电力市场中基于Attention-LSTM的短期负荷预测模型
|
摘要
电力市场环境下,精准的短期负荷预测可以保障电网安全稳定运行,但电价的实时波动增加了负荷变化的复杂性,加大了预测难度。针对这一问题,采用最大信息系数法分析电价及历史负荷与当前时刻负荷的相关性,为预测模型输入特征的确定提供依据。在此基础上,提出了基于Attention-LSTM (attention long short-term memory,Attention-LSTM)网络的短期负荷预测模型。该模型充分利用负荷的时序特性,并采用Attention机制突出对负荷预测起到关键作用的输入特征。以澳大利亚某地区真实数据为算例,分别应用Attention-LSTM模型与其他模型进行仿真实验。结果表明,所提方法在预测精度和算法鲁棒性方面均优于其他模型。
In electricity market environment, accurate short-term load forecasting can ensure safe and stable operation of power grid, and save a lot of cost for power generation, electricity sales and power consumption units. However, real-time fluctuations in electricity prices increase complexity of load changes and difficulty of forecasting. In view of such situation, the maximum information coefficient method was used to analyze the correlation of current time load with electricity price and historical load respectively in this paper. The input feature vector of the prediction model could be determined with this method. And on this basis, a short-term load forecasting model based on Attention-LSTM(Attention long short-term memory) network was proposed. The model took full advantage of the timing characteristics of the load and used Attention mechanism to highlight input features playing key role in load forecasting. Taking the real data of a certain area in Australia as an example, simulations were performed with the proposed Attention-LSTM model and other models. Results show that the proposed model outperforms other models in terms of prediction accuracy and algorithm robustness.
In electricity market environment, accurate short-term load forecasting can ensure safe and stable operation of power grid, and save a lot of cost for power generation, electricity sales and power consumption units. However, real-time fluctuations in electricity prices increase complexity of load changes and difficulty of forecasting. In view of such situation, the maximum information coefficient method was used to analyze the correlation of current time load with electricity price and historical load respectively in this paper. The input feature vector of the prediction model could be determined with this method. And on this basis, a short-term load forecasting model based on Attention-LSTM(Attention long short-term memory) network was proposed. The model took full advantage of the timing characteristics of the load and used Attention mechanism to highlight input features playing key role in load forecasting. Taking the real data of a certain area in Australia as an example, simulations were performed with the proposed Attention-LSTM model and other models. Results show that the proposed model outperforms other models in terms of prediction accuracy and algorithm robustness.
引文
[1]叶瑰昀,罗耀华,刘勇,等.基于ARMA模型的电力负荷预测方法研究[J].信息技术,2002(6):74-76.Ye Guiyun,Luo Yaohua,Liu Yong,et al.Research on method of power system load forecasting based on ARMA model[J].Information Technology,2002(6):74-76(in Chinese).
[2]彭鹏,彭佳红.基于多元线性回归模型的电力负荷预测研究[J].中国安全生产科学技术,2011,7(9):158-161.Peng Peng,Peng Jiahong.Power load based on multiple linear regression model prediction[J].Journal of Safety Science and Technology,2011,7(9):158-161(in Chinese).
[3]陈鸿琳,李欣然,冷华,等.运用PSO和GRNN的短期负荷二维组合预测[J].电力系统及其自动化学报,2018,30(2):86-89.Chen Honglin,Li Xinran,Leng Hua,et al.Bidirectional combined short-term load forecasting by using PSO and GRNN[J].Proceedings of the CSU-EPSA,2018,30(2):86-89(in Chinese).
[4]Xi Chen,Jing Qu,Zhao Yangdong.An improved load forecast model using factor analysis:an Australian case study[C]//Proceedings of the2017 IEEE International Conference on Information and Automation(ICIA).Macau SAR,China:IEEE,2017:903-908.
[5]何耀耀,刘瑞,撖奥洋.基于实时电价与支持向量分位数回归的短期电力负荷概率密度预测方法[J].中国电机工程学报,2017,37(3):768-774.He Yaoyao,Liu Rui,Han Aoyang.Short-term power load probability density forecasting method based on real time price and support vector quantile regression[J].Proceedings of the CSEE,2017,37(3):768-774(in Chinese).
[6]葛少云,贾鸥莎,刘洪.基于遗传灰色神经网络模型的实时电价条件下短期电力负荷预测[J].电网技术,2012,36(1):224-229.Ge Shaoyun,Jia Ousha,Liu Hong.A gray neural network model improved by genetic algorithm for short-term load forecasting in price-sensitive environment[J].Power System Technology,2012,36(1):224-229(in Chinese).
[7]孔祥玉,郑锋,鄂志君,等.基于深度信念网络的短期负荷预测方法[J].电力系统自动化,2018,42(5):133-138.Kong Xiangyu,Zheng Feng,E Zhijun,et al.Short term load forecasting method based on deep belief network[J].Automation of Electric Power Systems,2018,42(5):133-138(in Chinese).
[8]Graves A.Supervised sequence labeling with recurrent neural networks[M].Berlin:Springer,2012,32-39.
[9]朱乔木,李弘毅,王子琪,等.基于长短期记忆网络的风电场发电功率超短期预测[J].电网技术,2017,41(12):3797-3802.Zhu Qiaomu,Li Hongyi,Wang Ziqi,et al.Short-term wind power forecasting based on LSTM[J].Power System Technology,2017,41(12):3797-3802(in Chinese).
[10]张宇航,邱才明,贺兴.一种基于LSTM神经网络的短期用电负荷预测方法[J].电力信息与通信技术,2017,15(9):19-24.Zhang Yuhang,Qiu Caiming,He Xing.A short-term load forecasting method based on LSTM neural network[J].Electric Power Information and Communication Technology,2017,15(9):19-24(in Chinese).
[11]Reshef D N,Reshef Y A,Finucane H K,et al.Detecting novel associations in large data sets[J].Science,2011,334(6062):1518-1524.
[12]Justin B K,Gurinder S A.Equitability,mutual information,and the maximal information coefficient[J].PNAs,2014,111(9):3354-3359.
[13]王中伟,宋宏,李帅,等.基于对数变换和最大信息系数PCA的过程监测[J].科学技术与工程,2017,17(16):259-265.Wang Zhongwei,Song Hong,Li Shuai,et al.Process monitoring based on logarithmic transformation and maximum information coefficient PCA[J].Science Technology and Engineering,2017,17(16):259-265(in Chinese).
[14]Australian Energy Market Operator.AEMO’s first national gas forecasting report[R/OL].[2014-11-17].http://www.aemo.com.au.
[15]Kolen J,Kremer S,Frasconi P,et al.Gradient flow in recurrent nets:the difficulty of learning long-term dependencies[M].Wiley-IEEEPress,New York,2001:237-243.
[16]Hochreiter S,Schmidhuber J.Long short-term memory[J].Neural computation,1997,9(8):1735-1780.
[17]朱星嘉,李红莲,吕学强,等.一种改进的Attention-Based LSTM特征选择模型[J].北京信息科技大学学报(自然科学版),2018,33(2):54-59.Zhu Xingjia,Li Honglian,Lu Xueqiang,et al.An improved Attention-based LSTM feature selection model[J].Journal of Beijing Information Science&Technology University,2018,33(2):54-59(in Chinese).
[18]张冲.基于Attention-based LSTM模型的文本分类技术的研究[D].南京:南京大学,2016.
[2]彭鹏,彭佳红.基于多元线性回归模型的电力负荷预测研究[J].中国安全生产科学技术,2011,7(9):158-161.Peng Peng,Peng Jiahong.Power load based on multiple linear regression model prediction[J].Journal of Safety Science and Technology,2011,7(9):158-161(in Chinese).
[3]陈鸿琳,李欣然,冷华,等.运用PSO和GRNN的短期负荷二维组合预测[J].电力系统及其自动化学报,2018,30(2):86-89.Chen Honglin,Li Xinran,Leng Hua,et al.Bidirectional combined short-term load forecasting by using PSO and GRNN[J].Proceedings of the CSU-EPSA,2018,30(2):86-89(in Chinese).
[4]Xi Chen,Jing Qu,Zhao Yangdong.An improved load forecast model using factor analysis:an Australian case study[C]//Proceedings of the2017 IEEE International Conference on Information and Automation(ICIA).Macau SAR,China:IEEE,2017:903-908.
[5]何耀耀,刘瑞,撖奥洋.基于实时电价与支持向量分位数回归的短期电力负荷概率密度预测方法[J].中国电机工程学报,2017,37(3):768-774.He Yaoyao,Liu Rui,Han Aoyang.Short-term power load probability density forecasting method based on real time price and support vector quantile regression[J].Proceedings of the CSEE,2017,37(3):768-774(in Chinese).
[6]葛少云,贾鸥莎,刘洪.基于遗传灰色神经网络模型的实时电价条件下短期电力负荷预测[J].电网技术,2012,36(1):224-229.Ge Shaoyun,Jia Ousha,Liu Hong.A gray neural network model improved by genetic algorithm for short-term load forecasting in price-sensitive environment[J].Power System Technology,2012,36(1):224-229(in Chinese).
[7]孔祥玉,郑锋,鄂志君,等.基于深度信念网络的短期负荷预测方法[J].电力系统自动化,2018,42(5):133-138.Kong Xiangyu,Zheng Feng,E Zhijun,et al.Short term load forecasting method based on deep belief network[J].Automation of Electric Power Systems,2018,42(5):133-138(in Chinese).
[8]Graves A.Supervised sequence labeling with recurrent neural networks[M].Berlin:Springer,2012,32-39.
[9]朱乔木,李弘毅,王子琪,等.基于长短期记忆网络的风电场发电功率超短期预测[J].电网技术,2017,41(12):3797-3802.Zhu Qiaomu,Li Hongyi,Wang Ziqi,et al.Short-term wind power forecasting based on LSTM[J].Power System Technology,2017,41(12):3797-3802(in Chinese).
[10]张宇航,邱才明,贺兴.一种基于LSTM神经网络的短期用电负荷预测方法[J].电力信息与通信技术,2017,15(9):19-24.Zhang Yuhang,Qiu Caiming,He Xing.A short-term load forecasting method based on LSTM neural network[J].Electric Power Information and Communication Technology,2017,15(9):19-24(in Chinese).
[11]Reshef D N,Reshef Y A,Finucane H K,et al.Detecting novel associations in large data sets[J].Science,2011,334(6062):1518-1524.
[12]Justin B K,Gurinder S A.Equitability,mutual information,and the maximal information coefficient[J].PNAs,2014,111(9):3354-3359.
[13]王中伟,宋宏,李帅,等.基于对数变换和最大信息系数PCA的过程监测[J].科学技术与工程,2017,17(16):259-265.Wang Zhongwei,Song Hong,Li Shuai,et al.Process monitoring based on logarithmic transformation and maximum information coefficient PCA[J].Science Technology and Engineering,2017,17(16):259-265(in Chinese).
[14]Australian Energy Market Operator.AEMO’s first national gas forecasting report[R/OL].[2014-11-17].http://www.aemo.com.au.
[15]Kolen J,Kremer S,Frasconi P,et al.Gradient flow in recurrent nets:the difficulty of learning long-term dependencies[M].Wiley-IEEEPress,New York,2001:237-243.
[16]Hochreiter S,Schmidhuber J.Long short-term memory[J].Neural computation,1997,9(8):1735-1780.
[17]朱星嘉,李红莲,吕学强,等.一种改进的Attention-Based LSTM特征选择模型[J].北京信息科技大学学报(自然科学版),2018,33(2):54-59.Zhu Xingjia,Li Honglian,Lu Xueqiang,et al.An improved Attention-based LSTM feature selection model[J].Journal of Beijing Information Science&Technology University,2018,33(2):54-59(in Chinese).
[18]张冲.基于Attention-based LSTM模型的文本分类技术的研究[D].南京:南京大学,2016.