基于MT-BSGP的电力负荷预测方法  

Power Load Forecasting Method Based on MT-BSGP

作　　者：李智勇[1] 苏寅生[1] 李斌[1] 刘春晓[1] 李豹[1] 谭守标 LI Zhi-yong;SU Yin-sheng;LI Bin;LIU Chun-xiao;LI Bao;TAN Shou-biao(Power Dispatching Control Center,China Southern Power Grid,Guangzhou 510623,China;Key Laboratory of Intelligent Computing and Signal Processing,Anhui University,Hefei 230601,China)

机构地区：[1]中国南方电网电力调度控制中心,广州510623 [2]安徽大学计算智能与信号处理教育部重点实验室,合肥230601

出　　处：《哈尔滨理工大学学报》2021年第4期46-55,共10页Journal of Harbin University of Science and Technology

基　　金：国家自然科学基金青年科学基金(71601001).

摘　　要：为了预测家庭短期电力负荷,提出一种基于多任务贝叶斯时空高斯过程的电力负荷预测方法,有效地捕捉不同居住社区之间的关联性和家庭之间的局部空间变化。为了更好地捕捉不同居住社区之间关联性,提出了一种新的多任务学习方法,即低秩Dirty模型,通过使用“公共特征集”和“共享低秩结构”来改进多任务贝叶斯时空高斯过程的学习。为了克服低阶dirty模型和高斯过程参数联合估计的困难,提出了一种迭代算法。实验结果表明,与已有的几种方法相比,该方法可以有效地预测电力负荷。In order to forecast short-term household power load,a power load forecasting method based on multi-task Bayesian spatiotemporal Gaussian process(MT-BSGP)is proposed.The proposed method can effectively capture the relationship between different residential communities and the local spatial changes between families.In order to better capture the relevance between different residential communities,a new multi-task learning(MTL)method,namely low rank dirty model(LRDM),is proposed.The proposed model improves the learning of multi-task Bayesian spatiotemporal Gaussian process by using“common set of features”and“shared low-rank structure”.In order to overcome the difficulty of joint estimation of LRDM and Gaussian process(GP)parameters,an iterative algorithm is proposed.The experimental results show that the proposed method can effectively forecast the power load compared with the existing methods.

关 键 词：负荷预测 多任务学习 高斯过程 关联性 低秩结构 

分 类 号：TP391.3[自动化与计算机技术—计算机应用技术]