Research On Wind Power Prediction Based On LSTM

With the development of global economy,energy consumption is increasing,while the reserves of non-renewable energy such as oil,coal and natural gas are limited.At the same time,with the large-scale exploitation and consumption of fossil fuels,serious environmental problems are brought.Governments are paying more and more attention to the development of new energy sources.Wind power is developing rapidly under the policy encouragement,but the randomness of wind energy itself poses a threat to grid security.Electric energy is a kind of energy that can not be stored on a large scale.The power grid must maintain a general balance of absorption.The supply and demand of the power grid itself are balanced,but the fluctuation of wind energy is random.Large-scale access to the power grid poses a threat to the safe operation of the power grid.Wind power must be predicted for power dispatch.Accurate prediction of wind power can effectively mitigate the impact of w ind power fluctuation on the safe operation of power system.With the advancement of wind power parity,wind power enterprises have higher requirements.In order to avoid the waste of wind power resources,it is necessary to provide high quality and predictable power in order to improve the competitiveness of wind energy.This thesis presents a rolling prediction model based on LSTM network.The original data of wind power is preprocessed,the data missing is compensated reasonably and scientifically deleted to ensure the validity of the data.LSTM network is established and wind power is predicted by rolling prediction method.At the same time,in order to further improve the quality of wind power prediction,a prediction model of wind power generation base d on Bi-LSTM network is proposed.The main contents of this thesis are as follows:1.This thesis builds LSTM network based on TensorFlow deep learning framework.The pretreated data is intercepted,the training data is intercepted at the back end,the tes t data is intercepted at the front end,and the manufacturing time is different.Simulate the time sequence of real prediction scene data prediction.Then the model is trained,and finally the model is used to predict and test.2.According to the actual demand of wind power forecasting,combined with rolling forecasting method,LSTM forecasting model based on rolling method is constructed.The change of wind turbine power generation mainly depends on the change of weather factors,and the weather laws of different wind power plants are different,and the same wind farm has different weather laws in different seasons.Therefore,it is difficult to use a fixed parameter model to predict the power of arbitrary wind farms.The basic idea of rolling prediction i s to optimize the network parameters and obtain the latest wind power variation rules by constantly using the acquired new data,which is conducive to improving the accuracy of wind power prediction and improving the adaptability of the model to different wind power plants.3.To further improve the accuracy of wind power prediction,the inherent characteristics of historical data of wind power are studied.The prediction of wind power at the current time can not only use the changing law of historical wind power data,but also consider the changing law of future wind power data.Therefore,the model can predict the current wind power from two time directions.Bi-LSTM can predict the current data from two reverse time directions.A wind power prediction model based on Bi-LSTM is proposed.The experimental results show that the Bi-LSTM model further improves the accuracy of wind power prediction.