Research On Application Of Modal Decomposition Algorithm In Short-term Prediction Of Photovoltaic Power Generation

The traditional power grid emphasizes the real-time balance of the source-gridload link through the forecast of load demand;the connection of a large number of photovoltaic power generation will bring about random intermittency,reverse peak shaving and uncontrollable source and load.Therefore,it is important to research how to achieve the precise prediction of short-term photovoltaic power generation,so as to continuously increase the scale of new energy in the power grid.At present,short-term photovoltaic power generation mainly uses the historical data of photovoltaic power generation in the form of single mode,and directly uses time series method,neural network and other methods to predict.Although it has achieved good performance in the overall trend prediction,the non-linear and non-stationary multi-modal characteristics of photovoltaic power generation are ignored.This paper introduces the modal decomposition method into the photovoltaic data processing process,decomposes the complex photovoltaic output signal into sub-signals with different time-frequency characteristics,and then combines them with the prediction model in a targeted manner.Not only it overcomes the complexity of the prediction model,but also enhances the adaptability of the prediction model to the signal,thereby improving the prediction performance of short-term photovoltaic power generation.The main research work of this paper includes:(1)A short-term power prediction method for photovoltaic power generation systems based on the combination of improved Empirical Mode Decomposition(EMD)and ARIMA is proposed.On the basis of the traditional EMD-based method,a white noise test link is added to eliminate the components without important physical information,which solves the problem of modal mixing caused by noise,and is also used to test whether the model fully extracts useful information from the sequence in the subsequent modeling process.At the same time,predictions are made for multiple groups of photovoltaic power generation under three weather conditions.Compared with traditional prediction methods,the effectiveness of the proposed prediction method is verified.(2)Propose an Ensemble Empirical Mode Decomposition(EEMD)and inverse function distance weighted KNN photovoltaic power generation power combined prediction method.Using the EEMD method to improve the modal mixing problem existing in the traditional EMD method,and in view of the shortcomings of the average distribution of feature weights in the traditional K-Nearest Neighbor(KNN)regression algorithm,the idea of inverse function distance weighting is introduced to improve it.Finally,the prediction results of the combination method are verified by an empirical system.(3)The prediction model combined by Variational Mode Decomposition(VMD)and the inverse function distance weighted KNN is used to predict the short-term photovoltaic power generation under three weather conditions.Compared with the EEMD combined prediction model,the proposed model effectively improves the prediction accuracy.