Building an efficient PV output forecasting model can reduce the impact of PV output fluctuation on the power system. Considering the randomness and instability of photovoltaic power generation, a weighted Markov chain is proposed to modify the SVM prediction model to improve the prediction accuracy. Firstly, the SVM photovoltaic output prediction model is established to predict the output curve of the next day. Then, the relative errors are classified based on mean-mean square deviation method, and the weighted Markov chain model is used to predict the future state of the relative errors by weighting the autocorrelation of each order normalized error sequence. The threshold of state space is used to correct the prediction result of SVM. The model is applied to the output prediction of a photovoltaic power system. The simulation results show that the model modified by weighted Markov chain has higher prediction accuracy, and the model is feasible and effective. This work is supported by National Natural Science Foundation of China (No. 11601125).