Short-term photovoltaic power prediction with similar-day integrated by BP-AdaBoost based on the Grey-Markov model

A photovoltaic (PV) power prediction based on similar-day, Grey-Markov model and AdaBoost is proposed for the impact of similar days on solar output power and the limit of historical data. First, select similar days for target days using dual 2D calculation. Four groups of similar-day were selected through Euclidean distance (ED) and grey relational analysis (GRA), with irradiance and temperature as similar variables. Second, a grey model (GM) was used to predict PV output power based on four sets of similar-day data; the prediction errors of the GM (1,1) were corrected by Markov chains (MC) to obtain four groups of prediction results. Finally, the four sets of predictions are integrated with BP-AdaBoost to obtain better prediction results. The proposed method is vali-dated with the actual data of PV plants. The simulation results show that the integrated prediction results outperform the prediction results of the Grey-Markov model based on similar-day, which achieves comple-mentary information and can provide important reference information for grid-connected PV power plants. Finally, comparing the proposed method (SD-BPA-GM) with decision trees (DT), back propagation neural net-works (BP), fully-connected neural networks (FNN), long-term memory networks (LSTM), and convolutional neural networks (CNN), respectively, SD-BPA-GM shows better accuracy.