文章摘要
黄南天,齐斌,刘座铭,等.采用面积灰关联决策的高斯过程回归概率短期负荷预测[J].电力系统自动化,2018,42(23):64-71. DOI: 10.7500/AEPS20180118008.
HUANG Nantian,QI Bin,LIU Zuoming, et al.Probabilistic Short-term Load Forecasting Using Gaussian Process Regression with Area Grey Incidence Decision Making[J].Automation of Electric Power Systems,2018,42(23):64-71. DOI: 10.7500/AEPS20180118008.
采用面积灰关联决策的高斯过程回归概率短期负荷预测
Probabilistic Short-term Load Forecasting Using Gaussian Process Regression with Area Grey Incidence Decision Making
DOI:10.7500/AEPS20180118008
关键词: 概率预测  短期负荷预测  综合决策  面积灰关联决策  高斯过程回归  协方差函数
KeyWords: probabilistic prediction  short-term load forecasting  comprehensive decision making  area grey incidence decision making  Gaussian process regression  covariance function
上网日期:2018-10-15
基金项目:国家重点研发计划资助项目(2016YFB0900104);吉林省科技发展计划资助项目(20160411003XH)
作者单位E-mail
黄南天 现代电力系统仿真控制与绿色电能新技术教育部重点实验室(东北电力大学), 吉林省吉林市 132012 huangnantian@126.com 
齐斌 现代电力系统仿真控制与绿色电能新技术教育部重点实验室(东北电力大学), 吉林省吉林市 132012  
刘座铭 国网吉林省电力有限公司电力科学研究院, 吉林省长春市 130021  
蔡国伟 现代电力系统仿真控制与绿色电能新技术教育部重点实验室(东北电力大学), 吉林省吉林市 132012  
邢恩恺 现代电力系统仿真控制与绿色电能新技术教育部重点实验室(东北电力大学), 吉林省吉林市 132012  
摘要:
      为克服概率负荷预测各评价指标相互冲突,难以确定最优预测模型难题,提出采用面积灰关联决策的高斯过程回归(GPR)概率短期负荷预测新方法。首先,构建综合评价指标集合,全面评估基于不同协方差函数的GPR模型预测效果,得到综合评价矩阵。然后,采用熵权法对各指标客观赋权,并在此基础上,使用面积灰关联决策对各模型排序,确定最优GPR概率预测模型。最终,以该模型开展概率预测。实验表明,相较传统距离灰关联决策,面积灰关联决策更明确地分辨方案间差异,结论更可靠。最优GPR模型在保证确定性预测精度的同时,相较预测误差分布特性统计法,准确刻画了负荷的波动性,预测区间更加精确可靠,区间上限明显更低,有助于为决策提供更多有效信息。
Abstract:
      In order to overcome the limitation of determining the optimal forecasting model caused by the conflict between the criteria of probabilistic load forecasting, a new probabilistic short-term load forecasting method using Gaussian process regression(GPR)area grey incidence decision making is proposed. Firstly, a set of comprehensive evaluation criteria are constructed to comprehensively evaluate the prediction effect of GPR model based on different covariance functions, which obtain a comprehensive evaluation matrix. Then the criteria weights are determined by entropy method. On this basis, each model is ranked by the area grey incidence decision making to determine the optimal GPR model. Finally, the model is used to probabilistic prediction. The experiment results show that, compared with traditional distance grey incidence decision making, area gray incidence decision making can more clearly distinguish the difference between schemes and the conclusion is more reliable. While, compared with the statistics method of probability distribution of forecasting errors, the deterministic forecasting accuracy is ensured, the optimal GPR model accurately characterizes load variation, and prediction interval is sharper and more reliable, the upper limit of interval is lower, which helps to provide more effective information for decision making.
查看全文(Free!)   查看附录   查看/发表评论  下载PDF阅读器