群优化BP网在交通事故预测中的应用研究
|
摘要
随着经济的发展,各种机动交通工具已经成为一种十分重要的出行方式。然而,道路交通上频繁发生的各种交通事故却不断地威胁着人们的生命和财产安全。因此,交通局有关部门希望可以从长期累积的交通事故的历史数据中,提取出有价值的信息,.为关系到整个交通事业安全方面的相关工作提供科学的决策支持。
神经网络作为一种数据挖掘的方式,长期以来被广泛地应用于各种领域的数据分析和预测。粒子群算法是近年来比较新兴的算法,将粒子群算法应用到BP神经网络,作为训练算法,可以优化网络的性能,克服BP算法的局限性。
本文首先从研究道路交通事故预测的必要性入手,分析了从事交通事故预测的复杂性以及所要面对的问题。其次,论述了国内外常见的交通事故预测方法,并对这些方法进行必要的分析,分析了各自存在的特点以及不足之处。BP神经网络这一种带有前向反馈的神经网络模型由于在解决复杂非线性系统问题方面的优势以及能够进行广泛学习训练的特性,决定了它作为一种道路交通事故预测方法来进行预测建模会取得不一样的特性。针对神经网络的这一特点,提出了采用BP神经网络来进行道路交通事故预测的方法,并建立预测模型。但是,基本的BP神经网络预测模型具有收敛速度慢、易于陷入极小值的缺陷,采用了粒子群优化算法作为网络的训练算法解决这一问题。为了提高神经网络训练和检验的数据的可靠性和针对性,对样本数据进行预处理,采用与多元回归分析法相结合来确定神经网络的输入变量,提高输入变量与预测变量的相关度。
最后对使用标准BP神经网络和基于群优化算法的BP神经网络分别进行建模和检验,并对实验的结果进行比较分析。实验结果证明基于粒子群优化算法的BP神经网络模型拥有更高的精确度。
With economic development, all kinds of motor vehicles have become a very important way to travel. However, the frequent occurrence of road traffic accidents has continued to threaten all people's lives and property. Therefore, the Department of Transportation authorities hope the long term accumulation of historical data of traffic accidents, to extract valuable information, as relates to the transport safety related work to provide scientific decision support. Neural network as a data mining approach has long been widely used in various fields of data analysis and forecasting. PSO is a relatively new method in recent years, the particle swarm algorithm is applied to the BP neural network as its training algorithm, can optimize network performance, to overcome the limitations of BP algorithm. This paper studies the need for road traffic forecast to start, that projections in the traffic and the complexity of the problems faced; followed by the common accidents at home and abroad discussed the prediction method, and the necessary analysis of these methods, that their characteristics of the existing shortcomings; and BP neural network with this kind of feedback before the neural network model in solving the problems of complex nonlinear systems have the advantage of extensive training to learn the characteristics, as a road Accident prediction model will be made to different characteristics. This feature for the neural network is proposed by using BP neural network prediction method for road traffic accidents, and the establishment of forecasting model, the basic BP neural network model has a slow convergence and easy to fall into the minimum of defects, which PSO algorithm as the network training algorithms. In order to improve the neural network training and testing the reliability and relevance of the data, pretreatment of the sample data, using multiple regression analysis method with the combination to determine the neural network input variables, input variables and the soon to improve the relevance of predictor variables. Finally, using the standard BP neural networks and swarm optimization algorithm based on BP neural networks for modeling and testing, the experimental results were compared. Experimental results show that particle swarm optimization algorithm based on BP neural network model with higher accuracy.
神经网络作为一种数据挖掘的方式,长期以来被广泛地应用于各种领域的数据分析和预测。粒子群算法是近年来比较新兴的算法,将粒子群算法应用到BP神经网络,作为训练算法,可以优化网络的性能,克服BP算法的局限性。
本文首先从研究道路交通事故预测的必要性入手,分析了从事交通事故预测的复杂性以及所要面对的问题。其次,论述了国内外常见的交通事故预测方法,并对这些方法进行必要的分析,分析了各自存在的特点以及不足之处。BP神经网络这一种带有前向反馈的神经网络模型由于在解决复杂非线性系统问题方面的优势以及能够进行广泛学习训练的特性,决定了它作为一种道路交通事故预测方法来进行预测建模会取得不一样的特性。针对神经网络的这一特点,提出了采用BP神经网络来进行道路交通事故预测的方法,并建立预测模型。但是,基本的BP神经网络预测模型具有收敛速度慢、易于陷入极小值的缺陷,采用了粒子群优化算法作为网络的训练算法解决这一问题。为了提高神经网络训练和检验的数据的可靠性和针对性,对样本数据进行预处理,采用与多元回归分析法相结合来确定神经网络的输入变量,提高输入变量与预测变量的相关度。
最后对使用标准BP神经网络和基于群优化算法的BP神经网络分别进行建模和检验,并对实验的结果进行比较分析。实验结果证明基于粒子群优化算法的BP神经网络模型拥有更高的精确度。
With economic development, all kinds of motor vehicles have become a very important way to travel. However, the frequent occurrence of road traffic accidents has continued to threaten all people's lives and property. Therefore, the Department of Transportation authorities hope the long term accumulation of historical data of traffic accidents, to extract valuable information, as relates to the transport safety related work to provide scientific decision support. Neural network as a data mining approach has long been widely used in various fields of data analysis and forecasting. PSO is a relatively new method in recent years, the particle swarm algorithm is applied to the BP neural network as its training algorithm, can optimize network performance, to overcome the limitations of BP algorithm. This paper studies the need for road traffic forecast to start, that projections in the traffic and the complexity of the problems faced; followed by the common accidents at home and abroad discussed the prediction method, and the necessary analysis of these methods, that their characteristics of the existing shortcomings; and BP neural network with this kind of feedback before the neural network model in solving the problems of complex nonlinear systems have the advantage of extensive training to learn the characteristics, as a road Accident prediction model will be made to different characteristics. This feature for the neural network is proposed by using BP neural network prediction method for road traffic accidents, and the establishment of forecasting model, the basic BP neural network model has a slow convergence and easy to fall into the minimum of defects, which PSO algorithm as the network training algorithms. In order to improve the neural network training and testing the reliability and relevance of the data, pretreatment of the sample data, using multiple regression analysis method with the combination to determine the neural network input variables, input variables and the soon to improve the relevance of predictor variables. Finally, using the standard BP neural networks and swarm optimization algorithm based on BP neural networks for modeling and testing, the experimental results were compared. Experimental results show that particle swarm optimization algorithm based on BP neural network model with higher accuracy.
引文
[1]中国统计年鉴社.中国统计年鉴1998-2003[M].北京：中国统计年鉴社,1998-2003.
[2]段里仁.道路交通事故概论[M],北京：中国人民公安大学出版社,1997.
[3]王岩.道路交通事故危险点的判定和成因分析[D].长沙交通学院硕士学位论文,2002.
[4]谷志杰,丛国权.交通事故处理及其预防[M].北京：中国人民公安大学出版社,2002.
[5]过秀成.道路交通安全学[M].南京：东南大学出版社,2001.
[6]El-Sadig M. Road traffic accidents in the United Arab Emirates:trends of morbidity and mortality during 1977-1998[J]. Accident Analysis & Prevention, 2002,34(5):465-476.
[7]李相勇.道路交通事故预测方法研究[D].西南交通大学硕士学位论文,2004.
[8]陈明伟,袁晓华,潘敏等.从道路交通事故统计分析对比谈预防措施[J].中国安全科学学报,2004,32(8)：59-63.
[9]王慈光.二次指数平滑法中确定初始值的简便方法[J].西南交通大学学报,2004,25(6)：269-271.
[10]陈宁,朱美琪,余珍文.基于对数二次指数平滑的港口吞吐量预测[J].武汉理工大学学报,2005,21(9)：77-79.
[11]马晓坷,王慈光.三次指数平滑法在大秦铁路运量预测中的应用[J].华东交通大学学报,2005,24(6)：8-11.
[12]张启锐.实用回归分析[M].北京：地质出版社,1988.
[13]刘建齐,陈兰,刘建武.道路交通事故预测中的灰色预测GM(1,1)模型[J].广西交通科技,2003,19(4)：106-109.
[14]彭冬芝.基于灰平面的事故预测方法研究[J].中国安全科学学报,2004,10(4)：36-39.
[15]胡守仁.神经网络导论[M].长沙：国防科技大学出版社,1993.
[16]Engelbrecht A. P., Ismail A. Training product unit neural networks, Stability and Control [J]. Theory and Applications,1999,47(5):59-74.
[17]颜七笙,游泳和杨志辉.BP网络结构设计与算法探讨[J].科技广场,2005,12(3)：36-38
[18]Kennedy J, Eberhart R C. Particle swarm optimization [A]. Proceedings of IEEE International Conference on. Neutral Networks [C]. Australia:IEEE, 1995,39 (3):1942-1948.
[19]K. W. Chau. Particle swarm optimization training algorithm for ANNs in stage prediction of Shing Mun River [J]. Journal of Hydrology,2006,9(5):363-367.
[20]刘洪波,王秀坤和孟军.神经网络基于粒子群优化的学习算法研究[J].小型微型计算机系统,2005,26(4)：638-640.
[21]M Geethanjali, S. Mary Raja Slochanal&R. Bhavani. PSO trained ANN-based differential protection scheme for power transformers [J]. Neurocomputing,2007, doi:10.1016/j.neucom.2007.02.014.
[22]R Eberhart, J Kennedy. A new optimizer using particle swarm theory[C]. Proc of 6th Int Symposium on Micro Machine and Human Science,1995,34(2):39-43.
[23]R. Eberhart, J. Kennedy. Particle swarm optimization[C]. Proc. IEEE Int. Coni:on Neural Networks, Perth, Australia,1995,26(5):1942-1948.
[24]张志伟.粗糙集理论及其在数据预处理中的应用[D].硕士论文,华南理工大学,2004.
[25]闻新.MATLAB神经网络应用设计[M].北京：科学出版社,2002：207-232.
[26]Rumelhart D E, Hinton G E & Williams R J. Learning representations by back propagating errors [J]. Nature,1986,323(11):533-536.
[27]Shi Y, Eberhart R C. A modified particle swarm optimizer[C]. Proc. IEEE Int'l Conf. on Evolutionary Computation, NJ,1998,64(33):69-73.
[28]Shi Y H, Eberhart R C. Empirical study of particle swarm optimization [A]. Proceedings of'IEEE International Congress on Evolutionary Computation[C], USA: IEEE,1999,54(21):6-9.
[29]Shi Y H., Eberhart R. C. Parameter selection in particle swarm optimization [A]. The 7th Annual Conference on Evolutionary Programming[C], San Dlego, CA, USA: Springer Verlag,1998,45(14):591-601.
[30]Yoshida H, Kawata K & Yoshikazu F. A Particle swarm optimization for reactive power and voltage control considering voltage security assessment [J]. IEEE Transaction on Power System,2000,15(4):1232-1239.
[2]段里仁.道路交通事故概论[M],北京：中国人民公安大学出版社,1997.
[3]王岩.道路交通事故危险点的判定和成因分析[D].长沙交通学院硕士学位论文,2002.
[4]谷志杰,丛国权.交通事故处理及其预防[M].北京：中国人民公安大学出版社,2002.
[5]过秀成.道路交通安全学[M].南京：东南大学出版社,2001.
[6]El-Sadig M. Road traffic accidents in the United Arab Emirates:trends of morbidity and mortality during 1977-1998[J]. Accident Analysis & Prevention, 2002,34(5):465-476.
[7]李相勇.道路交通事故预测方法研究[D].西南交通大学硕士学位论文,2004.
[8]陈明伟,袁晓华,潘敏等.从道路交通事故统计分析对比谈预防措施[J].中国安全科学学报,2004,32(8)：59-63.
[9]王慈光.二次指数平滑法中确定初始值的简便方法[J].西南交通大学学报,2004,25(6)：269-271.
[10]陈宁,朱美琪,余珍文.基于对数二次指数平滑的港口吞吐量预测[J].武汉理工大学学报,2005,21(9)：77-79.
[11]马晓坷,王慈光.三次指数平滑法在大秦铁路运量预测中的应用[J].华东交通大学学报,2005,24(6)：8-11.
[12]张启锐.实用回归分析[M].北京：地质出版社,1988.
[13]刘建齐,陈兰,刘建武.道路交通事故预测中的灰色预测GM(1,1)模型[J].广西交通科技,2003,19(4)：106-109.
[14]彭冬芝.基于灰平面的事故预测方法研究[J].中国安全科学学报,2004,10(4)：36-39.
[15]胡守仁.神经网络导论[M].长沙：国防科技大学出版社,1993.
[16]Engelbrecht A. P., Ismail A. Training product unit neural networks, Stability and Control [J]. Theory and Applications,1999,47(5):59-74.
[17]颜七笙,游泳和杨志辉.BP网络结构设计与算法探讨[J].科技广场,2005,12(3)：36-38
[18]Kennedy J, Eberhart R C. Particle swarm optimization [A]. Proceedings of IEEE International Conference on. Neutral Networks [C]. Australia:IEEE, 1995,39 (3):1942-1948.
[19]K. W. Chau. Particle swarm optimization training algorithm for ANNs in stage prediction of Shing Mun River [J]. Journal of Hydrology,2006,9(5):363-367.
[20]刘洪波,王秀坤和孟军.神经网络基于粒子群优化的学习算法研究[J].小型微型计算机系统,2005,26(4)：638-640.
[21]M Geethanjali, S. Mary Raja Slochanal&R. Bhavani. PSO trained ANN-based differential protection scheme for power transformers [J]. Neurocomputing,2007, doi:10.1016/j.neucom.2007.02.014.
[22]R Eberhart, J Kennedy. A new optimizer using particle swarm theory[C]. Proc of 6th Int Symposium on Micro Machine and Human Science,1995,34(2):39-43.
[23]R. Eberhart, J. Kennedy. Particle swarm optimization[C]. Proc. IEEE Int. Coni:on Neural Networks, Perth, Australia,1995,26(5):1942-1948.
[24]张志伟.粗糙集理论及其在数据预处理中的应用[D].硕士论文,华南理工大学,2004.
[25]闻新.MATLAB神经网络应用设计[M].北京：科学出版社,2002：207-232.
[26]Rumelhart D E, Hinton G E & Williams R J. Learning representations by back propagating errors [J]. Nature,1986,323(11):533-536.
[27]Shi Y, Eberhart R C. A modified particle swarm optimizer[C]. Proc. IEEE Int'l Conf. on Evolutionary Computation, NJ,1998,64(33):69-73.
[28]Shi Y H, Eberhart R C. Empirical study of particle swarm optimization [A]. Proceedings of'IEEE International Congress on Evolutionary Computation[C], USA: IEEE,1999,54(21):6-9.
[29]Shi Y H., Eberhart R. C. Parameter selection in particle swarm optimization [A]. The 7th Annual Conference on Evolutionary Programming[C], San Dlego, CA, USA: Springer Verlag,1998,45(14):591-601.
[30]Yoshida H, Kawata K & Yoshikazu F. A Particle swarm optimization for reactive power and voltage control considering voltage security assessment [J]. IEEE Transaction on Power System,2000,15(4):1232-1239.