Contemporary power network planning faces critical challenges from intensifying climate variability,including greenhouse effect amplification,extreme precipitation anomalies,and persistent thermal extremes.These meteo...Contemporary power network planning faces critical challenges from intensifying climate variability,including greenhouse effect amplification,extreme precipitation anomalies,and persistent thermal extremes.These meteorological disruptions compromise the reliability of renewable energy generation forecasts,particularly in photovoltaic(PV)systems.However,current predictive methodologies exhibit notable deficiencies in extreme weather monitoring,systematic transient phenomena analysis,and preemptive operational strategies,especially for cold-wave weather.In order to address these limitations,we propose a dual-phase data enhancement protocol that takes advantage of Time-series Generative Adversarial Networks(TimeGAN)for temporal pattern expansion and the K-medoids clustering algorithm for synthetic data quality verification.In order to better extract the spatiotemporal features of the model input simultaneously,we develop a hybrid neural architecture integrating Convolutional Neural Networks with Long Short-Term Memory modules(CNN-LSTM).To avoid the problem of hyperparameters getting trapped in local optimal solutions,we use the Whale Optimization Algorithm(WOA)algorithm to obtain the global optimal solution by simulating the hunting of humpback whales,further enhancing the generalization ability of the model.Experimental validation demonstrates performance improvements,with the proposed model achieving 30%higher prediction accuracy compared to Genetic Algorithm-Backpropagation Neural Network(GA-BPNN)and Radial Basis Function-Support Vector Regression(RBF-SVR)benchmarks,promoting the renewable energy prediction in data-constrained extreme weather scenarios for future power networks.展开更多
风电出力与运行条件密切相关,寒潮作为一种极端天气事件,往往会导致电力负荷激增。该场景下的风电功率预测结果与实际出力存在较大偏差,从而对电力系统的安全稳定运行带来了挑战。文章提出了一种基于现实时间序列生成对抗网络(real-worl...风电出力与运行条件密切相关,寒潮作为一种极端天气事件,往往会导致电力负荷激增。该场景下的风电功率预测结果与实际出力存在较大偏差,从而对电力系统的安全稳定运行带来了挑战。文章提出了一种基于现实时间序列生成对抗网络(real-world time series generative adversarial network,RTSGAN)样本扩充和注意力动态权重集成的寒潮极端天气条件风电功率预测方法。首先针对寒潮天气下风电样本稀缺的问题,采用RTSGAN对寒潮事件原始数据进行样本扩充;然后基于扩充得到的样本数据集分别构建基础预测模型和注意力网络;最后通过注意力动态权重集成方法进行风电功率预测。算例结果表明,所提方法能够解决寒潮极端天气风电样本稀缺的问题,可有效提升寒潮极端天气条件下的风电功率预测精度。展开更多
基金supported by Science and Technology Projects of Jiangsu Province(No.BE2022003)Science and Technology Projects of Jiangsu Province(No.BE2022003-5).
文摘Contemporary power network planning faces critical challenges from intensifying climate variability,including greenhouse effect amplification,extreme precipitation anomalies,and persistent thermal extremes.These meteorological disruptions compromise the reliability of renewable energy generation forecasts,particularly in photovoltaic(PV)systems.However,current predictive methodologies exhibit notable deficiencies in extreme weather monitoring,systematic transient phenomena analysis,and preemptive operational strategies,especially for cold-wave weather.In order to address these limitations,we propose a dual-phase data enhancement protocol that takes advantage of Time-series Generative Adversarial Networks(TimeGAN)for temporal pattern expansion and the K-medoids clustering algorithm for synthetic data quality verification.In order to better extract the spatiotemporal features of the model input simultaneously,we develop a hybrid neural architecture integrating Convolutional Neural Networks with Long Short-Term Memory modules(CNN-LSTM).To avoid the problem of hyperparameters getting trapped in local optimal solutions,we use the Whale Optimization Algorithm(WOA)algorithm to obtain the global optimal solution by simulating the hunting of humpback whales,further enhancing the generalization ability of the model.Experimental validation demonstrates performance improvements,with the proposed model achieving 30%higher prediction accuracy compared to Genetic Algorithm-Backpropagation Neural Network(GA-BPNN)and Radial Basis Function-Support Vector Regression(RBF-SVR)benchmarks,promoting the renewable energy prediction in data-constrained extreme weather scenarios for future power networks.
文摘风电出力与运行条件密切相关,寒潮作为一种极端天气事件,往往会导致电力负荷激增。该场景下的风电功率预测结果与实际出力存在较大偏差,从而对电力系统的安全稳定运行带来了挑战。文章提出了一种基于现实时间序列生成对抗网络(real-world time series generative adversarial network,RTSGAN)样本扩充和注意力动态权重集成的寒潮极端天气条件风电功率预测方法。首先针对寒潮天气下风电样本稀缺的问题,采用RTSGAN对寒潮事件原始数据进行样本扩充;然后基于扩充得到的样本数据集分别构建基础预测模型和注意力网络;最后通过注意力动态权重集成方法进行风电功率预测。算例结果表明,所提方法能够解决寒潮极端天气风电样本稀缺的问题,可有效提升寒潮极端天气条件下的风电功率预测精度。
