Accurate photovoltaic power prediction is an important means to ensure the safe and stable operation of new energy power system. According to the output characteristics of photovoltaic system, a combined prediction method based on integrated clustering and improved Markov chain model is proposed. Firstly, the trend sequences representing periodicity and random sequences representing randomness are divided by integrated clustering. Secondly, the trend sequence is predicted by the Markov chain model improved by first-order difference processing, and the random sequence is predicted by the differential integrated moving average autoregressive model. Finally, through modeling the historical operation data of photovoltaic power stations in Jilin Province and Qinghai Province, the results show that the accuracy rate of the method based on integrated clustering and improved Markov chain model is 7.76% higher than that of the traditional Markov chain model, which verify the applicability of the model proposed in this paper under different output types and different geographical locations.