To enhance the precision of ultra-short-term wind power prediction,a model is proposed that combines SGMD-SE with an optimized TCN-BiLSTM/BiGRU framework.Firstly,highly correlated variables are selected using maximum information coefficient(MIC)as input features.Secondly,symplectic geometry modal decomposition(SGMD)is utilized to decompose the original signal into initial symplectic geometry components(SGC),effectively suppressing mode mixing and eliminating the need for decomposition parameters.Then,sample entropy(SE)is applied to reconstruct the initial components,which are classified into high and low complexity categories.Based on the different characteristics of these categories,separate TCN-BiLSTM and TCN-BiGRU models are built for prediction.To improve the predictive performance of BiLSTM and BiGRU,time convolutional networks(TCN)are utilized to extract features from the two component categories.Additionally,an improved optimization algorithm called IOOA,based on Tent chaotic mapping and Cauchy mutation,is proposed to optimize their key parameters.Finally,the final prediction result is superimposed by combining the predicted values of each component.The findings suggest that the proposed hybrid prediction model significantly enhances the accuracy of ultra-short-term wind power forecasting and holds substantial practical utility.
关键词
风功率预测/分解/长短期记忆网络/时间卷积网络/鱼鹰优化算法
Key words
wind power prediction/decomposition/long short-term memory/time convolutional network/osprey optimization algorithm
维普
万方数据