|
Photovoltaic (PV) power generation is highly random and volatile, especially in microgrid power systems where this volatility has a great impact on their safe and stable operation. Meanwhile, considering the problems of traditional photovoltaics power prediction methods in terms of incomplete analysis of influencing factors and uncertainty of point prediction errors, this paper proposes a PV power interval prediction model by combing Ensemble Empirical Mode Decomposition (EEMD) and the Long Short-Term Memory (LSTM) model. Firstly, the PV historical data are preprocessed based on the k-Nearest Neighbors (k-NN) algorithm; XGBoost is used to analyze the PV power influencing factors and extract the PV power sample features; then, EEMD is used to decompose the historical power series into a series of Intrinsic Mode Functions (IMFs) and residual components, and the components having strong correlation with the original power series are filtered by the Maximal Information Coefficient (MIC) and fed into the LSTM model; the prediction results of each component are superposed to obtain the PV power prediction results; finally, the distribution interval of the predicted power in the known confidence interval is figured out with by fitting t-distribution based on the prediction results. For the model, the photovoltaic power generation data of an actual park in 2018 was taken as the sample to compare with conventional prediction models, such as Random Forest (RF), Multiple Linear Regression (MLR), and Support Vector Machine (SVM). The experimental results show that the model is better than conventional prediction models in Root Mean Square Error (RMSE), Mean Absolute Error (MAE), coefficient of determination (R2), and other indicators, and can give the range of PV predicted power with higher prediction accuracy and adaptability, which can meet the demand of short-term prediction of PV power in microgrids and provide strong data support for the stable operation of microgrid power systems.
|
