Research On Short-term Wind Power Prediction Of Wind Farm Based On LSTM And NARX

Wind energy is the world's third largest renewable energy and has huge development potential.China has become the country and region with the largest installed capacity in the world.However,wind energy is random and volatile and it will seriously affects the security of the local power grid,especially when large-scale wind power is connected to the grid.At present,the most effective method to solve the safety problem of large-scale wind power grid integration is wind power forecasting,especially super short-term and short-term wind power forecasting.Therefore,whether for wind power producers or wind power transmission parties,the research on wind power forecasting methods has important practical needs,and higher-precision forecasting methods will be the long-term goal pursued by the wind power industry and researchers.The wind turbine is mainly composed of an impeller,a nacelle and a tower,among which the impeller is composed of blades and a hub.The main function of the blade is to absorb wind energy,convert the wind energy into mechanical energy,and then use gears to drive the generator,and then convert the mechanical energy into electrical energy.Wind power power is mainly affected by factors such as wind speed and wind direction,especially wind speed has a huge impact on wind power.Wind speed is affected by meteorological factors such as atmospheric temperature,air pressure,and atmospheric density.Therefore,improving the prediction accuracy of wind speed and direction for the location of the wind turbine as much as possible is the basis for establishing a high-precision wind power prediction model.Long Short-Term Memory(LSTM)is a practical cyclic neural network model with powerful time series processing capabilities.Nonlinear AutoRegressive with eXternal input(NARX)neural network is a dynamic neural network model.This model makes full use of the good non-linear mapping ability of the MLP(Multi-Layer Perceptron)network,and at the same time combines the derived regression time series model.This thesis introduces the LSTM neural network and NARX neural network into the wind power prediction model,and proposes a two-stage wind power prediction model(LSTM-NARX model)that predicts wind speed,wind direction and wind power r separately.Based on the historical wind speed and wind direction forecast data provided by weather service companies and the corresponding actual wind speed and wind direction data of wind farms,the LSTM model is used to establish wind speed and wind direction prediction models to improve the forecast accuracy of wind speed and wind direction data provided by weather service companies;Use high-precision wind speed and wind direction prediction data and the corresponding wind power data of the wind farm establish a wind power prediction model based on the NARX neural network to improve the overall prediction accuracy of the wind power model.Since LSTM has the limitation of processing fixed-length sequence data,it is more suitable to use the Seq2Seq model to process variable-length sequence data.When the model maps an input sequence to another output sequence,it mainly completes two basic steps:encoding input and decoding output.With reference to the aforementioned LSTM-NARX two-stage wind power model,a wind power prediction model based on Seq2Seq and NARX models(Seq2Seq-NARX model)is proposed.The Seq2Seq model is used to replace the LSTM model in the aforementioned one to achieve higher-precision prediction of the weather information provided by the weather service company for a period of time in the future.Based on the US WIND Toolkit 2012 data set,simulation experiments are carried out on the proposed wind power prediction model.The wind farm No(SID)is 120914,and the data collection year is the full-year data of 2012.After data resampling,normalization,discretization and other preprocessing,the LSTM-NARX wind power prediction model and the Seq2Seq-NARX wind power prediction model were simulated.The results of the simulation experiment show that the prediction error of the LSTM-NARX wind power prediction model proposed in this thesis is about 13%lower than that of the direct prediction method,the Seq2Seq-NARX model has a higher prediction accuracy than LSTM-NARX wind power prediction model when performing ultra-short-term wind power prediction.