基于遗传算法的RBF神经网络用于配电网线损计算
|
摘要
配电网线损作为电力系统的一项重要的技术经济指标,长期以来受到电力企业及相关部门的广泛重视。特别是电力的市场化改革以来,线损率己经直接与电价挂钩,影响着企业的经营效益。因此,准确而简便的线损计算,对于考核企业线损管理工作成效、制定各项有效地降损措施是极其重要的。
本文首先对现有的理论线损计算方法进行了深入分析,指出了目前各种线损计算方法的局限性。并针对我国配电网具有元器件数量众多、分布复杂,自动化程度普遍较低,原始数据不易收集等特点,提出了一种基于遗传算法的RBF神经网络用于配电网线损计算的新方法。该方法通过RBF神经网络的空间拟合性和径向基函数的局部响应特性映射配电线路参数与配电网线损之间的非线性复杂关系,并针对传统的RBF网络学习方法中,隐含层与输出层结构参数的确定相互独立,输出层权重训练容易陷入局部最小等缺点,应用遗传算法对整个RBF网络进行优化,将RBF网络不同的中心和其对应的宽度及各个调节权重统一编码,加强了RBF网络隐含层和输出层的合作关系,并利用遗传算法全局搜索的功能特性,使得整个网络模型达到全局最优。此外,对遗传算法本身的遗传机制作出了相应的改进,使遗传操作更加完善。
为了验证本文提出的方法的实用性和可行性,分别以某地区68条配电线路和天津滨海供电局67条配电线路的线路特征参数为样本,进行了线损的实例仿真计算。试验仿真结果表明以遗传算法优化的RBF网络,具有网络模型简单、训练速度快、计算精确度高等优点,并具有很强的实用性和推广性。利用神经网络的拟合能力合扩展能力,可以较为准确的记忆可获取的配电线路特征参数与线路线损之间的非线性映射关系,从而进行较为精确的线损计算。
最后,本文以Borland C++ Builder 5.0作为软件开发平台,基于面向对象程序设计的思想,开发出一套适应于配电网线损计算的可视化软件。软件具有配电线路图绘制,元件运行数据录入,数据库管理,线损计算,报表输出等功能。人机界面友好,并提供了大量的快捷操作,具有国际标准软件界面风格,易于用户学习掌握。软件设计后期,进行了多次测试和使用,确保了软件的完整性、安全性和可靠性。
The line losses of power distribution network, which had caught much attention of power enterprises and many related organizations, plays a very important role in evaluating the economics of a power system. Since the reform of the power industry, line losses has directly affected the power price, which is vital to the benefit of enterprises.Accurate and convenient methods of calculating the line losses is very important not only to the evaluation of line losses management, but also to the formulations of losses decreasing measurement.
Firstly existing methods for calculation of the theoretical line losses is investigated in detail in this paper, and the limitation of them is also pointed out. According to the characteristics of power distribution system in our country, this paper presents a RBFNN based on Genetic Algorithm, which is applied in calculating the line losses of power distribution systems. The radial basis fuction (RBF) neural network, due to its nonlinear processing ability, is used to map complex nonlinear relationship between the line losses and the feature factors in power distribution systems. But in the traditional methods, the parameters of the hidden layer and the output layer are trained separately. The Genetic Algorithm (GA), a global optimization algorithm simulating organic evolution, is used to optimize the parameters of the RBF network. The coding method put all the parameters in one chromosome optimize them synchronously. It can strengthen the cooperation between the hidden layer and the output layer, and optimizes the neural work to avoid the network parameters suffering into partial minimum.
In order to validate the practicability and feasibility of this method, this paper has two simulations respectively. A distribution network in some aera with 68 lines and the distribution network of Binhai power company with 67 lines are used as examples. The simulation recults show that the RBF neural network based on GA has so many advantages such as simple configuration, high convergence speed, high precision and so on. It can memory the complex nonlinear relationship between the line losses and the feature factors in distribution network accurately. So the method presented in this paper has good practicability and should be popularized in other areas.
At last, based on the OOP (Object Oriented Programming) ideology, this paper developed visual software for distribution network losses calculation on C++ Builder platform. The proposed software has many fuctions including that circuit diagram painting, parameter input, data base management, distribution network losses calculation, report output and so on. It has better performances in software visualization and friendly interface. Provided many shortcuts, which are according to the international standard styles, it is easy for users to study and operate. After the designing phrase, the software was tested for many times in order to make sure the integrality, security and dependability.
本文首先对现有的理论线损计算方法进行了深入分析,指出了目前各种线损计算方法的局限性。并针对我国配电网具有元器件数量众多、分布复杂,自动化程度普遍较低,原始数据不易收集等特点,提出了一种基于遗传算法的RBF神经网络用于配电网线损计算的新方法。该方法通过RBF神经网络的空间拟合性和径向基函数的局部响应特性映射配电线路参数与配电网线损之间的非线性复杂关系,并针对传统的RBF网络学习方法中,隐含层与输出层结构参数的确定相互独立,输出层权重训练容易陷入局部最小等缺点,应用遗传算法对整个RBF网络进行优化,将RBF网络不同的中心和其对应的宽度及各个调节权重统一编码,加强了RBF网络隐含层和输出层的合作关系,并利用遗传算法全局搜索的功能特性,使得整个网络模型达到全局最优。此外,对遗传算法本身的遗传机制作出了相应的改进,使遗传操作更加完善。
为了验证本文提出的方法的实用性和可行性,分别以某地区68条配电线路和天津滨海供电局67条配电线路的线路特征参数为样本,进行了线损的实例仿真计算。试验仿真结果表明以遗传算法优化的RBF网络,具有网络模型简单、训练速度快、计算精确度高等优点,并具有很强的实用性和推广性。利用神经网络的拟合能力合扩展能力,可以较为准确的记忆可获取的配电线路特征参数与线路线损之间的非线性映射关系,从而进行较为精确的线损计算。
最后,本文以Borland C++ Builder 5.0作为软件开发平台,基于面向对象程序设计的思想,开发出一套适应于配电网线损计算的可视化软件。软件具有配电线路图绘制,元件运行数据录入,数据库管理,线损计算,报表输出等功能。人机界面友好,并提供了大量的快捷操作,具有国际标准软件界面风格,易于用户学习掌握。软件设计后期,进行了多次测试和使用,确保了软件的完整性、安全性和可靠性。
The line losses of power distribution network, which had caught much attention of power enterprises and many related organizations, plays a very important role in evaluating the economics of a power system. Since the reform of the power industry, line losses has directly affected the power price, which is vital to the benefit of enterprises.Accurate and convenient methods of calculating the line losses is very important not only to the evaluation of line losses management, but also to the formulations of losses decreasing measurement.
Firstly existing methods for calculation of the theoretical line losses is investigated in detail in this paper, and the limitation of them is also pointed out. According to the characteristics of power distribution system in our country, this paper presents a RBFNN based on Genetic Algorithm, which is applied in calculating the line losses of power distribution systems. The radial basis fuction (RBF) neural network, due to its nonlinear processing ability, is used to map complex nonlinear relationship between the line losses and the feature factors in power distribution systems. But in the traditional methods, the parameters of the hidden layer and the output layer are trained separately. The Genetic Algorithm (GA), a global optimization algorithm simulating organic evolution, is used to optimize the parameters of the RBF network. The coding method put all the parameters in one chromosome optimize them synchronously. It can strengthen the cooperation between the hidden layer and the output layer, and optimizes the neural work to avoid the network parameters suffering into partial minimum.
In order to validate the practicability and feasibility of this method, this paper has two simulations respectively. A distribution network in some aera with 68 lines and the distribution network of Binhai power company with 67 lines are used as examples. The simulation recults show that the RBF neural network based on GA has so many advantages such as simple configuration, high convergence speed, high precision and so on. It can memory the complex nonlinear relationship between the line losses and the feature factors in distribution network accurately. So the method presented in this paper has good practicability and should be popularized in other areas.
At last, based on the OOP (Object Oriented Programming) ideology, this paper developed visual software for distribution network losses calculation on C++ Builder platform. The proposed software has many fuctions including that circuit diagram painting, parameter input, data base management, distribution network losses calculation, report output and so on. It has better performances in software visualization and friendly interface. Provided many shortcuts, which are according to the international standard styles, it is easy for users to study and operate. After the designing phrase, the software was tested for many times in order to make sure the integrality, security and dependability.
引文
[1] 吴安官,倪保珊,电力系统线损,北京:中国电力出版社,1996
[2] 虞忠年,陈星莺,刘吴,电力网电能损耗,北京:中国电力出版社,2000
[3] 杨秀台,电力网线损的理论计算和分析,北京:水利电力出版社,1985
[4] 许绍良,宋冶,苗竹梅等,电力网电能损耗计算导则,北京:中国电力出版社,2000
[5] 张利生,电力网电能损耗管理及降损技术,北京:中国电力出版社,2005
[6] 周文俊,电器设备实用手册,北京:中国水利水电出版社,1999
[7] 温步瀛,陈冲,改进最大电流法的配电网损耗计算,福建工程学院学报, 2004,2(1):13~15
[8] 陈海涵,程启诚,等值电阻法计算配电网损耗的理论和实践,广东电力,2004,17(3):5~8
[9] 周建英,宋奕冰,刘宪林,网络最大流法在配电网线损计算中的应用,电力系统及其自动化学报,2000,12(5):11~14
[10] 刘伟,李恒,张江等,基于馈线区间和GIS的配电网网损计算拓扑分析方法,继电器,2002,30(8):10~13
[11] 陈得治,郭志忠,基于负荷获取和匹配潮流方法的配电网理论线损计算,电网技术,2005,29(1):80~84
[12] 杨霖,王卫东,陈得治,郭志忠,基于测量的中压配电网线损计算研究,继电器,2005,33(3):28~33
[13] Hsu C T,Tzeng Y M,Chen C Z et al. Distribution feeder loss analysis by using an artificial neural network.Electric Power Systems Research,1995,34(2):85~90
[14] 王刚军,王承民,李恒等,基于实测数据的配网理论网损计算方法,电网技术,2002,26(12):18~20
[15] 张铁峰,苑津莎,刘建新,计及多种因素的配电网理论线损使用计算方法,华北电力大学学报,2005,32(3):5~7,31
[16] 许汉平,侯进峰,基于状态估计数据的电网线损理论计算方法,电网技术,2003,27(3):59~62
[17] 丁心海,罗毅方,刘巍等,配电网线损理论计算的实用方法——改进迭代法.电网技术,2000,24(1):39~42
[18] 马瑾,徐建源,王胜辉,林莘,辐射状配电网的潮流计算,沈阳工业大学学报,2004,26(1):37~40
[19] 张步涵,李可文,沙立华等,基于树状网潮流的配电网电能损耗计算,华中理工大学学报,2000,28(2):58~60
[20] 王成山,王守相,基于区间算法的配电网三相潮流计算及算例分析,中国电机工程学报,2002,22(3):58~62
[21] 孙宏斌,张伯明,相年德,配电匹配潮流技术及其在配电网状态估计中的应用,电力系统自动化,1998,22(7):18~22
[22] 李红梅,齐伟夫,徐建源,改进电量法在配电网线损计算中的应用,东北电力技术,2004,10(2):14~19
[23] 蔡君豪,人工神经网络用于配电网的线损计算,辽宁电机工程学报,1995,13(4):47~50
[24] 文福拴,韩祯祥,基于分群算法和人工神经元网络的配电网线损计算,中国电机工程学报,1993,13(3):41~50
[25] 辛开远,杨玉华,陈富.计算配电网线损的GA与BP结合的新方法,中国电机工程学报,2002,22(2): 79~82
[26] 袁慧梅,郭喜庆,用遗传算法辅助设计的人工神经网络计算配电网线损,电网技术,1998,22(12):17~19
[27] 陈 天 恩 , 于 鹏 , 用 函 数 型 神 经 网 络 计 算 配 网 的 线 损 , 甘 肃 电 力 , 1996,(1):40~41
[28] 张超,刘宪林,集成神经网络在配网线损分析中的应用,继电器,2002,30(7):3~6
[29] 蒋宗礼,人工神经网络导论,北京:高等教育出版社,2001
[30] 陈祥光,裴旭东,人工神经网络技术及应用,北京:中国电力出版社,2003
[31] 王树谦,魏德华,胡浩云等,多层前向人工神经网络,郑州:黄河水利出版社,1999
[32] 袁曾任,人工神经元网络及其应用,北京:清华大学出版社,1999
[33] J.Barry Gomm,Ding Li Yu.”Selecting radial basis function network centers with recursive orthogonal squares training”,IEEE Trans.Neural Networks.2000.11(2),pp:306~314
[34] Chen S et al.”Orthogonal least square learning algorithm for radial basis function networks”.IEEE Trans.Neural Networks.1991.2:302~309
[35] Y. Abe, Y. Figuni. Fast computation of RBF coefficients for regularly sampled inputs. IEE Electronics Letters, 2003, 39(6): 543~544
[36] 姜静清,宋初一,刘娜仁等, RBF神经网络的训练方法及分析,内蒙古民族大学学报(自然科学版),2003,18(4):301~303
[37] 朱明星,张德龙,RBF网络基函数中心选取算法的研究,安徽大学学报(自然科学版),2000, 24(1):72~78
[38] 李波,王成友等,用最陡下降法修正RBF隐层参数,空军工程大学学报(自然科学版),2004,5(4):67~69
[39] 周明,孙树栋,遗传算法原理及应用,北京:国防工业出版社,1996
[40] 陆金桂,李谋,王浩,遗传算法原理及其工程应用,北京:中国矿业大学出版社,1997
[41] 李敏强,寇纪淞,林丹,遗传算法的基本原理与应用,北京:科学出版社,2002
[42] 彭慧敏,赵菁,谢维廉,基于遗传算法的最优潮流,贵州工业大学学报,2003,32,(1):45~48
[43] 范宏,韦化,改进遗传算法在无功优化中的应用,电力系统及其自动化学报,2005,17(1):6~9
[44] 李财莲,贾永兴,岳振军,基于自适应遗传算法和BP网络的物重监测模型,系统工程与电子技术,2005,27(2):377~380
[45] 余健明,吴海峰,杨文宇,基于改进多种群遗传算法的配电网规划,电网技术,2005,29(7):36~40
[46] 唐伟,李群湛,余丹,基于改进遗传算法的电力系统机组优化组合,电气应用,2006,25(1):64~66
[47] 余明兴,吴明哲,Borland C++ Builder 5.0 实例精解,北京:清华大学出版社,2001
[48] 俞涓,宋丽红,Borland C++ Builder 5.0 实例教程,上海:浦东电子出版社,2000
[49] 袁辉,郝文宁,Borland C++ Builder 高级编程,北京:科学出版社,2000
[50] 刘光,C++ Builder 数据库系统设计与开发,北京:清华大学出版社,2003
[51] 杨宗志,C++ Builder 数据库程序设计,北京:清华大学出版社,2001
[2] 虞忠年,陈星莺,刘吴,电力网电能损耗,北京:中国电力出版社,2000
[3] 杨秀台,电力网线损的理论计算和分析,北京:水利电力出版社,1985
[4] 许绍良,宋冶,苗竹梅等,电力网电能损耗计算导则,北京:中国电力出版社,2000
[5] 张利生,电力网电能损耗管理及降损技术,北京:中国电力出版社,2005
[6] 周文俊,电器设备实用手册,北京:中国水利水电出版社,1999
[7] 温步瀛,陈冲,改进最大电流法的配电网损耗计算,福建工程学院学报, 2004,2(1):13~15
[8] 陈海涵,程启诚,等值电阻法计算配电网损耗的理论和实践,广东电力,2004,17(3):5~8
[9] 周建英,宋奕冰,刘宪林,网络最大流法在配电网线损计算中的应用,电力系统及其自动化学报,2000,12(5):11~14
[10] 刘伟,李恒,张江等,基于馈线区间和GIS的配电网网损计算拓扑分析方法,继电器,2002,30(8):10~13
[11] 陈得治,郭志忠,基于负荷获取和匹配潮流方法的配电网理论线损计算,电网技术,2005,29(1):80~84
[12] 杨霖,王卫东,陈得治,郭志忠,基于测量的中压配电网线损计算研究,继电器,2005,33(3):28~33
[13] Hsu C T,Tzeng Y M,Chen C Z et al. Distribution feeder loss analysis by using an artificial neural network.Electric Power Systems Research,1995,34(2):85~90
[14] 王刚军,王承民,李恒等,基于实测数据的配网理论网损计算方法,电网技术,2002,26(12):18~20
[15] 张铁峰,苑津莎,刘建新,计及多种因素的配电网理论线损使用计算方法,华北电力大学学报,2005,32(3):5~7,31
[16] 许汉平,侯进峰,基于状态估计数据的电网线损理论计算方法,电网技术,2003,27(3):59~62
[17] 丁心海,罗毅方,刘巍等,配电网线损理论计算的实用方法——改进迭代法.电网技术,2000,24(1):39~42
[18] 马瑾,徐建源,王胜辉,林莘,辐射状配电网的潮流计算,沈阳工业大学学报,2004,26(1):37~40
[19] 张步涵,李可文,沙立华等,基于树状网潮流的配电网电能损耗计算,华中理工大学学报,2000,28(2):58~60
[20] 王成山,王守相,基于区间算法的配电网三相潮流计算及算例分析,中国电机工程学报,2002,22(3):58~62
[21] 孙宏斌,张伯明,相年德,配电匹配潮流技术及其在配电网状态估计中的应用,电力系统自动化,1998,22(7):18~22
[22] 李红梅,齐伟夫,徐建源,改进电量法在配电网线损计算中的应用,东北电力技术,2004,10(2):14~19
[23] 蔡君豪,人工神经网络用于配电网的线损计算,辽宁电机工程学报,1995,13(4):47~50
[24] 文福拴,韩祯祥,基于分群算法和人工神经元网络的配电网线损计算,中国电机工程学报,1993,13(3):41~50
[25] 辛开远,杨玉华,陈富.计算配电网线损的GA与BP结合的新方法,中国电机工程学报,2002,22(2): 79~82
[26] 袁慧梅,郭喜庆,用遗传算法辅助设计的人工神经网络计算配电网线损,电网技术,1998,22(12):17~19
[27] 陈 天 恩 , 于 鹏 , 用 函 数 型 神 经 网 络 计 算 配 网 的 线 损 , 甘 肃 电 力 , 1996,(1):40~41
[28] 张超,刘宪林,集成神经网络在配网线损分析中的应用,继电器,2002,30(7):3~6
[29] 蒋宗礼,人工神经网络导论,北京:高等教育出版社,2001
[30] 陈祥光,裴旭东,人工神经网络技术及应用,北京:中国电力出版社,2003
[31] 王树谦,魏德华,胡浩云等,多层前向人工神经网络,郑州:黄河水利出版社,1999
[32] 袁曾任,人工神经元网络及其应用,北京:清华大学出版社,1999
[33] J.Barry Gomm,Ding Li Yu.”Selecting radial basis function network centers with recursive orthogonal squares training”,IEEE Trans.Neural Networks.2000.11(2),pp:306~314
[34] Chen S et al.”Orthogonal least square learning algorithm for radial basis function networks”.IEEE Trans.Neural Networks.1991.2:302~309
[35] Y. Abe, Y. Figuni. Fast computation of RBF coefficients for regularly sampled inputs. IEE Electronics Letters, 2003, 39(6): 543~544
[36] 姜静清,宋初一,刘娜仁等, RBF神经网络的训练方法及分析,内蒙古民族大学学报(自然科学版),2003,18(4):301~303
[37] 朱明星,张德龙,RBF网络基函数中心选取算法的研究,安徽大学学报(自然科学版),2000, 24(1):72~78
[38] 李波,王成友等,用最陡下降法修正RBF隐层参数,空军工程大学学报(自然科学版),2004,5(4):67~69
[39] 周明,孙树栋,遗传算法原理及应用,北京:国防工业出版社,1996
[40] 陆金桂,李谋,王浩,遗传算法原理及其工程应用,北京:中国矿业大学出版社,1997
[41] 李敏强,寇纪淞,林丹,遗传算法的基本原理与应用,北京:科学出版社,2002
[42] 彭慧敏,赵菁,谢维廉,基于遗传算法的最优潮流,贵州工业大学学报,2003,32,(1):45~48
[43] 范宏,韦化,改进遗传算法在无功优化中的应用,电力系统及其自动化学报,2005,17(1):6~9
[44] 李财莲,贾永兴,岳振军,基于自适应遗传算法和BP网络的物重监测模型,系统工程与电子技术,2005,27(2):377~380
[45] 余健明,吴海峰,杨文宇,基于改进多种群遗传算法的配电网规划,电网技术,2005,29(7):36~40
[46] 唐伟,李群湛,余丹,基于改进遗传算法的电力系统机组优化组合,电气应用,2006,25(1):64~66
[47] 余明兴,吴明哲,Borland C++ Builder 5.0 实例精解,北京:清华大学出版社,2001
[48] 俞涓,宋丽红,Borland C++ Builder 5.0 实例教程,上海:浦东电子出版社,2000
[49] 袁辉,郝文宁,Borland C++ Builder 高级编程,北京:科学出版社,2000
[50] 刘光,C++ Builder 数据库系统设计与开发,北京:清华大学出版社,2003
[51] 杨宗志,C++ Builder 数据库程序设计,北京:清华大学出版社,2001