear correlation of the input and compress data but also simplify the model structure. A novel hybrid chaos particle swarm optimization simulated annealing (CPSO-SA) algorithm is applied to optimize WLSSVM parameters to improve learning performance and generalization ability of the model. Furthermore, the KPCA-WLSSVM model is applied to the energy consumption prediction for an office building, and the results show that the KPCA-WLSSVM has better accuracy compared with WLSSVM model, LSSVM model and RBF neural network model. and the KPCA-WLSSVM is effective for building energy consumption prediction.