PV Power Probability Density Prediction Based on CEEMDAN-QRF

The photovoltaic output power is stochastic and variable, which makes it challenging to develop an accurate prediction system. A combined PV power probability density prediction model based on Complete EEMD with Adaptive Noise (CEEMDAN) and Quantile Regression Forest (QRF) was developed to improve the accuracy of PV power prediction due to the small base of PV power time series, high volatility and stochasticity, and the difficulty of obtaining high prediction accuracy. To begin, the CEEMDAN signal processing technique is utilized to decompose the original time series data of Photovoltaic power and reconstruct the modal function based on each modal function’s frequency magnitude. Second, for each reconstructed component, a QRF PV power prediction model is created, and the conditional distribution of predicted values is derived by superimposing the prediction results of distinct components. Finally, the probability density prediction results of PV power at any moment are output using kernel density estimation. The probability density prediction approach, as opposed to the deterministic point prediction method, can explain the likely fluctuation range and uncertainty of PV power in the future, and the model’s correctness is proven by the case test.