基于神经元网络的SFC注入束运线自动调束控制系统研究
|
摘要
兰州重离子加速器的调束是基于人工手动调试的方法,这种方式十分繁杂,不仅需要大量的人力物力,而且束流的稳定性还很难得到有效的控制。为了解决此问题,本文结合SFC注入束运线,提出了一种基于神经元网络的自动调束控制方案,为该方案设计了两个硬件平台。
根据兰州重离子加速器的调束特点,将HIRFL-CSR自动调束控制系统设计为两层：智能调束决策层和电源调节层。上层调束决策层主要通过采集和分析束流诊断元件反馈回来的束流参数,判断束流信息,并给定下层电源的参数。下层电源调节层的主要作用是接收到上层对电源参数的给定,并通过优化算法来调节电源参数,实现对电源的自动调节。为了满足通信的需求,本文设计了.NET工业控制计算机,采用的是智能网络交换机工业计算机的方式,集网络交换和工业控制于一体。微型工业可编程自动化控制器(μPAC)是为电源调节层设计的,采用了FPGA+ARM的结构,ARM用来作为核心处理器,FPGA用来实现电源的控制算法。
为了解决电源控制节点之间的通讯问题,本文将P2P技术嵌入到μPAC的核心控制器ARM中,运用UDP的打洞技术,使节点之间的数据不再通过中心Oracle数据库转发,而是在控制节点之间直接建立通讯,这种方式分散了中心Oracle数据库的通讯量,降低了网路堵塞的风险,提高了通讯的效率、有效性和可靠性。同时,为了满足自动调束控制系统实时性强和并行处理的算法需求,对基于FPGA的BP算法进行了研究和优化,并利用MATLAB中的Simulink工具进行了仿真。
The traditional beam tuning method for HIRFL-CSR which is based on manual debugging is very complicated, and this method not only needs lots of manpower and material resources but also difficult to achieve stable control aim. To resolve this problem, a neural network-based beam automatic modulating control system is proposed and two hardware platforms are designed for SFC injection beam line in this thesis.
According the beam tuning characteristics, the beam automatic modulating control system of HIRFL-CSR is designed into two layers:the intelligent beam tuning decision layer and power supply adjustment layer. The intelligent beam tuning decision layer collects and analyses the parameters from the feedback of beam diagnostic elements, then gives the power supply parameters. The main role of power supply adjustment layer is to receive the power parameters from the intelligent beam tuning decision layer, and then adjust the parameters by using optimization algorithm. The.net industrial control computer is designed for communication between the two layers, its function is network switching and industrial control, and has the advantages of intelligent network switches and industrial computer. The μPAC is designed for power adjustment layer by using the structure of FPGA+ARM. ARM is designed as the core processor, and FPGA is designed for power digital adjust algorithm.
In order to solve the communications blockage problem of the power supply controller in Lanzhou Heavy Ion Accelerator CSR, a new system architecture is adopted, P2P technology is embedded to the core processor ARM of the μPAC. Based on the technology of UDP holing, a direct communication connect is established between the controllers, so the data no longer forwarded by the main Oracle database system acting as a server. This method decentralized the communication responsibility of Oracle database system, the risk of network block is reduced, and the effectiveness and reliability of communication is improved. At the same time, beam automatic modulating control system has the need of strong real-time and parallel processing, so BP neural network is optimized in the FPGA, and the control system is simulated by the tools boxs of Simlulink in MATLAB.
根据兰州重离子加速器的调束特点,将HIRFL-CSR自动调束控制系统设计为两层：智能调束决策层和电源调节层。上层调束决策层主要通过采集和分析束流诊断元件反馈回来的束流参数,判断束流信息,并给定下层电源的参数。下层电源调节层的主要作用是接收到上层对电源参数的给定,并通过优化算法来调节电源参数,实现对电源的自动调节。为了满足通信的需求,本文设计了.NET工业控制计算机,采用的是智能网络交换机工业计算机的方式,集网络交换和工业控制于一体。微型工业可编程自动化控制器(μPAC)是为电源调节层设计的,采用了FPGA+ARM的结构,ARM用来作为核心处理器,FPGA用来实现电源的控制算法。
为了解决电源控制节点之间的通讯问题,本文将P2P技术嵌入到μPAC的核心控制器ARM中,运用UDP的打洞技术,使节点之间的数据不再通过中心Oracle数据库转发,而是在控制节点之间直接建立通讯,这种方式分散了中心Oracle数据库的通讯量,降低了网路堵塞的风险,提高了通讯的效率、有效性和可靠性。同时,为了满足自动调束控制系统实时性强和并行处理的算法需求,对基于FPGA的BP算法进行了研究和优化,并利用MATLAB中的Simulink工具进行了仿真。
The traditional beam tuning method for HIRFL-CSR which is based on manual debugging is very complicated, and this method not only needs lots of manpower and material resources but also difficult to achieve stable control aim. To resolve this problem, a neural network-based beam automatic modulating control system is proposed and two hardware platforms are designed for SFC injection beam line in this thesis.
According the beam tuning characteristics, the beam automatic modulating control system of HIRFL-CSR is designed into two layers:the intelligent beam tuning decision layer and power supply adjustment layer. The intelligent beam tuning decision layer collects and analyses the parameters from the feedback of beam diagnostic elements, then gives the power supply parameters. The main role of power supply adjustment layer is to receive the power parameters from the intelligent beam tuning decision layer, and then adjust the parameters by using optimization algorithm. The.net industrial control computer is designed for communication between the two layers, its function is network switching and industrial control, and has the advantages of intelligent network switches and industrial computer. The μPAC is designed for power adjustment layer by using the structure of FPGA+ARM. ARM is designed as the core processor, and FPGA is designed for power digital adjust algorithm.
In order to solve the communications blockage problem of the power supply controller in Lanzhou Heavy Ion Accelerator CSR, a new system architecture is adopted, P2P technology is embedded to the core processor ARM of the μPAC. Based on the technology of UDP holing, a direct communication connect is established between the controllers, so the data no longer forwarded by the main Oracle database system acting as a server. This method decentralized the communication responsibility of Oracle database system, the risk of network block is reduced, and the effectiveness and reliability of communication is improved. At the same time, beam automatic modulating control system has the need of strong real-time and parallel processing, so BP neural network is optimized in the FPGA, and the control system is simulated by the tools boxs of Simlulink in MATLAB.
引文
[1]夏佳文,詹文龙等.兰州重离子加速器冷却储存环[J].强激光与离子束.2008,20(11)：1183-1190.
[2]Xia Jiawen, Zhan Wenlong, Wei Baowen, et al. The heavy ion cooler-storage ring project(HIRFL-CSR) at Lanzhou [J]. Nuclear Insruments and Methods in Physics Research Section A.2002,488:11-25.
[3]尹全民,宋明涛,张金泉,赵红卫.SFC的新的轴向注入束流线[J].高能物理与核物理.2006,30(2)：117-119.
[4]唐靖宇,魏宝文.回旋加速器理论与设计[M].合肥：中国科学技术大学出版社.2008.
[5]张金泉.新的SFC轴向注入系统及SFC-CSR匹配的设计研究[硕士论文].兰州：中国科学院近代物理研究所.2003.
[6]蔡自兴.智能控制原理与应用[M].北京：清华大学出版社.2007.
[7]阎平凡,张长水.人工神经网络与模拟进化计算[M].北京：清华大学出版社.2000.
[8]Simon Haykin. Neural Networks:A Comprehensive Foundation, Second Edition[M]北京：清华大学出版社.2001.
[9]刘金馄.智能控制[M].北京：电子工业出版社.2005.
[10]盛荣菊,马建伟.人工神经网络FPGA硬件实现的研究进展[J].电气自动化.2009,31(5)：53-55.
[11]祝志文,吴东.一种基于FPGA的神经网络的实现[J].无线电工程.2000,30(5)：56-59.
[12]郝智泉,王贞松.BP算法的脉动阵列结构在FPGA上的实现[J].计算机工程.2006(21).
[13]Rafael Gade, Joaquln Cerdo, Franeiso Ballester, Antonio Moeholi. Artificial Neural Network Implementation on a single FPGA of a Pipelined on-line Back Propagation[J]. IEEE 13th International Symposium on System Synthesis (ISSS'00), (2000), p225-230.
[14]何旭春CSRe冷却装置磁铁电源系统的神经元网络控制设计[博士论文].兰州：中国科学院近代物理研究所.2009.
[15]焦喜香,敬岚,龙银东等.用于重离子加速器磁铁电源的数字调节器设计[J].强激光与粒子束.2011,23(1)：185-189.
[16]Analog Devices, Inc. CMOS 300 MHz Quadrature Complete-DDS,1999.
[17]TMS320C6000 DSP Programmer'Guide.Texas Instruments,2003.
[18]CIRRUS LOGIC Corporation Cirrus logic EP9315 data sheet.2004.
[19]Altera Corporation. Cyclone Ⅲ Device Handbook.2004.
[20]高旭Windows CE嵌入式操作系统下的CANopen主站实现方案及其应用[硕士论文].天津：河北工业大学.2007.
[21]刘太联.重离子治癌病灶定位嵌入式控制系统研究[博士论文].兰州：中国科学院近代物理研究所.2012.
[22]LEADTOOLS.LEADTOOLS Raster Imaging C++ Class Library Help.2011.
[23]薛大龙,陈世帝,王韵Windows CE嵌入式系统开发从基础到实践[M].北京：电子工业出版社,2008.
[24]孙天泽,袁文菊,张海峰编著.嵌入式设计及Linux驱动开发指南—基于ARM9处理器 [M].北京：电子工业出版社,2005.
[25]武安河Windows设备驱动程序WDF开发[M].北京：电子工业出版社,2009.
[26]杨阿锋.基于WDF的PCIe接口高速数据传输卡的驱动程序开发[硕十论文].长沙：国防科学技术大学.2008.
[27]张帆,史彩成Windows驱动开发技术详解[M].北京：电子工业出版社,2008.
[28]封岸松,陈谷,江滢Windows CE平台下排队机呼叫系统的移植研究[J].现代电子技术.2012,313(2)：28-30.
[29]周建设Windows CE设备驱动及BSP开发指南[M].北京：中国电力出版社.2009.
[30]潘磊磊,王勇.基于WinCE的USB设备流接口驱动研究与实现[J].仪器仪表用户,2010.17(1)：69-70.
[31]王黎明,陈双桥,闫晓玲.ARM9嵌入式系统开发与实践[M].北京：北京航空航天大学出版社,2008.
[32]徐杨,乔卫民,刘伍丰等HIRFL-CSR嵌入式数据库的设计实现[J].核电子学与探测器技术.2008,28(3)：590-592.
[33]曾贤强,敬岚,姚泽恩,郝焕锋,苟世哲.分离扇回旋加速器的等时场在线优化[J].强激光与粒子束.2012,24(6)：1458-1460.
[34]肖文君.重离子肿瘤治疗医学图像智能终端系统研究[博十论文].兰州：中国科学院近代物理研究所.2012.
[35]Takeshi Katayama, Morio Yoshizawa, shin-ichi Watanabe. RF Acceleration System at Cooler-Synchrotron [J]. Nuclear Instruments and Methods in Physics Research, July 1993.
[36]Zhang Wenzhi, Xu Zhe,Yang Xiaodong. RF accelerating system at HIRFL-CSR main ring[C]. Proceedings of the 2003 Particle Accelerator Conference. Portland,2003:1288-1290.
[37]刘伍丰.CSR控制系统的大规模多级分布式数据交互系统[博十论文].兰州：中国科学院近代物理研究所.2009.
[38]肖文君,敬岚,乔卫民.重离子治癌电源调节器的嵌入式系统[J].核电子学与探测技术,2012,32(2)：168-170.
[39]管磊.P2P技术揭秘-P2P网络技术原理与典型系统开发[M].北京：清华大学出版社.2011.
[40]Liang J, Kumar R, Xi Y, et al. Pollution in p2p file sharing systems[C]//IEEE Infocom, Miami, FL, USA,2005:1174-1185.
[41]黄佳庆,刘剑锋,程刚,荣义吕.大规模并行分布式P2P算法模拟器的设计与实现[J].华中科技大学学报(自然科学版).2012,40(2)：91-96.
[42]齐卫宁,王劲林.一种嵌入式P2P文件共享系统.计算机工程与应用[J].2008,44(15)：76-78.
[43]孙海民Windows Sockets网络开发—基于Visual C++实现[M].北京：人民邮电出版社,2008.
[44]汗兵等Windows CE嵌入式高级编程及其实例详解[M].北京：中国水利水电出版社,2008.
[45]焦李成.神经网络系统理论[M].西安：西安电子科技大学出版社.1990.
[46]胡守仁主编.神经网络应用技术[M].长沙：国防科技大学出版社,1993.
[47]张洁琦.人工神经网络控制在电力系统中的应用研究[硕士论文].济南：山东师范大学.2007.
[48]海金(Hayjin S)著,叶世民,史忠植译.神经网络原理[M].北京：机械工业出版社.2004.
[49]胡德文,王正志,王耀南.神经网络自适应控制[M].长沙：国防科技大学出版社.2006.
[50]董长虹.MATLAB神经网络与应用[M].北京：国防工业出版社.2005.
[51]M.Zak. Terminal Attractors For Addressable Memory in Neural Networks[J].Physics Letter A.1988,133(12).18-22.
[52]P. FBaldi, K. Hornik. Learning in Linear Neural Networks:A Survey[C]. IEEE Trans, on Neutral Networks.1995,6(4),837-857.
[53]D.J. Montana, L. Davis. Training Feed forward Network Using Genetic Algorithm[C]. Int.J. Conf. on AI,1989:762-767.
[54]M.R. Azimi-Sadjadi, R. J. Lion. Fast Learning Process of Multilayer Neural Networks Using Recursive Least Squares Method[C]. IEEE Trans, on SP,1992:40(3),446-450.
[55]周开利,康耀红.神经网络模型及其MATLAB仿真程序设计[M].北京：清华大学出版社,2004.
[56]韩少锋.神经网络控制的FPGA研究与应用[硕十论文].大连：大连海事大学.2005.
[57]加卢什会著,阎平凡译.神经网络理论[M].北京：清华大学出版社,2007.
[58]王保存.基于人工神经网络的水质监测拉曼光谱系统的研究[硕十论文].武汉：华中师范大学.2007.
[59]马建国,孟宪元FPGA现代数字系统设计[M].北京：清华大学出版社,2010.
[60]侯伯亨,顾新.VHDL硬件描述语言与数字逻辑电路设计[M].西安：西安电子科技大学出版社.1999.
[61]沈明发,易清明,黄伟英,周伟贤.用VHDL语言在CPLD/FPAG上实现浮点运算[J].暨南大学学报(自然科学版).2002,23(5)：19-24.
[62]Wolf, D.F., Romero, R.A.F., Marques, E. Using Embedded Processors in Hardware Models of Artificial Neural Networks[J]. In Proceedings of SBAI-Simposio Brasileiro de Automao Inteligente.2001.78-83.
[63]祝志文,吴东.一种基于FPGA的神经网络的实现[J].无线电工程.2000,30(5)：56-59.
[64]YJ Chen and WP du Plessis, Neural Network Implementation on a FPGA[C]. In Proceedings of the IEEE African 2002 Conference, IEEE,337-342.
[2]Xia Jiawen, Zhan Wenlong, Wei Baowen, et al. The heavy ion cooler-storage ring project(HIRFL-CSR) at Lanzhou [J]. Nuclear Insruments and Methods in Physics Research Section A.2002,488:11-25.
[3]尹全民,宋明涛,张金泉,赵红卫.SFC的新的轴向注入束流线[J].高能物理与核物理.2006,30(2)：117-119.
[4]唐靖宇,魏宝文.回旋加速器理论与设计[M].合肥：中国科学技术大学出版社.2008.
[5]张金泉.新的SFC轴向注入系统及SFC-CSR匹配的设计研究[硕士论文].兰州：中国科学院近代物理研究所.2003.
[6]蔡自兴.智能控制原理与应用[M].北京：清华大学出版社.2007.
[7]阎平凡,张长水.人工神经网络与模拟进化计算[M].北京：清华大学出版社.2000.
[8]Simon Haykin. Neural Networks:A Comprehensive Foundation, Second Edition[M]北京：清华大学出版社.2001.
[9]刘金馄.智能控制[M].北京：电子工业出版社.2005.
[10]盛荣菊,马建伟.人工神经网络FPGA硬件实现的研究进展[J].电气自动化.2009,31(5)：53-55.
[11]祝志文,吴东.一种基于FPGA的神经网络的实现[J].无线电工程.2000,30(5)：56-59.
[12]郝智泉,王贞松.BP算法的脉动阵列结构在FPGA上的实现[J].计算机工程.2006(21).
[13]Rafael Gade, Joaquln Cerdo, Franeiso Ballester, Antonio Moeholi. Artificial Neural Network Implementation on a single FPGA of a Pipelined on-line Back Propagation[J]. IEEE 13th International Symposium on System Synthesis (ISSS'00), (2000), p225-230.
[14]何旭春CSRe冷却装置磁铁电源系统的神经元网络控制设计[博士论文].兰州：中国科学院近代物理研究所.2009.
[15]焦喜香,敬岚,龙银东等.用于重离子加速器磁铁电源的数字调节器设计[J].强激光与粒子束.2011,23(1)：185-189.
[16]Analog Devices, Inc. CMOS 300 MHz Quadrature Complete-DDS,1999.
[17]TMS320C6000 DSP Programmer'Guide.Texas Instruments,2003.
[18]CIRRUS LOGIC Corporation Cirrus logic EP9315 data sheet.2004.
[19]Altera Corporation. Cyclone Ⅲ Device Handbook.2004.
[20]高旭Windows CE嵌入式操作系统下的CANopen主站实现方案及其应用[硕士论文].天津：河北工业大学.2007.
[21]刘太联.重离子治癌病灶定位嵌入式控制系统研究[博士论文].兰州：中国科学院近代物理研究所.2012.
[22]LEADTOOLS.LEADTOOLS Raster Imaging C++ Class Library Help.2011.
[23]薛大龙,陈世帝,王韵Windows CE嵌入式系统开发从基础到实践[M].北京：电子工业出版社,2008.
[24]孙天泽,袁文菊,张海峰编著.嵌入式设计及Linux驱动开发指南—基于ARM9处理器 [M].北京：电子工业出版社,2005.
[25]武安河Windows设备驱动程序WDF开发[M].北京：电子工业出版社,2009.
[26]杨阿锋.基于WDF的PCIe接口高速数据传输卡的驱动程序开发[硕十论文].长沙：国防科学技术大学.2008.
[27]张帆,史彩成Windows驱动开发技术详解[M].北京：电子工业出版社,2008.
[28]封岸松,陈谷,江滢Windows CE平台下排队机呼叫系统的移植研究[J].现代电子技术.2012,313(2)：28-30.
[29]周建设Windows CE设备驱动及BSP开发指南[M].北京：中国电力出版社.2009.
[30]潘磊磊,王勇.基于WinCE的USB设备流接口驱动研究与实现[J].仪器仪表用户,2010.17(1)：69-70.
[31]王黎明,陈双桥,闫晓玲.ARM9嵌入式系统开发与实践[M].北京：北京航空航天大学出版社,2008.
[32]徐杨,乔卫民,刘伍丰等HIRFL-CSR嵌入式数据库的设计实现[J].核电子学与探测器技术.2008,28(3)：590-592.
[33]曾贤强,敬岚,姚泽恩,郝焕锋,苟世哲.分离扇回旋加速器的等时场在线优化[J].强激光与粒子束.2012,24(6)：1458-1460.
[34]肖文君.重离子肿瘤治疗医学图像智能终端系统研究[博十论文].兰州：中国科学院近代物理研究所.2012.
[35]Takeshi Katayama, Morio Yoshizawa, shin-ichi Watanabe. RF Acceleration System at Cooler-Synchrotron [J]. Nuclear Instruments and Methods in Physics Research, July 1993.
[36]Zhang Wenzhi, Xu Zhe,Yang Xiaodong. RF accelerating system at HIRFL-CSR main ring[C]. Proceedings of the 2003 Particle Accelerator Conference. Portland,2003:1288-1290.
[37]刘伍丰.CSR控制系统的大规模多级分布式数据交互系统[博十论文].兰州：中国科学院近代物理研究所.2009.
[38]肖文君,敬岚,乔卫民.重离子治癌电源调节器的嵌入式系统[J].核电子学与探测技术,2012,32(2)：168-170.
[39]管磊.P2P技术揭秘-P2P网络技术原理与典型系统开发[M].北京：清华大学出版社.2011.
[40]Liang J, Kumar R, Xi Y, et al. Pollution in p2p file sharing systems[C]//IEEE Infocom, Miami, FL, USA,2005:1174-1185.
[41]黄佳庆,刘剑锋,程刚,荣义吕.大规模并行分布式P2P算法模拟器的设计与实现[J].华中科技大学学报(自然科学版).2012,40(2)：91-96.
[42]齐卫宁,王劲林.一种嵌入式P2P文件共享系统.计算机工程与应用[J].2008,44(15)：76-78.
[43]孙海民Windows Sockets网络开发—基于Visual C++实现[M].北京：人民邮电出版社,2008.
[44]汗兵等Windows CE嵌入式高级编程及其实例详解[M].北京：中国水利水电出版社,2008.
[45]焦李成.神经网络系统理论[M].西安：西安电子科技大学出版社.1990.
[46]胡守仁主编.神经网络应用技术[M].长沙：国防科技大学出版社,1993.
[47]张洁琦.人工神经网络控制在电力系统中的应用研究[硕士论文].济南：山东师范大学.2007.
[48]海金(Hayjin S)著,叶世民,史忠植译.神经网络原理[M].北京：机械工业出版社.2004.
[49]胡德文,王正志,王耀南.神经网络自适应控制[M].长沙：国防科技大学出版社.2006.
[50]董长虹.MATLAB神经网络与应用[M].北京：国防工业出版社.2005.
[51]M.Zak. Terminal Attractors For Addressable Memory in Neural Networks[J].Physics Letter A.1988,133(12).18-22.
[52]P. FBaldi, K. Hornik. Learning in Linear Neural Networks:A Survey[C]. IEEE Trans, on Neutral Networks.1995,6(4),837-857.
[53]D.J. Montana, L. Davis. Training Feed forward Network Using Genetic Algorithm[C]. Int.J. Conf. on AI,1989:762-767.
[54]M.R. Azimi-Sadjadi, R. J. Lion. Fast Learning Process of Multilayer Neural Networks Using Recursive Least Squares Method[C]. IEEE Trans, on SP,1992:40(3),446-450.
[55]周开利,康耀红.神经网络模型及其MATLAB仿真程序设计[M].北京：清华大学出版社,2004.
[56]韩少锋.神经网络控制的FPGA研究与应用[硕十论文].大连：大连海事大学.2005.
[57]加卢什会著,阎平凡译.神经网络理论[M].北京：清华大学出版社,2007.
[58]王保存.基于人工神经网络的水质监测拉曼光谱系统的研究[硕十论文].武汉：华中师范大学.2007.
[59]马建国,孟宪元FPGA现代数字系统设计[M].北京：清华大学出版社,2010.
[60]侯伯亨,顾新.VHDL硬件描述语言与数字逻辑电路设计[M].西安：西安电子科技大学出版社.1999.
[61]沈明发,易清明,黄伟英,周伟贤.用VHDL语言在CPLD/FPAG上实现浮点运算[J].暨南大学学报(自然科学版).2002,23(5)：19-24.
[62]Wolf, D.F., Romero, R.A.F., Marques, E. Using Embedded Processors in Hardware Models of Artificial Neural Networks[J]. In Proceedings of SBAI-Simposio Brasileiro de Automao Inteligente.2001.78-83.
[63]祝志文,吴东.一种基于FPGA的神经网络的实现[J].无线电工程.2000,30(5)：56-59.
[64]YJ Chen and WP du Plessis, Neural Network Implementation on a FPGA[C]. In Proceedings of the IEEE African 2002 Conference, IEEE,337-342.