Short-term Photovoltaic Power Forecasting Using the Combined KSVD and LSTM Method

A sparse representation model strategy based on feature extraction techniques is proposed for predicting solar power. To begin, all of these factors that affect photovoltaic power output are considered input data for the model, and dictionary learning techniques such as the K-mean singular value decomposition (KSVD) algorithm are applied to extract the corresponding sparse coding from all of the input data. Second, the sparse coding vectors are utilized as the model's input, and the Long Short-Term Memory (LSTM) algorithm is employed to construct the global prediction model. The proposed approach is used to a solar power forecasting to demonstrate the usefulness of the combined KSVD and LSTM method. Experiments reveal that several combination sparse representation approaches generate superior prediction results than SVM and LSTM under the same settings, with the combined KSVD and LSTM method providing the best prediction outcomes and model accuracy.