摘要
针对传统基线负荷估计方法在面对多变的电力负荷消耗时准确度不足的挑战,提出一种结合k-shape聚类算法和基于时间卷积网络(TCN)、注意力机制和极端梯度提升(XGBoost)的估计方法。首先利用TCN-Attention模型进行时间序列预测,提取负荷的时间特征;然后采用k-shape聚类算法对用户负荷数据进行聚类,以找出在形状和趋势上相似度高的负荷序列,并采用XGBoost算法捕捉负荷的空间特征。最后,将两者的估计值进行误差倒数法组合得到最终的估计结果。利用实际算例进行验证,并将所提方法与其他方法进行对比,结果表明了所提方法的有效性。
To address the challenge of insufficient accuracy in traditional baseline load estimation methods under dynamic electricity consumption patterns,this study proposes an integrated approach combining the k-shape clustering algorithm with a hybrid model based on temporal convolutional network(TCN),attention mechanism,and extreme gradient boosting(XGBoost).Firstly,the TCN-Attention model is employed for time series forecasting to extract temporal features of load profiles.Subsequently,the k-shape clustering algorithm is applied to group user load data into clusters with high similarity in shape and trend,while the XGBoost algorithm is utilized to capture spatial characteristics of load patterns.Finally,the individual estimates from both components are integrated using an inverse error weighting method to generate the final load estimation.The proposed method is validated through real-world case studies and compared other methods.The results demonstrate the effectiveness of the proposed method.
作者
沈杰
邢海军
俞钱
施怡沁
Shen Jie;Xing Haijun;Yu Qian;Shi Yiqin(School of Electrical Engineering,Shanghai Electric Power University,Shanghai 200090,China)
出处
《太阳能学报》
北大核心
2025年第8期676-685,共10页
Acta Energiae Solaris Sinica
基金
国家自然科学基金(52477106)。
作者简介
通信作者:邢海军(1979-),男,博士、讲师,主要从事新型电力系统、综合能源系统优化方面的研究。xinghj@shiep.edu.cn。
