Research On Wind Power Prediction Based On Improved VMD And Deep Learning

With the transformation of the power system towards clean and low-carbon,wind energy,as one of the most promising clean and renewable energy sources,is increasingly gaining penetration in the power system.However,due to the randomness,intermittency,and volatility of wind power generation,large-scale wind power grid integration will have a huge impact on the safe and stable operation of the power system.Therefore,high-precision prediction of wind power is of great significance for optimizing power system scheduling and improving the economy and utilization of wind power.This article focuses on the research of ultra short term wind power prediction.In response to the difficulties caused by non-stationary fluctuations in wind power and the insufficient research in the field of multi-step power prediction,a wind power prediction model is constructed by combining data decomposition technology and deep learning methods to improve the accuracy of single step and multi-step power prediction.The main research content of this article is as follows:Firstly,feature selection and data preprocessing of wind power data were carried out.The Maximal Information Coefficient(MIC)is used to evaluate the correlation between wind power and meteorological factors.According to the correlation analysis,the weak correlation features and redundant features are eliminated,and the wind speed and wind direction data at the hub height are retained as the input of the prediction model.Subsequently,abnormal and missing data in wind power data were identified and corrected to improve data quality and avoid difficulties in model training.Secondly,in order to reduce the impact of non-stationary fluctuations in wind power on prediction,Variational Mode Decomposition(VMD)is used to stabilize the wind power series and effectively extract local information from the power series.And in response to the problem of selecting preset parameters for VMD,the average envelope entropy is introduced as the evaluation index for decomposition,and the Squirrel Search Algorithm(SSA)is used to automatically optimize the parameters of VMD.Example analysis shows that SSA-VMD can effectively optimize the decomposition effect,and its decomposed subsequences have the highest accuracy when applied to prediction.Afterwards,a combined prediction model based on SSA-VMD and Convolutional Neural Network(CNN)-Bidirectional Gated Recurrent Unit(BiGRU)was proposed.The sub sequences obtained through SSA-VMD decomposition are combined with historical wind power data to form a feature map;Under the premise of not damaging the original temporal structure,pass 1×1 Convolutional and channel pooling extract and highlight the coupling information between different features,and use BiGRU to mine temporal information from the front and back directions,thereby outputting prediction results.Through multiple sets of instance analysis,it has been verified that the proposed model can fully mine the implicit information of input data and effectively improve the accuracy of prediction.Finally,in order to effectively predict the continuous power of future time periods and make the predicted information more valuable,a multi-step wind power prediction model based on Sequence to Sequence(Seq2Seq)is proposed.Use CNN-BiGRU as the encoder to encode historical wind power data into intermediate state vectors,and another GRU as the decoder to gradually decode the intermediate state vectors and output a prediction sequence.Through example analysis,it is shown that the Seq2Seq model has better multi-step prediction performance compared to other prediction methods,and the advantage becomes more obvious as the prediction step size increases.