Research On Ultra-short-term Wind Power Prediction Based On Deep Neural Network

Under the backdrop of "double carbon",the new energy field has achieved unprecedented development,and the proportion of wind power grid connection has increased year by year,but the intermittent wind speed has also brought a great test to the wind power grid.Therefore,one of the methods to solve this problem is to rise the prediction accuracy.Based on the analysis and summary of wind energy forecasting research status at home and abroad,an indirect method to obtain wind power forecast is proposed.Through the comparative analysis of various models,a wind speed prediction model based on the combination of ELMAN neural network variational modal decomposition and improved sparrow algorithm(ISSA)optimization is established.The power curve indirectly obtains the pred icted value of single wind power.The specific work content is as follows:1.Interpretation of predictive models.The prediction principles of feed forward neural network,feedback neural network and least squares support vector machine are deduced,and the advantages and disadvantages of each model are analyzed.2.Determine the input of the wind speed prediction model.Using the mutual information method and the experimental method to establish the input number of the wind speed prediction model,the above four single wind speed prediction models are established.The prediction results show that the ELMAN neural network has obvious prediction advantages.3.Improve predictive models.Considering that the ELMAN neural network cannot obtain the optimal solution and the key to rising the prediction accuracy of the LSSVM model is to reasonably set the parameters,in this article,based on the introduction of the SSA to optimize ELMAN and LSSVM models,improved Tent mapping and reverse learning are used to improve the SSA algorithm,and 10 benchmark functions are used to check feasibility of the ISSA.Finally,the ISSA-ELMAN and ISSA-LSSVM optimization prediction models are established.Both prediction errors are reduced and the latter has the smallest prediction error in a single model.4.Introduce decomposition techniques in wind speed prediction models.Reduce wind speed prediction error and improve wind speed stability.This paper analyzes empirical mode decomposition(EMD),ensemble empirical mode decomposition(EEMD),empirical mode decomposition Adaptive Noise Set(CEEMDAN),variational The advantages and disadvantages of the mode decomposition algorithm(VMD)are analyzed and the actual wind speed data are decomposed by CEEMDAN and VMD,and the decomposition results are analyzed.ISSA-ELMAN is established.5.Indirect method to realize wind power prediction.Use the quartile method and four-point interpolation method to clean the abnormal wind power data,then use the ISSA-ELMAN model to fit the processed historical wind speed-power curve,and finally,use the wind speed forecast value and the curve to get the power forecast value...