摘要
电量需求可以映射经济发展趋势,上下游行业发展导致不同行业用电量需求间存在因果关系。现有行业用电量预测缺乏行业间因果传导路径分析,且预测算法忽略了时序与特征之间的耦合关系,导致行业用电量预测模型存在输入因果传导路径不明确、政策参考性差、预测精度低等问题。为此,文中提出一种考虑行业发展因果传导路径的行业用电量预测方法。首先,为挖掘行业上下游生产关系,采用贪婪等价搜索算法构建47个行业间因果本质图,明确不同行业间的因果传导路径。然后,将因果传导路径作为收敛交叉映射算法的约束条件,进而自适应计算因果传导路径的最佳时滞期。最后,提出时序-特征混合预测算法,将不同行业间因果传导路径依次作为预测模型的输入,预测不同行业的用电量需求。通过全国47个行业的实际用电量数据,验证了所提方法的适用性和有效性。
Electricity demand can reflect the economic development trends,and the development of upstream and downstream industries leads to the causal relationships between the electricity demands of various industries.Existing industrial electricity consumption prediction lacks the analysis of causal transmission pathways among industries,and the prediction algorithms ignore the coupling relationship between time series and features.This results in issues such as unclear input causal transmission pathways,limited policy relevance,and low prediction accuracy in industrial electricity consumption prediction model.Therefore,this paper proposes a industrial electricity consumption prediction method considering causal transmission pathways in industry development.First,to explore the production relationships among upstream and downstream industries,the greedy equivalence search algorithm is employed to construct a causal essential graph among 47 industries,clarifying the causal transmission pathways among various industries.Then,these causal transmission pathways are used as constraints for the convergent cross-mapping algorithm to adaptively calculate the optimal time lag for causal transmission pathways.Finally,a temporal-feature mixing prediction algorithm is proposed,which sequentially incorporates causal transmission pathways among various industries as inputs to the prediction model to predict the electricity demand across various industries.The applicability and effectiveness of the proposed method are validated using actual electricity consumption data from 47 industries in China.
作者
马伟
刘曌
和敬涵
王小君
李佳明
窦嘉铭
MA Wei;LIU Zhao;HE Jinghan;WANG Xiaojun;LI Jiaming;DOU Jiaming(School of Electrical Engineering,Beijing Jiaotong University,Beijing 100044,China;Electric Power Planning&Engineering Institute,Beijing 100120,China)
出处
《电力系统自动化》
北大核心
2025年第21期98-107,共10页
Automation of Electric Power Systems
基金
国家自然科学基金委员会-联合基金集成项目:“新型配电系统形态演化与安全高效运行的基础理论及方法”(U23B6007)。
关键词
因果传导
预测
行业用电量
贪婪等价搜索
收敛交叉映射
causal transmission
prediction
industrial electricity consumption
greedy equivalence search
convergent cross-mapping
