| With the increasing population and rapid economic expansion,Chinese electricity production and consumption are on the rise.Therefore,reasonable planning and effective management of photovoltaic(PV)power generation is very important,and the premise of achieving these is to predict one hour ahead PV power generation accurately and quickly.However,the traditional PV power prediction model can not meet the requirements of the current power system.And here are the problems with PV power forecasting: First,the PV power generation series is non-stationary due to the influence of solar illumination changes,and the deep learning model is difficult to predict;Secondly,the variation of PV power generation is affected by many factors,and it is difficult to make reasonable use of meteorological factors as a feature.How to effectively stabilize the PV power generation data and reasonably analyze the impact of meteorological factors,and accurately predict the PV power generation by using improved deep learning method,has become the focus of many scholars in PV power related fields.This thesis systematically analyzes the background,and then builds a PV power generation prediction model based on signal decomposition and LSTM.The main research contents are as follows:In this thesis,the Clustering Empirical Mode Decomposion(CEMD)algorithm is proposed to decompose the PV power generation series.Combining this algorithm with Long Short-Term Memory(LSTM)network,a CEMD-LSTM PV power generation prediction algorithm is proposed.Firstly,the Intrinsic Mode Functions(IMFs)are obtained by the decomposition of the original PV power generation sequence using empirical mode decomposition algorithm.Then,the improved k-Shape clustering algorithm proposed in this thesis is used to cluster IMFs to ensure the constant number of sub-series.Finally,LSTM network is used to predict the sub-series and the output of each network is added to obtain the final prediction result.The simulation results show that CEMD-LSTM model can reduce the training time of the model while ensuring the prediction accuracy.In addition,the CEMD-LSTM PV power generation prediction model only uses historical PV power data for prediction,so there is still a gap for improvement in the prediction results under non-ideal weather conditions.Therefore,this thesis proposes an optimization algorithm for PV power generation prediction results based on feature attention mechanism.Firstly,the algorithm analyze the correlation between meteorological features and PV power generation series,and reconstructed the phase space of the historical PV power generation series.Then,the processed meteorological features,historical PV power generation sequence and CEMD-LSTM prediction results were splicing to form the combined data.Finally,the combined data are used as the input into Attn-LSTM model based on feature attention mechanism and LSTM.Through simulation experiments,it is proved that the prediction result optimization algorithm proposed in this paper can effectively improve the prediction accuracy of the model. |
