Abstract: Focusing on the defect of raw data segmentation and random sampling for the existing Markov chains model of PV output, a simulation method of PV output time series which is based on a new type of scenario division and considering temporal correlation is proposed. Firstly, the DBI clustering effectiveness index is introduced to optimize fuzzy C-mean clustering method, and the scenes are divided into different situations and the data sets of historical PV output series with more obvious data characteristics are established. Then, a number of state transition matrixes in different scenes are generated and PV output series are simulated through Markov chains Monte Carlo method. During this process, the state-value for the next moment can be got using the Copula theory to conduct the conditional probability sampling, so as to reduce the randomness of the traditional Monte Carlo sampling. According to actual case calculation, in the new method in this article, the PV output series are not only more accurate than the existing model in probabilistic statistical features of the data, but also preserve better autocorrelation of the original sequence.

Key words: PV output time series, Markov chains Monte Carlo method, scenario division, temporal correlation, Copula method