Research On Ultrashort-Term Prediction Method Of Photovoltaic Power Based On EEMD And LSTM

In recent years,with the year-on-year increase of installed PV capacity,the proportion of PV power generation in the power system is increasing,and accurate and effective PV prediction results are of great significance for the safe and stable operation of the power system.However,the PV power output is affected by many factors,which leads to its strong volatility and randomness,which makes the PV power prediction difficult.The traditional PV power prediction methods are not very adaptable,and it is difficult to meet the current requirements of the power system for the accuracy and time scale of PV power prediction,therefore,it is important to study the accurate PV power ultra-short-term prediction methods.With the continuous research in the field of deep learning,more and more prediction methods use neural networks to build prediction models.Long and short-term memory neural networks have unique memory mechanisms and good learning ability,and they perform well in dealing with prediction problems with temporal correlation.In addition,PV power contains a variety of fluctuation patterns in different time scales,and extracting the characteristic patterns in the data in advance will make the training of the prediction model more satisfactory and improve the prediction accuracy.Therefore,this paper proposes two different prediction models for PV power prediction by combining two basic theoretical approaches: ensemble empirical mode decomposition(EEMD)and long short-term memory(LSTM).To address the problem of PV power prediction in the absence of meteorological data,an EEMD-PSO-LSTM-based ultra-short-term deterministic PV power prediction method is proposed,which firstly decomposes the PV power historical time series into EEMD data,extracts the component signals representing different time scale change patterns,and then establishes particle The results show that the proposed model can achieve better prediction results under different power fluctuation characteristics,and the average prediction accuracy of the three regions reaches The average prediction accuracy for the three regions is more than 95%,which is significantly better than other single models and has stronger applicability and accuracy,and the introduction of EEMD and PSO helps to improve the prediction accuracy.PV power is greatly influenced by surrounding meteorological factors,and it is difficult to reflect the influence brought by weather changes by using only the historical time series of PV power as input features,so the meteorological features around the power plant are used in the input of the prediction model and filtered by Pearson correlation coefficients,and the variables strongly correlated with PV output power are considered with each component of EEMD decomposition as input features,and then combined with the quantile regression theory,a probabilistic interval prediction model is established,and a probabilistic PV power prediction method based on EEMD-QRLSTM is proposed.The method is able to obtain interval prediction results under each quantile,and the proposed prediction method has high accuracy by comparing the cases,and the reliability and comprehensive performance of the proposed prediction method are improved compared with other models by combining the evaluation indexes of probabilistic prediction.