Generation Method for Medium and Long-Term Photovoltaic Power Time Series Considering Variable Order Time Series Characteristics

For PV power time series generation in a single station, the traditional fixed order Markov chain model has the advantage of using limited training data for time series modeling. However, it has the problem that the fixed order and high state space complexity lead to the low restoration of the statistical characteristics of the generated time series. To solve the problem, the paper proposes a variable order Markov model (VOMM) for generating power time series. First, to reduce the computational complexity, a finite alphabet is obtained by traversing the training series, and Trie is used to represent the transition relationship between the power levels. Then, the Prediction by Partial Match (PPM) algorithm is used to compute the conditional transition probability between the nodes in Trie. The “escape” mechanism is introduced in PPM to characterize accidental events. Finally, the power time series is generated by the trained VOMM. The numerical example shows that compared with the current mainstream statistical learning methods, VOMM can improve the statistical measurements, such as the mean, variance, and quartile of the power time series, and are suitable for the medium and long-term power time series of PV simulation and generation.