Research On Short-Term Wind Power Prediction Algorithm For Wind Farms In The Bohai Rim Region Based On Deep Learning

With the increasingly urgent problem of global energy demand,a lot of countries regard offshore wind energy development as one of the effective measures to alleviate energy pressure and optimize energy supply structure.However,due to the special geographical environment of the offshore wind farm,the randomness of wind energy is affected by various factors,which makes the wind power have the fluctuation characteristics and increases lots of issues for the power grid operation.Therefore,it is particularly important to how to predict the short-term prediction of wind power and formulate preventive measures for power fluctuation.Taking the Bohai Rim region as the research object,this paper selects a wind farm and a test fan in the region to study the short-term prediction algorithm of wind power.First of all,considering that the wind power in the coastal-land interface area is susceptible to the influence of environmental factors and fluctuates greatly,this paper uses the empirical wavelet transform algorithm to extract the characteristics of non-stationary power series for smoothing.Aiming at the defects of long training time and insufficient prediction accuracy of deep belief network model,the particle swarm algorithm is selected to optimize the number of hidden layer nodes of the network.Finally,a joint prediction model of deep belief network based on empirical wavelet transform and particle swarm optimization is proposed to realize short-term wind power prediction.Compared with other prediction models,the effectiveness of the proposed method is verified.Secondly,in view of the error problem caused by the fixed parameters in each prediction process of the traditional direct prediction method,this paper adopts the rolling prediction model and uses the new data obtained in each step to optimize the network parameters.Each rolling of each parameter can conform to the change rule of instantaneous wind power and improve the accuracy of power prediction.In order to reduce the operation time of traditional signal decomposition,the convolution neural network with the advantages of high efficiency data feature extraction and the long-short term memory neural network with the advantages of time memory function are selected.The wind power rolling prediction model of convolution fusion long-short term memory network is constructed,and the error analysis is compared with each single model.It is proved that the combination model can improve the accuracy and computational efficiency of prediction.Finally,based on the establishment of wind power prediction model,in order to facilitate the application of the model to practice,a wind power prediction software based on deep learning is developed by calling the function and interface design module in MATLAB R2019 b.The software integrates a variety of prediction methods,with reasonable design and simple operation.It can realize data acquisition,power prediction,error analysis and other functions,and has certain practicability.