Optimizing CNN-LSTM Model for Short-term PV Power Prediction using Northern Goshawk Optimization

PV power is unstable and time-varying. In order to improve the accuracy and stability of PV power prediction, this paper proposes a short-term PV power prediction method based on the Northern Goshawk optimization algorithm (NGO)-optimized CNN-LSTM model. Firstly, the data were correlated using the Pearson correlation coefficient, and features with low correlation were eliminated. Then, the CNN-LSTM model was established, with the convolutional neural network (CNN) responsible for extracting the important features in the data, and the long-short term memory (LSTM) trained on the data. To further optimize the model, the NGO algorithm was introduced to optimize the hyperparameters such as the number of convolutional kernels, the step size of the pooling layer, and the number of neurons in the hidden layer of LSTM in the CNN-LSTM model, aiming to improve the prediction performance of the model. The prediction results show that the CNN-LSTM model optimized by NGO achieves higher accuracy compared to CNN-LSTM, SSA-CNN-LSTM, and other models, thereby providing technical support for grid scheduling and energy management.