Power Generation Prediction Based On Empirical Mode Decomposition And Neural Networks

Photovoltaic power generation is characterized by volatility and intermittency.Accurate forecasting of photovoltaic power generation power can help coordinate and coordinate the use of conventional energy and photovoltaic power generation,which has great benefits in terms of energy use and the needs of the people.The important influence factor of photovoltaic power generation is solar irradiance,and the instability of solar irradiance will bring difficulties to the prediction of photovoltaic power generation.Therefore,how to accurately predict photovoltaic power generation will be the focus of research.In combination with the above background,this paper uses different photovoltaic power generation data sets to model and analyze the characteristics of the output power of photovoltaic power generation and its meteorological impact factors.At the same time,based on traditional statistical time series models,artificial intelligence algorithms,data processing methods,and other related theories,this paper proposes different combined forecasting models for power generation.First,this paper studies the generation forecasting based on the single variable model,which is the single variable time series model.Through the historical data of the output power of a photovoltaic power plant in Gansu Province,the EMD(Empirical Mode Decomposition)algorithm is used to decompose the generation data,reducing the complexity and noise of the original data,and obtaining five intrinsic mode functions and one trend function,Then,four of the intrinsic mode functions are normalized,and the processed data are fitted with different Long short-term memory neural networks LSTM(Long Short Term Memory).The third intrinsic mode function is analyzed according to its own characteristics and selected to be fitted with the moving average autoregressive model ARMA(Auto Regression and Moving Average)to obtain the prediction results of each component,Then,the results of component prediction are reconstructed to get the final prediction value.The proposed model is the EMD-LSTM-ARMA combined prediction model.Then,the combined prediction model is compared with the single model through model evaluation indicators.The results show that the proposed combined prediction model has better prediction accuracy.Secondly,based on the theoretical basis of the EMDLSTM-ARMA combined model,the EMD-LSTM combined prediction model is proposed,And it is applied to datasets with larger sample sizes.According to the evaluation indicators of the model,it can be seen that the proposed combination model still has high prediction accuracy,and it also indicates that the EMD-LSTM combination model has good generalization performanceSecondly,this paper proposes a multi variable model based power generation forecasting,which is a multi-variable time series model.Using the data from the China New Energy Competition,explore the mutual information of seven meteorological factors that affect power generation,establish a deep neural network DNN(Deep Neural Networks)to perform regression analysis on the data set,obtain the power generation output data predicted from the relationship between the seven meteorological factors,and then calculate the residual of the prediction data.Carry out stationarity and randomness tests on the residual,and the results show that the residual sequence is not a purely random sequence,This indicates that there are still relationships within the residual sequence that have not been learned by the deep neural network,so the LSTM model is used to modify the residual sequence.Before the residual sequence is input into the LSTM network,the EMD method is still used to process the residual sequence,and then the decomposed residual components are grouped and input into different LSTM models for prediction to obtain the prediction data of each component.After linear superposition,the prediction correction value of the residual sequence is reconstructed,Finally,the predicted value of the residual series is added to the predicted value of the DNN model to obtain the final prediction result.After evaluating the model on the test set,it is found that the proposed combined prediction model has a good fitting effect.At the same time,compared with a single DNN model and a DNN-LSTM model,the results show that the DNN-EMD-LSTM combined prediction model has better prediction accuracy and generalization ability.At the end of this paper,the summary is made.Through the research of this paper,it is found that the proposed combined forecasting model has a higher prediction effect than the single model,and then combined with the noise reduction processing of the data,the prediction accuracy of the model will be further improved.Therefore,through the analysis of the experimental results,it can be concluded that the proposed combined forecasting model has a better prediction effect and fitting situation.