基于多表融合数据的用户短期用电量预测
|
摘要
用电量预测是用户用能分析的一个重要研究内容,提高预测精度对用户用能分析以及异常检测具有重要意义。利用用电信息系统采集的电、水、气三表数据,提出了基于支持向量机的短期用电量预测方法。该方法首先利用通径分析计算出影响用户用电量的日特征向量的权重以及模糊相似矩阵;然后,通过模糊聚类传递闭包法选取相似日,并将它们作为样本训练支持向量机模型,实现对用户用电量的预测。采用杭州地区2016年的多表融合数据对提出方法的性能进行测试。实验结果表明,多表融合预测相对于单表预测方法,其单用户用电量和小区多用户总用电量的预测相对误差分别减小了6%和1%以上。
Electricity consumption forecasting is one of the most important research topics in the user energy consumption analysis,and the improvement of forecasting accuracy is of great significance to both the user energy consumption analysis and abnormal detection. Based on the data collected by electricity,water and gas meters in an electricity consumption information acquisition system,a short-term electricity consumption forecasting method based on support vector machine(SVM)is proposed. First,path analysis is used to calculate the weights of daily features that affect the user's electricity consumption quantity,as well as the corresponding fuzzy similar matrix. Then,similar days are selected by fuzzy clustering and transitive closure algorithm,and they are further used as samples to train the SVM model,thus realizingtheforecastingofelectricityconsumption.Finally,themulti-meterfusiondatacollectedfromHangzhouCityin 2016 are used to test the performance of the proposed method. Experimental results show that compared with the single-meter forecasting method,the forecasting method based on multi-meter fusion data can reduce the relative forecasting errors of single-andmulti-userelectricityconsumptioninonecertaincommunitybymorethan 6% andmorethan1%,respectively.
Electricity consumption forecasting is one of the most important research topics in the user energy consumption analysis,and the improvement of forecasting accuracy is of great significance to both the user energy consumption analysis and abnormal detection. Based on the data collected by electricity,water and gas meters in an electricity consumption information acquisition system,a short-term electricity consumption forecasting method based on support vector machine(SVM)is proposed. First,path analysis is used to calculate the weights of daily features that affect the user's electricity consumption quantity,as well as the corresponding fuzzy similar matrix. Then,similar days are selected by fuzzy clustering and transitive closure algorithm,and they are further used as samples to train the SVM model,thus realizingtheforecastingofelectricityconsumption.Finally,themulti-meterfusiondatacollectedfromHangzhouCityin 2016 are used to test the performance of the proposed method. Experimental results show that compared with the single-meter forecasting method,the forecasting method based on multi-meter fusion data can reduce the relative forecasting errors of single-andmulti-userelectricityconsumptioninonecertaincommunitybymorethan 6% andmorethan1%,respectively.
引文
[1]康重庆,夏清,张伯明(Kang Chongqing,Xia Qing,Zhang Boming).应用于负荷预测中的回归分析的特殊问题(Research on short term off-line load forecasting in power system)[J].电力系统自动化(Automation of Electric Power Systems),1998,22(10):38-41,49.
[2]宗文婷,卫志农,孙国强,等(Zong Wenting,Wei Zhi?nong,Sun Guoqiang,et al).基于改进高斯过程回归模型的短期负荷区间预测(Short-term load interval predic?tion based on improved Gaussian process regression mod?el)[J].电力系统及其自动化学报(Proceedings of the CSU-EPSA),2017,29(8):22-28.
[3] Trudnowski D J,McReynolds W L,Johnson J M. Realtime very short-term load prediction for power-system au?tomatic generation control[J]. IEEE Trans on Power Sys?tems Technology,2001,9(2):254-260.
[4]张俊芳,吴伊昂,吴军基(Zhang Junfang,Wu Yiang,Wu Junji).基于灰色理论负荷预测的应用研究(Applica?tion of gray system theory in load forecasting)[J].电力自动化设备(Electric Power Automation Equipment),2004,24(5):24-27.
[5] Chen Ying,Luh Peter B,Guan Che,et al. Short-term load forecasting:similar day-based wavelet neural networks[J].IEEE Trans on Power Systems,2010,25(1):322-330.
[6]谢宏,陈志业,牛东晓,等(Xie Hong,Chen Zhiye,Niu Dongxiao,et al).基于小波分解与气象因素影响的电力系统日负荷预测模型研究(The research of daily load forecasting model based on wavelet decomposing and cli?matic influence)[J].中国电机工程学报(Proceedings of the CSEE),2001,21(5):5-10.
[7]牛东晓,魏亚楠(Niu Dongxiao,Wei Yanan).基于FHNN相似日聚类自适应权重的短期电力负荷组合预测(Short-term power load combinatorial forecast adaptively weighted by FHNN similar-day clustering)[J].电力系统自动化(Automation of Electric Power Systems),2013,37(3):54-57.
[8]潘锋,程浩忠,杨镜非,等(Pan Feng,Cheng Haozhong,Yang Jingfei,et al).基于支持向量机的电力系统短期负荷预测(Power system short-term load forecasting based on support vector machines)[J].电网技术(Power System Technology),2004,28(21):39-42.
[9]刘晶,朱锋峰,林辉,等(Liu Jing,Zhu Fengfeng,Lin Hui,et al).基于相似日负荷修正算法的短期负荷预测(Short-term load forecasting based on algorithm of similar days’load modification)[J].计算机工程与设计(Com?puter Engineering and Design),2010,31(6):1279-1282.
[10] Cortes C,Vapnik V. Support-vector networks[J]. Machine Learning,1995,20(3):273-297.
[11]周志华.机器学习[M].北京:清华大学出版社,2016.
[12]黎灿兵,李晓辉,赵瑞,等(Li Canbing,Li Xiaohui,Zhao Rui,et al).电力短期负荷预测相似日选取算法(A nov?el algorithm of selecting similar days for short-term power load forecasting)[J].电力系统自动化(Automation of Electric Power Systems),2008,32(9):69-73.
[13]敬艳辉,邢留伟(Jing Yanhui,Xing Liuwei).通径分析及其应用(Path analysis and its applications)[J].统计教育(Statistical Education),2006(2):24-26.
[14]高新波.模糊聚类分析及其应用[M].西安:西安电子科技大学出版社,2004.
[2]宗文婷,卫志农,孙国强,等(Zong Wenting,Wei Zhi?nong,Sun Guoqiang,et al).基于改进高斯过程回归模型的短期负荷区间预测(Short-term load interval predic?tion based on improved Gaussian process regression mod?el)[J].电力系统及其自动化学报(Proceedings of the CSU-EPSA),2017,29(8):22-28.
[3] Trudnowski D J,McReynolds W L,Johnson J M. Realtime very short-term load prediction for power-system au?tomatic generation control[J]. IEEE Trans on Power Sys?tems Technology,2001,9(2):254-260.
[4]张俊芳,吴伊昂,吴军基(Zhang Junfang,Wu Yiang,Wu Junji).基于灰色理论负荷预测的应用研究(Applica?tion of gray system theory in load forecasting)[J].电力自动化设备(Electric Power Automation Equipment),2004,24(5):24-27.
[5] Chen Ying,Luh Peter B,Guan Che,et al. Short-term load forecasting:similar day-based wavelet neural networks[J].IEEE Trans on Power Systems,2010,25(1):322-330.
[6]谢宏,陈志业,牛东晓,等(Xie Hong,Chen Zhiye,Niu Dongxiao,et al).基于小波分解与气象因素影响的电力系统日负荷预测模型研究(The research of daily load forecasting model based on wavelet decomposing and cli?matic influence)[J].中国电机工程学报(Proceedings of the CSEE),2001,21(5):5-10.
[7]牛东晓,魏亚楠(Niu Dongxiao,Wei Yanan).基于FHNN相似日聚类自适应权重的短期电力负荷组合预测(Short-term power load combinatorial forecast adaptively weighted by FHNN similar-day clustering)[J].电力系统自动化(Automation of Electric Power Systems),2013,37(3):54-57.
[8]潘锋,程浩忠,杨镜非,等(Pan Feng,Cheng Haozhong,Yang Jingfei,et al).基于支持向量机的电力系统短期负荷预测(Power system short-term load forecasting based on support vector machines)[J].电网技术(Power System Technology),2004,28(21):39-42.
[9]刘晶,朱锋峰,林辉,等(Liu Jing,Zhu Fengfeng,Lin Hui,et al).基于相似日负荷修正算法的短期负荷预测(Short-term load forecasting based on algorithm of similar days’load modification)[J].计算机工程与设计(Com?puter Engineering and Design),2010,31(6):1279-1282.
[10] Cortes C,Vapnik V. Support-vector networks[J]. Machine Learning,1995,20(3):273-297.
[11]周志华.机器学习[M].北京:清华大学出版社,2016.
[12]黎灿兵,李晓辉,赵瑞,等(Li Canbing,Li Xiaohui,Zhao Rui,et al).电力短期负荷预测相似日选取算法(A nov?el algorithm of selecting similar days for short-term power load forecasting)[J].电力系统自动化(Automation of Electric Power Systems),2008,32(9):69-73.
[13]敬艳辉,邢留伟(Jing Yanhui,Xing Liuwei).通径分析及其应用(Path analysis and its applications)[J].统计教育(Statistical Education),2006(2):24-26.
[14]高新波.模糊聚类分析及其应用[M].西安:西安电子科技大学出版社,2004.