Research on improved BP neural network in forecasting traffic accidents
-
摘要: 交通事故严重程度受多种因素的影响,适合用人工神经网络来建模预测.因为标准BP(Back Propagation,BP)神经网络具有收敛较慢的缺陷,所以在自适应学习和附加动量因子改进BP神经网络[1]基础上做了进一步的优化改进,使附加动量因子也具有自学习性.利用改进后的BP神经网络算法,选取英国利兹市的公开交通事故数据集,用影响交通事故严重程度的多种影响因素和事故严重程度构建并训练神经网络,并对最新数据进行预测.通过大量的实验对比收敛速度和预测结果,验证了改进后的算法具有更快的收敛速度和更高的预测准确率.Abstract: The traffic accident severity is affected by many factors. It is suitable for modeling and forecasting by using the artificial neural network (ANN). Because standard BP (back propagation) neural network has the defect of slow convergence, based on the improved BP neural network with adaptive learning and additional momentum factor[1], so the additional momentum factor was made to be self-learning for further optimization and improvement. Using the improved BP neural network algorithm, the public traffic accident data set in Leeds of England was selected to construct and train the neural network to predict the latest records. The data set includes many kinds of influencing factors and accident severity. After a lot of experiments, by comparing the convergence rate and prediction results, it has been proved that the improved algorithm has faster convergence rate and higher forecasting accuracy rate.
-
Key words:
- BP neural network /
- momentum factor /
- self-learning /
- traffic accident
-
表 1 不同方法的预测准确率
Tab. 1 The prediction accuracy of different methods
网络类别 平均准确率 (动量因子η取0到1之间) 标准网络 0.87 固定动量因子 0.9 自学习动量因子 0.913 -
[1] ZHU P, LIN C Y, WU P, et al. Permeability prediction of tight sandstone reservoirs using improved BP neural network[J]. The Open Petroleum Engineering Journal, 2015, 8(1):288-292. doi: 10.2174/1874834101508010288 [2] ZONG F, XU H G, ZHANG H Y. Prediction for traffic accident severity:Comparing the Bayesian network and regression models[J]. Mathematical Problems in Engineering, 2013(3):206-226. https://www.researchgate.net/publication/258394291_Prediction_for_Traffic_Accident_Severity_Comparing_the_Bayesian_Network_and_Regression_Models [3] 李铃铃, 仇蕾.基于灰色神经网络组合模型的水上交通事故预测[J].交通信息安全, 2014, 32(3):110-118. http://www.cnki.com.cn/Article/CJFDTOTAL-JTJS201403023.htm [4] 唐秋生, 杜营营.基于BP神经网络的交通事故预测模型及仿真[J].交通信息安全, 2011, 29(1):68-70. http://www.cnki.com.cn/Article/CJFDTOTAL-JTJS201101019.htm [5] ABELLÁN J, LÓPEZ G, OÑA J D. Analysis of traffic accident severity using decision rules via decision trees[J]. Expert Systems with Applications, 2013, 40(15):6047-6054. doi: 10.1016/j.eswa.2013.05.027 [6] 赵玲, 许宏科.基于改进的灰色马尔可夫链模型的交通事故预测[J].数学的实践与认识, 2013, 43(20):92-98. http://www.cnki.com.cn/Article/CJFDTOTAL-SSJS201320013.htm [7] 邢晓敏, 商国敬.基于改进共轭梯度理论神经网络优化算法研究[J].电器与仪表, 2014, 51(19):36-40. http://www.cnki.com.cn/Article/CJFDTOTAL-DCYQ201419007.htm [8] 张峤, 邓贵仕. Levenberg-Marquardt神经网络在煤矿作业人员人因可靠性评价中应用研究[J].大连理工大学学报, 2015, 55(4):424-430. http://www.cnki.com.cn/Article/CJFDTOTAL-DLLG201504014.htm [9] 焦纲领, 王士星.基于遗传算法的神经网络参数优化研究[J].数学的实践与认识, 2012, 42(11):176-183. http://www.cnki.com.cn/Article/CJFDTOTAL-SSJS201211026.htm [10] 刘春艳, 凌建春, 寇林元, 等. GA-BP神经网络与BP神经网络性能比较[J].中国卫生统计, 2013, 30(2):173-181. http://www.cnki.com.cn/Article/CJFDTOTAL-ZGWT201302005.htm [11] 高玉明, 张仁津.基于遗传算法和BP神经网络的房价预测分析[J].计算机工程, 2014, 40(4):187-191. http://www.cnki.com.cn/Article/CJFDTOTAL-JSJC201404037.htm [12] 许榕, 周东, 蒋士正, 等.自适应粒子群神经网络交通流预测模型[J].西安交通大学学报, 2015, 49(10):103-109. http://www.cnki.com.cn/Article/CJFDTOTAL-XAJT201510017.htm [13] 张德慧, 张德育, 刘清云, 等.基于粒子群算法的BP神经网络优化技术[J].计算机工程与设计, 2015, 36(5):1321-1326. [14] 江丽, 王爱平.基于粒子群优化与BP算法的协同神经网络学习方法[J].计算机应用与软件, 2013, 30(9):19-20. http://www.cnki.com.cn/Article/CJFDTOTAL-JYRJ201309007.htm [15] 刘卫宁, 王鹏, 孙棣华, 等.基于改进BP神经网络的道路交通事故预测[J].计算机系统应用, 2010, 19(10):177-181. http://www.cnki.com.cn/Article/CJFDTOTAL-XTYY201010039.htm