人工神经网络在泥沙运动中的应用与研究
|
摘要
近些年来，随着经济的发展和对大自然无休止的掠夺，我国大江大河流域内水土流失严重，造成大量的泥沙输入河道，加剧了江河整治的复杂性并带来严重的灾害。为了真正落实科学发展观与建设和谐社会，对于泥沙深入的研究迫在眉睫。虽然泥沙运动力学已形成了一些比较成熟的理论，但是至今仍有许多问题尚未得到解决。
本文首先对泥沙的起动和推移质研究成果做简要的回顾性总结，并在此基础上详细阐述了泥沙起动问题和推移质问题联系的纽带——推移质最大起动粒径研究的意义以及就目前研究水平所面临的问题；与此同时，人工神经网络理论则模仿人脑的思维，做出判断和预测，它可以借助有限的实测资料获取经验，揭示输入资料潜在的效应和变化趋势，无需用户建立函数关系，对复杂的非线性系统尤为有效。其次，选择目前采用最为广泛的BP神经网络模型、通过VC#．NET开发平台和SQL2000数据库并结合平衡输沙状态下的水槽输沙试验数据，开发出界面友好、基于BP及其改进算法的多输入单输出三层神经网络预测软件。针对BP网络算法所固有的缺点，提出并实现的两种改进算法分别为自适应调整学习率算法和模拟退火算法。再次，利用作者开发的带有自主知识产权的预测软件和Matlab神经网络工具箱，对实验数据做计算、预测和分析；与此同时，针对BP算法的转换函数中参数t对网络模型输出精度和稳定性的影响展开探索性的研究，并在大量实验数据的基础上提出了参数t的取值公式与适用范围。最后，通过对基于经典BP网络算法、自适应调整学习率改进算法、模拟退火改进算法预测结果与Matlab神经网络工具箱预测结果之间科学详尽的横向与纵向的数据比较与稳定性分析得出本文的结论。
本文的结论为：对于平衡状态下的宽级配非均匀沙水槽输沙试验中最大起动粒径的推求，利用VC#．NET开发出界面友好、基于BP及其改进算法的多输入单输出三层神经网络预测软件，其计算精度满足工程要求，特别对模拟退火改进算法，其计算精度比成熟的Matlab工具箱神经网络工具箱计算精度还要高，显示出光明的应用前景与商业价值，对于泥沙起动和推移质的研究有一定的辅助作用和参考价值。
Recent years, along with the development of economy and the endless depredation of nature, the phenomenon of water and soil loss in our country is more and more serious, which causes much more sediment into the rivers .That arouses the complexity of management to the rivers and many more severe disasters. For the sake of scientific development outlook and a harmonious society, deeper in investigation of sediment is imperatively. But, unfortunately, the problems about sediment science have not been solved completely until now.
The main contents are as follows: First of all, on the basis of comprehensive review of the study achievements on sediment motion and transport, this dissertation elaborates the maximal grain size during incipient motion (MGSIM), which is the connection between sediment motion and sediment transport, moreover, this dissertation also analyses the traditional investigations of MGSIM, which are also faced with a lots of problems; at the same time, artificial neural network (ANN) is an approximate simulation of biologic nerve system, which is a network model with a special algorithm got from biologic prototype after abstractly research. This dissertation chooses the BP algorithm which is a most popular and mature artificial neural network model. The author uses VC#.NET platform, SQL2000 database system and the experiment data (the gross bed-load transport rate of Non-uniform sediment with a wide distribution in flume) to develop the forecast software, which is friendly-interface, multi-input, single output, three layers' artificial neural network base on BP and improved BP algorithms. Aiming at the classical BP algorithm's limitation, the dissertation introduces and realizes two improved BP algorithms, which are called self-adapting adjust rate algorithm and simulated annealing algorithm. Secondly, the applications of BP and improved BP algorithms in sediment science are by using the self-determination intellectual property forecast software and Matlab neural network toolbox to compute, forecast and analyze the experiment data; on the other hand, the dissertation probes into a parameter (t) of diversion function about BP algorithm. It discusses and analyses the relationship between t and computation precision of BP algorithm on the basis of abundant experiment data, furthermore it proposes a formula for values of t independently. Thirdly, the results which are computed by classical BP algorithm, self-adapting adjust rate algorithm, simulated annealing algorithm and Matlab neural network toolbox respectively compare with each other in transverse and longitudinal ways.
Lastly, it draws conclusion as follows: computation in Non-uniform sediment with a wide distribution in flume experiment of stead sediment transportation by the software which is developed by VC#.NET platform satisfies with requirement of engineering, especially the simulated annealing algorithm, its stability of the network and precision of computation are superior to Matlab neural network toolbox. Those conclusions show that the applications of BP and improved BP algorithms in sediment science are feasible and also both have more values of investigation and commerce, besides, the software can be an auxiliary and referenced tool for sediment science research.
本文首先对泥沙的起动和推移质研究成果做简要的回顾性总结，并在此基础上详细阐述了泥沙起动问题和推移质问题联系的纽带——推移质最大起动粒径研究的意义以及就目前研究水平所面临的问题；与此同时，人工神经网络理论则模仿人脑的思维，做出判断和预测，它可以借助有限的实测资料获取经验，揭示输入资料潜在的效应和变化趋势，无需用户建立函数关系，对复杂的非线性系统尤为有效。其次，选择目前采用最为广泛的BP神经网络模型、通过VC#．NET开发平台和SQL2000数据库并结合平衡输沙状态下的水槽输沙试验数据，开发出界面友好、基于BP及其改进算法的多输入单输出三层神经网络预测软件。针对BP网络算法所固有的缺点，提出并实现的两种改进算法分别为自适应调整学习率算法和模拟退火算法。再次，利用作者开发的带有自主知识产权的预测软件和Matlab神经网络工具箱，对实验数据做计算、预测和分析；与此同时，针对BP算法的转换函数中参数t对网络模型输出精度和稳定性的影响展开探索性的研究，并在大量实验数据的基础上提出了参数t的取值公式与适用范围。最后，通过对基于经典BP网络算法、自适应调整学习率改进算法、模拟退火改进算法预测结果与Matlab神经网络工具箱预测结果之间科学详尽的横向与纵向的数据比较与稳定性分析得出本文的结论。
本文的结论为：对于平衡状态下的宽级配非均匀沙水槽输沙试验中最大起动粒径的推求，利用VC#．NET开发出界面友好、基于BP及其改进算法的多输入单输出三层神经网络预测软件，其计算精度满足工程要求，特别对模拟退火改进算法，其计算精度比成熟的Matlab工具箱神经网络工具箱计算精度还要高，显示出光明的应用前景与商业价值，对于泥沙起动和推移质的研究有一定的辅助作用和参考价值。
Recent years, along with the development of economy and the endless depredation of nature, the phenomenon of water and soil loss in our country is more and more serious, which causes much more sediment into the rivers .That arouses the complexity of management to the rivers and many more severe disasters. For the sake of scientific development outlook and a harmonious society, deeper in investigation of sediment is imperatively. But, unfortunately, the problems about sediment science have not been solved completely until now.
The main contents are as follows: First of all, on the basis of comprehensive review of the study achievements on sediment motion and transport, this dissertation elaborates the maximal grain size during incipient motion (MGSIM), which is the connection between sediment motion and sediment transport, moreover, this dissertation also analyses the traditional investigations of MGSIM, which are also faced with a lots of problems; at the same time, artificial neural network (ANN) is an approximate simulation of biologic nerve system, which is a network model with a special algorithm got from biologic prototype after abstractly research. This dissertation chooses the BP algorithm which is a most popular and mature artificial neural network model. The author uses VC#.NET platform, SQL2000 database system and the experiment data (the gross bed-load transport rate of Non-uniform sediment with a wide distribution in flume) to develop the forecast software, which is friendly-interface, multi-input, single output, three layers' artificial neural network base on BP and improved BP algorithms. Aiming at the classical BP algorithm's limitation, the dissertation introduces and realizes two improved BP algorithms, which are called self-adapting adjust rate algorithm and simulated annealing algorithm. Secondly, the applications of BP and improved BP algorithms in sediment science are by using the self-determination intellectual property forecast software and Matlab neural network toolbox to compute, forecast and analyze the experiment data; on the other hand, the dissertation probes into a parameter (t) of diversion function about BP algorithm. It discusses and analyses the relationship between t and computation precision of BP algorithm on the basis of abundant experiment data, furthermore it proposes a formula for values of t independently. Thirdly, the results which are computed by classical BP algorithm, self-adapting adjust rate algorithm, simulated annealing algorithm and Matlab neural network toolbox respectively compare with each other in transverse and longitudinal ways.
Lastly, it draws conclusion as follows: computation in Non-uniform sediment with a wide distribution in flume experiment of stead sediment transportation by the software which is developed by VC#.NET platform satisfies with requirement of engineering, especially the simulated annealing algorithm, its stability of the network and precision of computation are superior to Matlab neural network toolbox. Those conclusions show that the applications of BP and improved BP algorithms in sediment science are feasible and also both have more values of investigation and commerce, besides, the software can be an auxiliary and referenced tool for sediment science research.
引文
[1] 夏建新，吉祖稳．水沙环境学．北京：海洋出版社，2004．
[2] 苑希民，李鸿雁．神经网络和遗传算法在水科学领域的应用．北京：中国水利水电出版社，2002．
[3] 王兴奎，王光谦．河流动力学．北京：科学出版社，2004．
[4] 黄才安．水泥沙运动基本规律．北京：海洋出版社，2004．
[5] 杨志达．泥沙输送理论与实践．北京：中国水利水电出版社，2003．
[6] 韩力群．人工神经网络教程．北京：北京邮电大学出版社，2006．
[7] 王光谦．河流泥沙研究进展．泥沙研究，2007，2：64-74．
[8] 谢鉴衡．河流泥沙工程学．北京：水利出版社，1981．
[9] 钱宁，万兆惠．泥沙运动力学．北京：科学出版社，1983．
[10] 郭庆超．天然河道水流挟沙能力研究．泥沙研究，2006，5：45-51．
[11] 唐日长．泥沙研究．北京：水利电力出版社，1990．
[12] 唐村本．泥沙起动规律．水利学报，1963，2：125-129．
[13] 黄才安，严恺．论泥沙运动理论的流派及其相互关系．水道港口，2003，3：105-110．
[14] He wenshe, Cao Shuyou, Liu Xinnniu. Study on critical shear stress of incipientmotion of sedinent particles. Acta Mechanica Sinica, 2003 35(3): 328-331.
[15] 侯晖昌．河流动力学基本问题．北京：水利出版社，1982．
[16] 窦国仁．再论泥沙起动流速．水利学报，1996，6：1-9．
[17] Qian N, Wan Z. Mechanics of sediment transport. New York:ASCE Press, 1999.
[18] 高建恩．推移质输沙规律的再探讨．水利学报，1993，4：62-68．
[19] 黄才安，奚斌．水流强度指标与推移质输沙率．扬州大学学报，1999，2(1)：66-71．
[20] 韩其为，为何民．泥沙运动统计理论．北京：科学出版社，1984．
[21] 何文社，曹叔尤．非均匀推移质输移特性研究．四川大学学报，2006，38(6)：1-5．
[22] 王勇涛，倪汉根，崔莉．非均匀推移质级配研究．泥沙研究，2004，1：50-55．
[23] 孙志林，田林．非均匀沙分级推移质公式．泥沙研究，2001，5：65-69．
[24] Guo, Qingchao. EshimaLingcoel Ticienhs in one-dimensional depth-averaged sediment transport model. Canadian Journal of Civil Ennineerinn, 2001,28:536-540.
[25] WeigendAS, BumelhartDE, HubennanBA. Preclicting the future: ac-onnectionista Proaching. International, Tournaloileural Spstpans, 1990,1(3):193-209.
[26] Chien Ning. Meyer-Peter formula for bed load transport and Einstein bed load function. Missouri River Division, 1954(7).
[27] 胡手仁．神经网络应用技术．北京：国防科技大学出版社，1993．
[28] 焦李成．神经网络的应用与实现．西安：西安电子科技大学出版社，1995．
[29] 史忠植．神经计算．北京：电子工业出版社，1993．
[30] 罗晓曙．人工神经网络理论．桂林：广西师范大学出版社，2005．
[31] 陈刚，刘发升．基于BP神经网络的数据挖掘方法．计算机与现代化，2006，134(10)：20-22．
[32] 潘华．数据挖掘及神经网络在土木工程中的应用：(硕士学位论文)．重庆：重庆大学，2005．
[33] 王伟．人工神经网络原理．北京：北京航空航天大学出版，1995．
[34] 于东生，严以新，田淳．基于BP算法的泥沙含量预测研究．三峡大学学报，2003，25(1)：47-51．
[35] 张先起，刘慧卿，梁川．流域产沙量预测的神经网络模型．云南水力发电，21(6)：11-14．
[36] 杨华民，王文成．一种BP改进算法．微型计算机，1997，17(2)：53-54．
[37] Magnitskii N.A .Some New Approaches to the Construction and Leading of Artificial Neural Network. Computation Mathematics and Modeling, 2001,12(4):293-304.
[38] M R A Sadjadii, R J Liou Fast learning process of multilayer neural networks using recursive learsl squares method. IEEE. Trans SP, 1992,40(2):446-450.
[39] 张涛，席道瑛．模拟退火BP网络理论与算法．电脑开发与应用，1997，11(2)：23-25．
[40] 李丙春，汪仲文．基于模拟退火的神经网络预测模型．喀什师范学院学报，2004，25(3)：63-65．
[41] Nagy, H.M.,Watanabe. K. Prediction of Sediment Load Concentration in Rivers using Artificial Neural NetworkModel. Journal of Hydraulic Engineering, 2002,128(6):588-596.
[42] 刘兴年，曹叔尤．宽级配推移质输移特性研究．泥沙研究，2002，6：42-48．
[43] 曹祖圣，吴明哲．Visual C#．NET程序设计经典．北京：科学出版社，2004．
[44] 肖玉刚．一种新型模拟退火神经网络及其应用．控制工程，2006，11(6)：551-556．
[45] Jain Ashu, KumarVarshney Ashish, Chandra Joshi Umesh. Short-TermWater Demand Forecast Modeling at IIT Kan-pur using Artificial Neural Network. Water Resource Management, 2001, 15(5):299-321.
[46] 何文社．非均匀沙运动规律的研究：(硕士学位论文)．重庆：四川大学，2002．
[47] 唐造造．宽级配下非均匀沙输移规律的试验研究：(博士学位论文)．重庆：四川大学，1996．
[48] 闻新．MATLAB神经网络仿真与应用．北京：科学出版社，2003．
[49] 飞思科技．MATLAB辅助神经网络分析与设计．北京：电子工业出版社，2003．
[2] 苑希民，李鸿雁．神经网络和遗传算法在水科学领域的应用．北京：中国水利水电出版社，2002．
[3] 王兴奎，王光谦．河流动力学．北京：科学出版社，2004．
[4] 黄才安．水泥沙运动基本规律．北京：海洋出版社，2004．
[5] 杨志达．泥沙输送理论与实践．北京：中国水利水电出版社，2003．
[6] 韩力群．人工神经网络教程．北京：北京邮电大学出版社，2006．
[7] 王光谦．河流泥沙研究进展．泥沙研究，2007，2：64-74．
[8] 谢鉴衡．河流泥沙工程学．北京：水利出版社，1981．
[9] 钱宁，万兆惠．泥沙运动力学．北京：科学出版社，1983．
[10] 郭庆超．天然河道水流挟沙能力研究．泥沙研究，2006，5：45-51．
[11] 唐日长．泥沙研究．北京：水利电力出版社，1990．
[12] 唐村本．泥沙起动规律．水利学报，1963，2：125-129．
[13] 黄才安，严恺．论泥沙运动理论的流派及其相互关系．水道港口，2003，3：105-110．
[14] He wenshe, Cao Shuyou, Liu Xinnniu. Study on critical shear stress of incipientmotion of sedinent particles. Acta Mechanica Sinica, 2003 35(3): 328-331.
[15] 侯晖昌．河流动力学基本问题．北京：水利出版社，1982．
[16] 窦国仁．再论泥沙起动流速．水利学报，1996，6：1-9．
[17] Qian N, Wan Z. Mechanics of sediment transport. New York:ASCE Press, 1999.
[18] 高建恩．推移质输沙规律的再探讨．水利学报，1993，4：62-68．
[19] 黄才安，奚斌．水流强度指标与推移质输沙率．扬州大学学报，1999，2(1)：66-71．
[20] 韩其为，为何民．泥沙运动统计理论．北京：科学出版社，1984．
[21] 何文社，曹叔尤．非均匀推移质输移特性研究．四川大学学报，2006，38(6)：1-5．
[22] 王勇涛，倪汉根，崔莉．非均匀推移质级配研究．泥沙研究，2004，1：50-55．
[23] 孙志林，田林．非均匀沙分级推移质公式．泥沙研究，2001，5：65-69．
[24] Guo, Qingchao. EshimaLingcoel Ticienhs in one-dimensional depth-averaged sediment transport model. Canadian Journal of Civil Ennineerinn, 2001,28:536-540.
[25] WeigendAS, BumelhartDE, HubennanBA. Preclicting the future: ac-onnectionista Proaching. International, Tournaloileural Spstpans, 1990,1(3):193-209.
[26] Chien Ning. Meyer-Peter formula for bed load transport and Einstein bed load function. Missouri River Division, 1954(7).
[27] 胡手仁．神经网络应用技术．北京：国防科技大学出版社，1993．
[28] 焦李成．神经网络的应用与实现．西安：西安电子科技大学出版社，1995．
[29] 史忠植．神经计算．北京：电子工业出版社，1993．
[30] 罗晓曙．人工神经网络理论．桂林：广西师范大学出版社，2005．
[31] 陈刚，刘发升．基于BP神经网络的数据挖掘方法．计算机与现代化，2006，134(10)：20-22．
[32] 潘华．数据挖掘及神经网络在土木工程中的应用：(硕士学位论文)．重庆：重庆大学，2005．
[33] 王伟．人工神经网络原理．北京：北京航空航天大学出版，1995．
[34] 于东生，严以新，田淳．基于BP算法的泥沙含量预测研究．三峡大学学报，2003，25(1)：47-51．
[35] 张先起，刘慧卿，梁川．流域产沙量预测的神经网络模型．云南水力发电，21(6)：11-14．
[36] 杨华民，王文成．一种BP改进算法．微型计算机，1997，17(2)：53-54．
[37] Magnitskii N.A .Some New Approaches to the Construction and Leading of Artificial Neural Network. Computation Mathematics and Modeling, 2001,12(4):293-304.
[38] M R A Sadjadii, R J Liou Fast learning process of multilayer neural networks using recursive learsl squares method. IEEE. Trans SP, 1992,40(2):446-450.
[39] 张涛，席道瑛．模拟退火BP网络理论与算法．电脑开发与应用，1997，11(2)：23-25．
[40] 李丙春，汪仲文．基于模拟退火的神经网络预测模型．喀什师范学院学报，2004，25(3)：63-65．
[41] Nagy, H.M.,Watanabe. K. Prediction of Sediment Load Concentration in Rivers using Artificial Neural NetworkModel. Journal of Hydraulic Engineering, 2002,128(6):588-596.
[42] 刘兴年，曹叔尤．宽级配推移质输移特性研究．泥沙研究，2002，6：42-48．
[43] 曹祖圣，吴明哲．Visual C#．NET程序设计经典．北京：科学出版社，2004．
[44] 肖玉刚．一种新型模拟退火神经网络及其应用．控制工程，2006，11(6)：551-556．
[45] Jain Ashu, KumarVarshney Ashish, Chandra Joshi Umesh. Short-TermWater Demand Forecast Modeling at IIT Kan-pur using Artificial Neural Network. Water Resource Management, 2001, 15(5):299-321.
[46] 何文社．非均匀沙运动规律的研究：(硕士学位论文)．重庆：四川大学，2002．
[47] 唐造造．宽级配下非均匀沙输移规律的试验研究：(博士学位论文)．重庆：四川大学，1996．
[48] 闻新．MATLAB神经网络仿真与应用．北京：科学出版社，2003．
[49] 飞思科技．MATLAB辅助神经网络分析与设计．北京：电子工业出版社，2003．