Research Of Ultra-short-term Wind Power Prediction Method Based On Quadratic Decomposition And Machine Learning

With the rapid development of the wind power market,wind power installed capacity and wind power generation continue to increase,and the development of wind power energy has become an inevitable trend.Taking into account the random volatility of wind power,accurate prediction of the future wind power value of wind farms can enhance the safety,controllability and economy of the power system.This paper studies a quadratic decomposition strategy for processing wind power sequences to reduce the volatility of wind power sequences.The least squares support vector regression is used to establish the ultra-short-term prediction model of wind power,and the long short term memory network is used to establish the ultra-short-term prediction model of wind power.The main work is as follows:An ultra-short-term wind power least square support vector regression prediction method based on complementary ensemble empirical mode decomposition is studied.Aiming at the random volatility of the wind power sequence,the complementary ensemble empirical mode decomposition method is used to decompose the wind power sequence to obtain a series of sub-sequences with different feature scales,which are respectively used as the training input of the least squares support vector regression model.The least squares support vector regression prediction model was established,and the satin bowerbird optimization algorithm was used to optimize the regularization parameters and kernel function width of the least squares support vector regression model.The simulation results show that the proposed prediction method not only effectively reduces the complexity of prediction,but also ensures that the original wind power sequence has small reconstruction errors after being processed by complementary ensemble empirical mode decomposition,and has high accuracy in ultrashort-term wind power prediction.An ultra-short-term wind power least square support vector regression prediction method based on quadratic decomposition strategy is studied.Use the ensemble empirical mode decomposition method to decompose the wind power sequence to obtain a series of intrinsic mode function sequence and a residual component.Then use the wavelet packet decomposition method to decompose the highest frequency sub-sequence in the intrinsic modal function to obtain multiple sub-components with easing fluctuations,and through the t-mean test method,the wind power sub-sequence except the highest frequency sub-sequence generated by the ensemble empirical mode decomposition is dynamically divided and reconstructed.An intermediate frequency component and a low frequency component are obtained,and a quadratic decomposition strategy for processing the original wind power data is constructed.Established a least square support vector regression ultra-short-term wind power prediction model.In order to improve the diversity of the satin bowerbird optimization algorithm in the iterative process and the convergence of the algorithm,an improved satin bowerbird optimization algorithm is proposed.It is used to optimize the regularization parameters and kernel function width of least square support vector regression.The simulation results show that the proposed least square support vector regression ultra-short-term wind power forecasting method based on quadratic decomposition strategy has better forecasting performance than other forecasting models.An ultra-short-term wind power prediction method based on long and short term memory network that takes into account meteorological factors is studied.The distance correlation coefficient methodis used to screen out of wind speed and wind direction sine factors that have a strong correlation with wind power,and use them together with wind power as the input variables of the prediction model.Using the quadratic decomposition strategy,the original wind power and wind speed series are decomposed,divided and reconstructed,and multiple sub-components,intermediate frequency components and low frequency components with gentle fluctuations are obtained respectively.The obtained component quantities are trained under the long and short term memory network,and the prediction model of each component quantity is established.Finally,the prediction results of each component are superimposed to obtain the ultra-short-term wind power prediction value.The deep learning framework Tensor Flow and the programming language Python are used for simulation experiments.The simulation results show that the ultra-short-term wind power prediction method based on long and short term memory network with meteorological factors proposed in this chapter can predict wind power in a rolling manner.