重庆理工大学学报2025，Vol.39Issue(7)：217-226,10.DOI:10.3969/j.issn.1674-8425(z).2025.04.027
差分进化优化的风电功率混合预测模型
Wind power hybrid prediction model optimized by differential evolution
摘要
Abstract
Globally,wind energy resources are heterogeneously distributed,with significant potential observed in regions including China's eastern coastal areas,Inner Mongolia,and Gansu.Meteorological and topographical factors significantly influence the spatiotemporal variability of wind energy,resulting in substantial intermittency,fluctuations,and inherent uncertainty.These intrinsic characteristics contribute to the instability of wind power generation,thereby posing considerable challenges to the efficacy of power system scheduling and operational security.As a result,accurate wind power forecasting methodologies are crucial for ensuring reliable energy management. The dominant wind power prediction methodologies are broadly categorized into three principal frameworks:physical models,statistical models,and machine learning models.Physical models leverage numerical weather prediction and ancillary methodologies to procure requisite data for predictive modeling.However,they are inherently limited in their capacity to fully accommodate the inherent uncertainties within the meteorological forecast data.Conversely,statistical models,while adept at analyzing historical datasets,often exhibit diminished efficacy in the context of intricate,non-linear interdependencies and high-dimensional data structures.Traditional prediction models exhibit limited performance and face challenges in accurately representing the nonlinear and non-stationary characteristics of wind power time series.In recent years,deep learning models have rapidly developed and gained widespread application.Due to the high volatility,strong randomness,and complex nature of wind power data,the majority of current research employs signal decomposition algorithms to process non-stationary sequence data.However,further optimization of parameter configurations is essential to enhance the predictive accuracy and generalization capabilities of these models. To address the above challenges,this paper develops a hybrid forecasting model.It amalgamates Variational Mode Decomposition,Convolutional Recurrent Neural Networks and attention mechanism.First,the Pearson correlation coefficient evaluates the numerical weather prediction data and anemometer tower measurements for feature selection,thereby facilitating the prioritization of variables and demonstrating a strong correlation with wind power generation.Then,Variational Mode Decomposition decomposes the original time series into a suite of modal components,each characterized by distinct frequencies,consequently addressing the inherent complexity and non-stationarity of the sequence data.Meanwhile,a Differential Evolution algorithm optimizes parameters,specifically to ascertain the optimal number of modes.The resultant feature matrix,along with the decomposed modal components,inputs into a Convolutional Recurrent Neural Network architecture,which effectively extracts both the spatial features and the temporal dependencies inherent within the data.Finally,attention mechanism is integrated to augment the model's capacity to selectively capture salient information within the sequence,thereby enhancing the accuracy of wind power output predictions.The model rigorously assesses predictive efficacy within a Python-based simulation environment.It incorporates systematic ablation studies and comparative analyses to ascertain forecasting capabilities.Furthermore,the model evaluates generalization aptitude utilizing empirical data acquired across a spectrum of seasonal variabilities.Experimental results demonstrate the proposed model achieves higher forecasting accuracy and better generalization performance,effectively meeting the requirements of wind power forecasting.These findings confirm its advantages in handling wind power forecasting tasks.关键词
风电功率预测/变分模态分解/卷积循环神经网络/注意力机制/差分进化算法Key words
wind power forecasting/variational mode decomposition/convolutional recurrent neural network/attention mechanism/differential evolution algorithm分类
信息技术与安全科学引用本文复制引用
陈梦娇,陈为真,张岳..差分进化优化的风电功率混合预测模型[J].重庆理工大学学报,2025,39(7):217-226,10.基金项目
湖北省自然科学基金项目(2022CFB449) （2022CFB449）
湖北省教育厅科技项目(B2020061) （B2020061）
