Research On Short-term Wind Power Forecasting Method Based On Improved CNN-LST

With the development of the country,environmental protection issues will gradually attract attention,especially in the energy industry,the environment caused by traditional energy is becoming increasingly prominent,the urgency of strategic transformation is increasing,hydropower,wind power,photovoltaic and other clean energy industry will gradually replace the position of fossil energy,so the ensuing power prediction accuracy will also be related to the safety of power supply,And electricity supply and demand balance.In order to make the predicted value of wind power as close as possible to the actual situation,this paper firstly deduces the formula of wind energy according to the kinetic energy formula,and then obtains several factors affecting wind power:wind direction,wind speed and air density.Secondly,Pearson correlation coefficient method is used to make a quantitative analysis of air density,pressure,temperature,wind speed,wind direction and other parameters that affect wind power.Through the physical data of a wind farm in Yunnan Province,the main parameters that affect the output power of wind turbines are verified,including wind speed,wind direction,temperature,humidity,air density,etc.Finally,combined with the actual situation,the wind direction and wind speed are determined as the main characteristic factors that affect the wind power output.Then,in the process of data preprocessing,two data processing methods are used,one is the removal of abnormal data by the Laida criterion,the other is based on CEEMDAN empirical mode decomposition method,through the comparison of the two methods,the advantages of the method based on CEEMDAN empirical mode decomposition method are obtained,and then CNN is used to obtain the features of the input date sequence.Through feature training of the pre-processed data,the original data has the data samples of training set,test set and verification set.On this basis,an improved CNN-LSTM model is designed to enhance the feature extraction function and prediction accuracy of input sequence in the model by increasing or decreasing the LSTM network gate.Finally,through the data validation of a wind farm in Yunnan province,through the comparison of the LSTM model,the improved CNN-LSTM model and the actual value comparison,through RSEM,MAE,R~2 three indexes,the superiority of the new combination form module proposed in this paper is proved.