Short-term Prediction of Solar Irradiance Based on EEMD-Lasso-MLP

As solar irradiance is affected by weather conditions and other factors, the uncertainty of it leads to strong randomness and intermittency of photovoltaic power generation, which further leads to the instability of output power of photovoltaic power stations, and brings severe challenges to the dispatching and safe operation of the power system. Therefore, high-precision prediction of solar irradiance is a necessary prerequisite to ensure the safety, stability and reliability of power grid operation. In this paper, a short-term solar irradiance prediction method based on Ensemble Empirical Mode Decomposition (EEMD), Least Absolute Shrinkage and Selection Operator (Lasso) and Multi-layer Perceptron (MLP) is proposed, so as to extract the optimal feature set and improve the prediction accuracy of solar irradiance effectively. Firstly, EEMD is applied to decompose the original input variables into a series of decomposition sequences, in order to reflect the changes and inherent features of the original signal on different scales; Then, Lasso is used to extract the features of every decomposition sequence, eliminate the redundancy between them, reduce the dimension of the input, and extract the optimal feature set that can best express the solar irradiance; Finally, taking the extracted optimal feature set as the input, MLP is applied to predict the solar irradiance. The prediction and comparative results show that the application of EEMD and Lasso improves the prediction accuracy by at least 74% and 16%, and improves the prediction accuracy effectively.