| In recent years,in order to cope with the dual challenges of global energy shortage and environmental pollution,a series of renewable clean energy sources such as wind energy,solar energy,geothermal energy and so on have been increasing in the proportion of energy production.Among them,wind power generation has received extensive attention in the field of energy technology due to its high efficiency,safety and cleanliness.However,wind power generation is subject to meteorological conditions and is affected by geographical location factors.The output power of the wind power system will have great volatility,which makes it more difficult for the power grid dispatching center to make plans.It will also have a great impact on the safe and stable operation of the power grid.Therefore,in order to enhance the prediction level of the system for wind power,this thesis proposes a BiGRU-Att-1d CNN wind power prediction model based on IBA-VMD method for wind power data preprocessing.The main work is as follows :Firstly,aiming at the non-stationary characteristics of wind power,an improved bat algorithm(IBA)is proposed to optimize the penalty factor α and layer number K in variational mode decomposition(VMD),which effectively improves the decomposition effect of VMD.Secondly,aiming at the problem of gradient explosion and overfitting in the traditional neural network wind power prediction model.Bidirectional Gating Recurrent Unit(BiGRU)based on Attention Mechanism(AM)and Convolutional Neural Network(CNN)based on Time Sliding Window(TSW)are used to extract the features of wind power signal and wind speed signal.At the same time,a 1d CNN framework based on two-layer convolution is constructed,and the output features of the above two models are trained by using the 1d CNN framework to obtain a more accurate prediction value of wind power.Finally,this study uses the Python platform for example simulation analysis to test the BiGRU-Att-1d CNN wind power prediction model,and introduces four traditional models as a control group for comparison.The MSE,R2,MRE,RMSE errors were counted and residual analysis was performed.Experiments show that the wind power prediction model established in this thesis has high accuracy.Compared with the traditional wind power prediction method,the model effectively avoids the occurrence of gradient explosion and over-fitting,and the prediction accuracy and operation stability of the model are greatly improved. |
