Short-Term Power Prediction for Centralized Photovoltaic Plants Based on LSTNet-Attention

Short-term photovoltaic (PV) power forecast, as a fundamental foundation for the operation of new power systems, is critical for the safe and economical operation of power systems. In order to improve the short-term power prediction accuracy of PV power plants, this paper proposes a prediction model based on a long-term and short-term time series network-attention mechanism (LSTNet-Attn), which is formed by the fusion of multiple deep learning methods. The LSTNet-Attn model is established to forecast the short-term power generation of centralized PV power plants. This model first uses a convolutional neural network to extract local dependencies between power data, then captures the long-term trends of the power time series using a long-term and short-term memory (LSTM) network, fully learns the ultra-long-term repetitive patterns of the power series using a recurrent-skip structure, and then a fully connected layer and an autoregressive (AR) model are linked together for combined prediction. Finally, the effectiveness of the proposed method is verified using actual measurement data from a central PV plant. In comparison with four PV power prediction methods, the experimental results show that the proposed prediction model outperforms other methods and has higher prediction accuracy and robustness.