基于神经网络的地铁短时客流预测服务
|
摘要
短时客流预测在为人类构建智慧城市,提供风险预警,保证出行安全中扮演着重要的角色.本文在神经网络算法的基础上,结合卡尔曼滤波,提出了一种新型有效的地铁客流短时预测算法.对于要预测的时刻t,算法利用它之前24小时的客流量作为输入特征.由于实验数据存在噪声,本文利用卡尔曼滤波对实验数据进行去躁平滑处理.最后算法利用BP神经网络和LSTM递归神经网络进行建模与预测.我们利用杭州地铁提供的真实购票数据进行大量实验,证明了BP神经网络(基于adam算法和relu激活函数)以及LSTM递归神经网络(基于adam算法和tanh激活函数的)准确度最高,预测的平均绝对误差最小(5%左右).另外,实验还证明了卡尔曼滤波能够有效减少预测的平均绝对误差.相比于不使用卡尔曼滤波的神经网络,使用卡尔曼滤波后的神经网络算法可以降低相对25%的MAE.
Subway short-term passenger flowforecast plays an important role in construction of intelligent city,providing risk warning and ensuring the safety. In this paper,based on the neural network,combined with Kalman filter,we propose an effective subway shortterm passenger flowforecast algorithm. The algorithm uses the 24 hours before the prediction time as the input feature. After filtering through the Kalman filter,the BP neural network and the LSTMrecurrent neural network are used to predict. The Finally,we use the real purchase data provided by Hangzhou subway to carry out a large number of experiments. The results showthat the BP neural network based on adam algorithm and relu activation function has a similar MAE( mean absolute error) with the recursive neural network algorithm based on adam algorithm and tanh activation function. Both of them have the minimum MAE( about 5%). In addition,it is proved that Kalman filter can effectively reduce the MAE of prediction. Compared with the non-use of Kalman filter neural network,the neural network algorithm with Kalman filter can reduce the relative 25% of the MAE.
Subway short-term passenger flowforecast plays an important role in construction of intelligent city,providing risk warning and ensuring the safety. In this paper,based on the neural network,combined with Kalman filter,we propose an effective subway shortterm passenger flowforecast algorithm. The algorithm uses the 24 hours before the prediction time as the input feature. After filtering through the Kalman filter,the BP neural network and the LSTMrecurrent neural network are used to predict. The Finally,we use the real purchase data provided by Hangzhou subway to carry out a large number of experiments. The results showthat the BP neural network based on adam algorithm and relu activation function has a similar MAE( mean absolute error) with the recursive neural network algorithm based on adam algorithm and tanh activation function. Both of them have the minimum MAE( about 5%). In addition,it is proved that Kalman filter can effectively reduce the MAE of prediction. Compared with the non-use of Kalman filter neural network,the neural network algorithm with Kalman filter can reduce the relative 25% of the MAE.
引文
[1] Li Guo-yong. Intelligent predictive control and its realization of MATLAB[M]. Beijing:Electronic Industry Press,2010.
[2]GoodfellowI,Bengio Y,Courville A. Deep learning[M]. USA,MIT Press,2016.
[3] Ye Xiao-zhou,Tao Fei-fei,Qi Rong-zhi,et al. Improvement of activation function in cyclic neural network structure[J]. Computer and Modernization,2016,(12):29-33.
[4]Le Q V,Jaitly N,Hinton G E. A simple way to initialize recurrent networks of rectified linear units[J]. ar Xiv Preprint ar Xiv:1504,00941,2015.
[5]Kaysi I. An integrated approach to vehicle routing and congestion prediction for real-time driver guidance[D]. Massachusetts Institute of Technology Cambridge,Mass,1993.
[6]Williams B M,Hoel L A. Modeling and forecasting vehicular traffic flowas a seasonal ARIMA process:theoretical basis and empirical results[J].Journal of Transportation Engineering,2003,129(6):664-672.
[7] ChingFu Chen,YuHern Chang,YuWei Chang. Seasonal ARIMA forecasting of inbound air travel arrivals to Taiwan[J]. Transportmetrica,2009,5(5):125-140.
[8]Cheng Cheng,Du Yu-chuan,Liu Xin. A forecast model of traffic hub traffic considering holiday effect[J]. Journal of Transportation Systems Engineering and Information,2015,15(5):202-207.
[9] Kavousi-Fard A,Khosravi A,Nahavandi S. A newfuzzy-based combined prediction interval for wind power forecasting[J]. IEEE Transactions on Power Systems,2016,31(1):18-26.
[10]Fang Li-jun,Ye Xia-fei,Ming Rui-li. A comparative analysis of the development trend of urban rail transit in Shanghai Seoul and[J].Journal of Traffic and Transportation,2007,(1):105-107.
[11]Chen Z,Deng T,Gao L,et al. A novel spatial-temporal detection method of dim infrared moving small target[J]. Infrared Physics&Technology,2014,66(9):84-96.
[12]Engle R F. Dynamic conditional beta[J]. Journal of Financial Econometrics,2016,14(4):643-667.
[13]Nahmias Biran B,Shiftan Y. Towards a more equitable distribution of resources:using activity-based models and subjective well-being measures in transport project evaluation[J]. Transportation Research Part A:Policy and Practice,2016,94:672-684.
[14]Varagouli ESmith B L,Williams B M,Oswald R K. Comparison of parametric and nonparametric models for traffic flowforecasting[J]. Transportation Research Part C Emerging Technologies,2002,10(4):303-321.
[15]G,Simos T E,Xeidakis G S. Fitting a multiple regression line to travel demand forecasting:The case of the prefecture of Xanthi,northern greece[J]. Mathematical&Computer Modelling,2005,42(7-8):817-836.
[16] Abdel-Aty MA,Pemmanaboina R. Calibrating a real-time traffic crash-prediction model using archived weather and ITS traffic data[J]. IEEE Transactions on Intelligent Transportation Systems,2006,7(2):167-174.
[17] Hu W,Yan L,Liu K,et al. A short-term traffic flowforecasting method based on the hybrid PSO-SVR[J]. Neural Processing Letters,2016,43(1):155-172.
[18]Deng W,Li W,Yang X H. A novel hybrid optimization algorithm of computational intelligence techniques for highway passenger volume prediction[J]. Expert Systems with Applications,2011,38(4):4198-4205.
[19] Dia H. An object-oriented neural network approach to short-term traffic forecasting[J]. European Journal of Operational Research,2001,131(2):253-261.
[20]Xiao J M,Wang X H. Study on traffic flowprediction using RBF neural network[C]. International Conference on Machine Learning and Cybernetics,Guangzhou,China,2004:2672-2675.
[1]李国勇.智能预测控制及其MATLAB实现[M].北京:电子工业出版社,2010.
[3]叶小舟,陶飞飞,戚荣志,等.循环神经网络结构中激活函数的改进[J].计算机与现代化,2016,(12):29-33.
[8]成诚,杜豫川,刘新.考虑节假日效应的交通枢纽客流量预测模型[J].交通运输系统工程与信息,2015,15(5):202-207.
[10]方礼君,叶霞飞,明瑞利.上海、首尔、东京城市轨道交通客流发展趋势对比分析[J].交通与运输,2007,(1):105-107.
[2]GoodfellowI,Bengio Y,Courville A. Deep learning[M]. USA,MIT Press,2016.
[3] Ye Xiao-zhou,Tao Fei-fei,Qi Rong-zhi,et al. Improvement of activation function in cyclic neural network structure[J]. Computer and Modernization,2016,(12):29-33.
[4]Le Q V,Jaitly N,Hinton G E. A simple way to initialize recurrent networks of rectified linear units[J]. ar Xiv Preprint ar Xiv:1504,00941,2015.
[5]Kaysi I. An integrated approach to vehicle routing and congestion prediction for real-time driver guidance[D]. Massachusetts Institute of Technology Cambridge,Mass,1993.
[6]Williams B M,Hoel L A. Modeling and forecasting vehicular traffic flowas a seasonal ARIMA process:theoretical basis and empirical results[J].Journal of Transportation Engineering,2003,129(6):664-672.
[7] ChingFu Chen,YuHern Chang,YuWei Chang. Seasonal ARIMA forecasting of inbound air travel arrivals to Taiwan[J]. Transportmetrica,2009,5(5):125-140.
[8]Cheng Cheng,Du Yu-chuan,Liu Xin. A forecast model of traffic hub traffic considering holiday effect[J]. Journal of Transportation Systems Engineering and Information,2015,15(5):202-207.
[9] Kavousi-Fard A,Khosravi A,Nahavandi S. A newfuzzy-based combined prediction interval for wind power forecasting[J]. IEEE Transactions on Power Systems,2016,31(1):18-26.
[10]Fang Li-jun,Ye Xia-fei,Ming Rui-li. A comparative analysis of the development trend of urban rail transit in Shanghai Seoul and[J].Journal of Traffic and Transportation,2007,(1):105-107.
[11]Chen Z,Deng T,Gao L,et al. A novel spatial-temporal detection method of dim infrared moving small target[J]. Infrared Physics&Technology,2014,66(9):84-96.
[12]Engle R F. Dynamic conditional beta[J]. Journal of Financial Econometrics,2016,14(4):643-667.
[13]Nahmias Biran B,Shiftan Y. Towards a more equitable distribution of resources:using activity-based models and subjective well-being measures in transport project evaluation[J]. Transportation Research Part A:Policy and Practice,2016,94:672-684.
[14]Varagouli ESmith B L,Williams B M,Oswald R K. Comparison of parametric and nonparametric models for traffic flowforecasting[J]. Transportation Research Part C Emerging Technologies,2002,10(4):303-321.
[15]G,Simos T E,Xeidakis G S. Fitting a multiple regression line to travel demand forecasting:The case of the prefecture of Xanthi,northern greece[J]. Mathematical&Computer Modelling,2005,42(7-8):817-836.
[16] Abdel-Aty MA,Pemmanaboina R. Calibrating a real-time traffic crash-prediction model using archived weather and ITS traffic data[J]. IEEE Transactions on Intelligent Transportation Systems,2006,7(2):167-174.
[17] Hu W,Yan L,Liu K,et al. A short-term traffic flowforecasting method based on the hybrid PSO-SVR[J]. Neural Processing Letters,2016,43(1):155-172.
[18]Deng W,Li W,Yang X H. A novel hybrid optimization algorithm of computational intelligence techniques for highway passenger volume prediction[J]. Expert Systems with Applications,2011,38(4):4198-4205.
[19] Dia H. An object-oriented neural network approach to short-term traffic forecasting[J]. European Journal of Operational Research,2001,131(2):253-261.
[20]Xiao J M,Wang X H. Study on traffic flowprediction using RBF neural network[C]. International Conference on Machine Learning and Cybernetics,Guangzhou,China,2004:2672-2675.
[1]李国勇.智能预测控制及其MATLAB实现[M].北京:电子工业出版社,2010.
[3]叶小舟,陶飞飞,戚荣志,等.循环神经网络结构中激活函数的改进[J].计算机与现代化,2016,(12):29-33.
[8]成诚,杜豫川,刘新.考虑节假日效应的交通枢纽客流量预测模型[J].交通运输系统工程与信息,2015,15(5):202-207.
[10]方礼君,叶霞飞,明瑞利.上海、首尔、东京城市轨道交通客流发展趋势对比分析[J].交通与运输,2007,(1):105-107.