基于强化自组织映射和径向基神经网络的短期负荷预测
|
摘要
径向基(radial basis function,RBF)神经网络因其泛化能力强、收敛速度快的特点广泛应用于负荷预测。但传统采用K-means和自组织映射(self-organizing map,S O M)训练R B F径向基中心的方法因其全局搜索能力偏弱,仍然存在容易陷入局部最优解的问题,严重制约了RBF预测精度的提高。针对此问题,提出了一种基于强化学习(reinforcement learning,RL)改进的RBF短期负荷预测方法。强化学习通过环境的反馈不断完善搜索策略,具有非常突出的全局搜索能力。所提方法将强化学习以环境反馈修正搜索策略的机制应用于SOM,大幅增强了SOM的全局搜索能力,使其获得逼近最优的径向基中心,提高RBF负荷预测精度。以英国某地区2016年5～9月的负荷数据进行仿真实验。结果显示,与采用K-means和SOM方法训练径向基中心的RBF相比,所提的强化学习改进RBF方法的负荷预测平均相对误差分别由4.58%和4.37%降低至3.30%。
Radial basis function(RBF) neural network is widely applied in short-term load forecasting because of its strong generalization ability and fast convergence speed. However, the traditional method of using K-means and self-organizing map(SOM) for the training of the radial basis center of RBF possess certain shortcomings. Due to the weak global searching capability, this method falls into local optimal solution easily, which seriously restricts the improvement of the precision of load forecasting of RBF. To relieve the restriction, an improved RBF based on reinforcement learning(RL) is proposed for short-term load forecasting. The proposed method dramatically enhances the global searching capability of SOM by applying the feedback-correction mechanism of RL in SOM, which drives it to approach to the optimal radical basis center. As a result, the precision of short-term load forecasting of RBF is improved. To verify the proposed method, simulation case is carried out based on the load data of a certain area in UK from May to September 2016.Comparing with K-means method and SOM method, the simulation results show that the average relative error is notably reduced by using proposed method, which demonstrates the correctness and superiority of the proposed method.
Radial basis function(RBF) neural network is widely applied in short-term load forecasting because of its strong generalization ability and fast convergence speed. However, the traditional method of using K-means and self-organizing map(SOM) for the training of the radial basis center of RBF possess certain shortcomings. Due to the weak global searching capability, this method falls into local optimal solution easily, which seriously restricts the improvement of the precision of load forecasting of RBF. To relieve the restriction, an improved RBF based on reinforcement learning(RL) is proposed for short-term load forecasting. The proposed method dramatically enhances the global searching capability of SOM by applying the feedback-correction mechanism of RL in SOM, which drives it to approach to the optimal radical basis center. As a result, the precision of short-term load forecasting of RBF is improved. To verify the proposed method, simulation case is carried out based on the load data of a certain area in UK from May to September 2016.Comparing with K-means method and SOM method, the simulation results show that the average relative error is notably reduced by using proposed method, which demonstrates the correctness and superiority of the proposed method.
引文
[1]牛东晓,曹树华,赵磊,等.电力负荷预测技术及其应用[M].北京:中国电力出版社,1998.
[2]黎静华,赖昌伟.考虑气象因素的短期光伏出力预测的奇异谱分析方法[J].电力自动化设备,2018,38(5):50-57,76.Li Jinghua,Lai Changwei.Singular spectrum analysis method for short-term photovoltaic output prediction considering meteorological factors[J].Electric Power Automation Equipment,2018,38(5):50-57,76(in Chinese).
[3]赵唯嘉,张宁,康重庆,等.光伏发电出力的条件预测误差概率分布估计方法[J].电力系统自动化,2015,39(16):8-15.Zhao Weijia,Zhang Ning,Kang Chongqing,et al.A method of probabilistic distribution of conditional forecast error for photovoltaic power generation[J].Automation of Electric Power Systems,2015,39(16):8-15(in Chinese).
[4]兰飞,桑川川,梁浚杰,等.基于条件Copula函数的风电功率区间预测[J].中国电机工程学报,2016,36(S1):79-86.Lan Fei,Sang Chuanchuan,Liang Junjie,et al.Interval prediction for wind power based on conditional Copula function[J].Proceedings of the CSEE,2016,36(S1):79-86(in Chinese).
[5]单英浩,付青,耿炫,等.基于改进BP-SVM-ELM与粒子化SOM-LSF的微电网光伏发电组合预测方法[J].中国电机工程学报,2016,36(12):3334-3342.Shan Yinghao,Fu Qing,Geng Xuan,et al.Combined forecasting of photovoltaic power generation in microgrid based on the improved BP-SVM-ELM and SOM-LSF with particlization[J].Proceedings of the CSEE,2016,36(12):3334-3342(in Chinese).
[6]Malvoni M,Giorgi M G D,Congedo P M.Photovoltaic forecast based on hybrid PCA-LSSVM using dimensionality reducted data[J].Neurocomputing,2016,211:72-83.
[7]王东,史晓霞,尹交英.不同核函数的支持向量机用于空调负荷预测的对比研究[J].电工技术学报,2015(S1):531-535.Wang Dong,Shi Xiaoxia,Yin Jiaoying.Prediction on hourly load of air conditioning by RBF support vector machine[J].Transactions of China Electrotechnical Society,2015(S1):531-535(in Chinese).
[8]彭显刚,胡松峰,吕大勇.基于RBF神经网络的短期负荷预测方法综述[J].电力系统保护与控制,2011,39(17):144-148.Peng Xiangang,Hu Songfeng,Lyu Dayong.Review on grid short-term load forecasting methods based on RBF neural network power[J].System Protection and Control,2011,39(17):144-148(in Chinese).
[9]何耀耀,许启发,杨善林,等.基于RBF神经网络分位数回归的电力负荷概率密度预测方法[J].中国电机工程学报,2013,33(1):93-98.He Yaoyao,Xu Qifa,Yang Shanlin,et al.A power load probability density forecasting method based on RBF neural network quantile regression[J].Proceedings of the CSEE,2013,33(1):93-98(in Chinese).
[10]张智晟,于道林.考虑需求响应综合影响因素的RBF-NN短期负荷预测模型[J].中国电机工程学报,2018,36(6):1631-1638.Zhang Zhisheng,Yu Daolin.RBF-NN based short-term load forecasting model considering comprehensive factors affecting demand response[J].Proceedings of the CSEE,2018,36(6):1631-1638(in Chinese).
[11]付强,陈特放,朱佼佼.采用自组织RBF网络算法的变压器故障诊断[J].高电压技术,2012,38(6):1368-1375.Fu Qiang,Chen Tefang,Zhu Jiaojiao.Transformer fault diagnosis using self-adaptive RBF neural network algorithm[J].High Voltage Engineering,2012,38(6):1368-1375(in Chinese).
[12]韩红桂,乔俊飞,薄迎春.基于信息强度的RBF神经网络结构设计研究[J].自动化学报,2012,38(7):1083-1090.Han Honggui,Qiao Junfei,Bo Yingchun.On structure design for RBF neural network based on information strength[J].Acta Automatice Sinice,2012,38(7):1083-1090(in Chinese).
[13]张毅,姜思博,李铮.改进的遗传灰色RBF模型的短期电力负荷预测[J].电测与仪表,2014,51(5):1-4.Zhang Yi,Jiang Sibo,Li Zheng.A gray RBF model improved by genetic algorithm for short-term load forecasting[J].Electrical Measurement&Instrumentation,2014,51(5):1-4(in Chinese).
[14]师彪,李郁侠,于新花,等.基于改进粒子群-径向基神经网络模型的短期电力负荷预测[J].电网技术,2009,33(17):180-184.Shi Biao,Li Yuxia,Yu Xinhua,et al.Short-term load forecasting based on modified particle swarm optimization and radial basis function neural network model[J].Power System Technology,2009,33(17):180-184(in Chinese).
[15]回立川,于淼,梁芷睿.应用近邻传播算法改进RBF的短期负荷预测[J].电力系统及其自动化学报,2015,27(1):69-73.Hui Lichuan,Yu Miao,Liang Zhirui.Research on electrical load short-term forecasting via radial basis function neural network designed by affinity propagation[J].Proceedings of the CSU-EPSA,2015,27(1):69-73(in Chinese).
[16]黄震华,向阳,张波,等.一种进行K-Means聚类的有效方法[J].模式识别与人工智能,2010,23(4):516-521.Huang Zhenhua,Xiang Yang,Zhang Bo,et al.An efficient method for K-means clustering[J].PR&AI,2010,23(4):516-521(in Chinese).
[17]高山,单渊达.基于径向基函数网络的短期负荷预测[J].电力系统自动化,1999,23(5):31-34.Gao Shan,Shan Yuanda.A new short-term load forecasting algorithm based on radial basis function network[J].Automation of Electric Power Systems,1999,23(5):31-34(in Chinese).
[18]李智勇,吴晶莹,吴为麟,等.基于自组织映射神经网络的电力用户负荷曲线聚类[J].电力系统自动化,2008,32(15):66-70.Li Zhiyong,Wu Jingying,Wu Weilin,et al.Power customers load profile clustering using the SOM neural network[J].Automation of Electric Power Systems,2008,32(15):66-70(in Chinese).
[19]刘威,张东霞,王新迎,等.基于深度强化学习的电网紧急控制策略研究[J].中国电机工程学报,2018,38(1):109-119,347.Liu Wei,Zhang Dongxia,Wang Xinying,et al.Research on grid emergency control strategy based on deepening reinforcement learning[J].Proceedings of the CSEE,2018,38(1):109-119,347(in Chinese).
[20]韩传家,张孝顺,余涛,等.风险调度中引入知识迁移的细菌觅食强化学习优化算法[J].电力系统自动化,2017,41(8):69-77,97.Han Chuanjia,Zhang Xiaoshun,Yu Tao,et al.Bacterial foraging reinforcement learning optimization algorithm with knowledge transfer introduced into risk scheduling[J].Automation of Electric Power Systems,2017,41(8):69-77,97(in Chinese).
[21]Williams R J.Simple statistical gradient-following algorithms for connectionist reinforcement learning[J].Machine Learning,1992,8(3-4):229-256.
[22]郭跃东,宋旭东.梯度下降法的分析和改进[J].科技展望,2016,26(15):115-117.Guo Yuedong,Song Xudong.Analysis and improvement of gradient descent method[J].Technology Outlook,2016,26(15):115-117(in Chinese).
[23]黎静华,黄乾,韦善阳,等.基于S-BGD和梯度累积策略的改进深度学习方法及其在光伏出力预测中的应用[J].电网技术,2017,41(10):3292-3300.Li Jinghua,Huang Qian,Wei Shanyang,et al.Improved deep learning algorithm based on S-BGD and gradient pile strategy and its application in PV power forecasting[J].Power System Technology,2017,41(10):3292-3300(in Chinese).
[24]刘颖超,张纪元.梯度下降法[J].南京理工大学学报,1993(2):12-16.Liu Yingchao,Zhang Jiyuan.Gradient descent method[J].Journal of Nanjing University of Science and Technology,1993(2):12-16(in Chinese).
[2]黎静华,赖昌伟.考虑气象因素的短期光伏出力预测的奇异谱分析方法[J].电力自动化设备,2018,38(5):50-57,76.Li Jinghua,Lai Changwei.Singular spectrum analysis method for short-term photovoltaic output prediction considering meteorological factors[J].Electric Power Automation Equipment,2018,38(5):50-57,76(in Chinese).
[3]赵唯嘉,张宁,康重庆,等.光伏发电出力的条件预测误差概率分布估计方法[J].电力系统自动化,2015,39(16):8-15.Zhao Weijia,Zhang Ning,Kang Chongqing,et al.A method of probabilistic distribution of conditional forecast error for photovoltaic power generation[J].Automation of Electric Power Systems,2015,39(16):8-15(in Chinese).
[4]兰飞,桑川川,梁浚杰,等.基于条件Copula函数的风电功率区间预测[J].中国电机工程学报,2016,36(S1):79-86.Lan Fei,Sang Chuanchuan,Liang Junjie,et al.Interval prediction for wind power based on conditional Copula function[J].Proceedings of the CSEE,2016,36(S1):79-86(in Chinese).
[5]单英浩,付青,耿炫,等.基于改进BP-SVM-ELM与粒子化SOM-LSF的微电网光伏发电组合预测方法[J].中国电机工程学报,2016,36(12):3334-3342.Shan Yinghao,Fu Qing,Geng Xuan,et al.Combined forecasting of photovoltaic power generation in microgrid based on the improved BP-SVM-ELM and SOM-LSF with particlization[J].Proceedings of the CSEE,2016,36(12):3334-3342(in Chinese).
[6]Malvoni M,Giorgi M G D,Congedo P M.Photovoltaic forecast based on hybrid PCA-LSSVM using dimensionality reducted data[J].Neurocomputing,2016,211:72-83.
[7]王东,史晓霞,尹交英.不同核函数的支持向量机用于空调负荷预测的对比研究[J].电工技术学报,2015(S1):531-535.Wang Dong,Shi Xiaoxia,Yin Jiaoying.Prediction on hourly load of air conditioning by RBF support vector machine[J].Transactions of China Electrotechnical Society,2015(S1):531-535(in Chinese).
[8]彭显刚,胡松峰,吕大勇.基于RBF神经网络的短期负荷预测方法综述[J].电力系统保护与控制,2011,39(17):144-148.Peng Xiangang,Hu Songfeng,Lyu Dayong.Review on grid short-term load forecasting methods based on RBF neural network power[J].System Protection and Control,2011,39(17):144-148(in Chinese).
[9]何耀耀,许启发,杨善林,等.基于RBF神经网络分位数回归的电力负荷概率密度预测方法[J].中国电机工程学报,2013,33(1):93-98.He Yaoyao,Xu Qifa,Yang Shanlin,et al.A power load probability density forecasting method based on RBF neural network quantile regression[J].Proceedings of the CSEE,2013,33(1):93-98(in Chinese).
[10]张智晟,于道林.考虑需求响应综合影响因素的RBF-NN短期负荷预测模型[J].中国电机工程学报,2018,36(6):1631-1638.Zhang Zhisheng,Yu Daolin.RBF-NN based short-term load forecasting model considering comprehensive factors affecting demand response[J].Proceedings of the CSEE,2018,36(6):1631-1638(in Chinese).
[11]付强,陈特放,朱佼佼.采用自组织RBF网络算法的变压器故障诊断[J].高电压技术,2012,38(6):1368-1375.Fu Qiang,Chen Tefang,Zhu Jiaojiao.Transformer fault diagnosis using self-adaptive RBF neural network algorithm[J].High Voltage Engineering,2012,38(6):1368-1375(in Chinese).
[12]韩红桂,乔俊飞,薄迎春.基于信息强度的RBF神经网络结构设计研究[J].自动化学报,2012,38(7):1083-1090.Han Honggui,Qiao Junfei,Bo Yingchun.On structure design for RBF neural network based on information strength[J].Acta Automatice Sinice,2012,38(7):1083-1090(in Chinese).
[13]张毅,姜思博,李铮.改进的遗传灰色RBF模型的短期电力负荷预测[J].电测与仪表,2014,51(5):1-4.Zhang Yi,Jiang Sibo,Li Zheng.A gray RBF model improved by genetic algorithm for short-term load forecasting[J].Electrical Measurement&Instrumentation,2014,51(5):1-4(in Chinese).
[14]师彪,李郁侠,于新花,等.基于改进粒子群-径向基神经网络模型的短期电力负荷预测[J].电网技术,2009,33(17):180-184.Shi Biao,Li Yuxia,Yu Xinhua,et al.Short-term load forecasting based on modified particle swarm optimization and radial basis function neural network model[J].Power System Technology,2009,33(17):180-184(in Chinese).
[15]回立川,于淼,梁芷睿.应用近邻传播算法改进RBF的短期负荷预测[J].电力系统及其自动化学报,2015,27(1):69-73.Hui Lichuan,Yu Miao,Liang Zhirui.Research on electrical load short-term forecasting via radial basis function neural network designed by affinity propagation[J].Proceedings of the CSU-EPSA,2015,27(1):69-73(in Chinese).
[16]黄震华,向阳,张波,等.一种进行K-Means聚类的有效方法[J].模式识别与人工智能,2010,23(4):516-521.Huang Zhenhua,Xiang Yang,Zhang Bo,et al.An efficient method for K-means clustering[J].PR&AI,2010,23(4):516-521(in Chinese).
[17]高山,单渊达.基于径向基函数网络的短期负荷预测[J].电力系统自动化,1999,23(5):31-34.Gao Shan,Shan Yuanda.A new short-term load forecasting algorithm based on radial basis function network[J].Automation of Electric Power Systems,1999,23(5):31-34(in Chinese).
[18]李智勇,吴晶莹,吴为麟,等.基于自组织映射神经网络的电力用户负荷曲线聚类[J].电力系统自动化,2008,32(15):66-70.Li Zhiyong,Wu Jingying,Wu Weilin,et al.Power customers load profile clustering using the SOM neural network[J].Automation of Electric Power Systems,2008,32(15):66-70(in Chinese).
[19]刘威,张东霞,王新迎,等.基于深度强化学习的电网紧急控制策略研究[J].中国电机工程学报,2018,38(1):109-119,347.Liu Wei,Zhang Dongxia,Wang Xinying,et al.Research on grid emergency control strategy based on deepening reinforcement learning[J].Proceedings of the CSEE,2018,38(1):109-119,347(in Chinese).
[20]韩传家,张孝顺,余涛,等.风险调度中引入知识迁移的细菌觅食强化学习优化算法[J].电力系统自动化,2017,41(8):69-77,97.Han Chuanjia,Zhang Xiaoshun,Yu Tao,et al.Bacterial foraging reinforcement learning optimization algorithm with knowledge transfer introduced into risk scheduling[J].Automation of Electric Power Systems,2017,41(8):69-77,97(in Chinese).
[21]Williams R J.Simple statistical gradient-following algorithms for connectionist reinforcement learning[J].Machine Learning,1992,8(3-4):229-256.
[22]郭跃东,宋旭东.梯度下降法的分析和改进[J].科技展望,2016,26(15):115-117.Guo Yuedong,Song Xudong.Analysis and improvement of gradient descent method[J].Technology Outlook,2016,26(15):115-117(in Chinese).
[23]黎静华,黄乾,韦善阳,等.基于S-BGD和梯度累积策略的改进深度学习方法及其在光伏出力预测中的应用[J].电网技术,2017,41(10):3292-3300.Li Jinghua,Huang Qian,Wei Shanyang,et al.Improved deep learning algorithm based on S-BGD and gradient pile strategy and its application in PV power forecasting[J].Power System Technology,2017,41(10):3292-3300(in Chinese).
[24]刘颖超,张纪元.梯度下降法[J].南京理工大学学报,1993(2):12-16.Liu Yingchao,Zhang Jiyuan.Gradient descent method[J].Journal of Nanjing University of Science and Technology,1993(2):12-16(in Chinese).