GAO Shengqiang, ZHANG Lin, WANG Haipeng, SONG Yu, YAN Hao, LIU Zining, ZHOU Weiwei, BU Shuaiyu
To significantly enhance the prediction accuracy of the output power of photovoltaic (PV) power station, this paper developed an intelligent prediction model for PV output through incorporating CCM, IGRA, PSO and BiLSTM into a general framework. Firstly, the convergent cross mapping (CCM) algorithm was employed to extract key meteorological elements affecting PV output, where they are considered as major evaluation criteria of similar day selection and critical input variables of subsequently established prediction model; secondly, an improved grey relational analysis method (IGRA) based on entropy weight method was utilized to select historical similar days that closely match meteorological characteristics of the day to be predicted. And then, selecting the critical weather parameters and PV power generation sequence of similar days as the training samples, the particle swarm optimization (PSO) algorithm was used to determine optimal hyperparameters combination for the bidirectional long short-term memory (Bi-LSTM) network. A high-precision PV output prediction model based on CCM-IGRA-PSO-BiLSTM for the predicted days was established. Three criteria, including mean absolute error (MAE), mean absolute percentage error (MAPE) and root mean square error(RMSE), were selected as the evaluation metrics for model performance. The obtained results indicate that, taking the sunny weather in spring as an example, the proposed combined model achieved MAPE, MAE and RMSE of 0.38%, 0.06 and 0.07 MW, respectively, all of which surpass those of several baseline models. This way provides scientific guidance and support for the station to formulate reasonable production plan and electricity market participation strategy.
