Short-term PV power prediction study based on EEMD-KPCA-LSTM

Short-term PV power prediction is important for PV power generation stability and power dispatch. In order to cope with the noise disturbances occurring in PV power generation, this paper proposes a prediction model with an improved algorithm. Firstly, the annual PV data are decomposed by using EEMD to obtain the eigenmode function components; KPCA is used to reduce the dimensionality of the original sequences brought by EEMD, eliminate redundancy, and divide the test set and training set; The LSTM neural network is used to model the multivariate feature sequences in dynamic time to achieve the prediction of PV power. The model compares the prediction model of a single algorithm, and the accuracy of the prediction is improved.