基于SVR的话务量预测模型研究
|
摘要
移动通信话务量预测问题一直是各大移动通信运营商关注的焦点,它对于掌握系统容量、网络扩容等情况具有直接的指导意义。本文采用SA-SVR预测话务量,并且结合资费水平应用到话务量预测模型当中,提高了预测精度。文章的主要内容包括:
1.分析并比较当前常用的各种时间序列分析和预测算法,用实验验证了本文采用模型的优势;
2.基于SVR坚实的理论支持和很强的非线性拟合回归能力,采用SVR模型对移动通信话务量时间序列进行预测;
3.分析讨论了在SVR预测模型中核参数选择问题的重要性以及各个核参数对模型的影响情况,比较了当前常用的几种SVR核参数选择方法,并且首次采用基于模拟退火算法进行核参数选择的SVR模型对话务量序列进行预测,取得了很好的预测效果;
4.通过实验讨论和分析了嵌入维数在时间序列分析预测过程中的重要性,并结合实验结果关于序列嵌入维数的选择得出结论;
5.首次提出基于资费-SASVR模型的话务量预测模型,它充分利用话务量数据与话务资费水平之间的关系提取出话务量序列的总趋势,对提取后的序列采用SASVR模型进行预测。实验结果表明此模型具有很好的预测效果。
The forecasting of mobile telephone traffic has always been the major mobile operators’focus, it is of directive meaning to master the system capacity and network expansion and so on. SA-SVR is utilized to forecast mobile traffic in this paper, what’s more, the tariff levels are combined in the model which improved the prediction accuracy. The contents are included in the article as follows:
1. The common algorithms of time series analysis and prediction are analyzed, and experiments prove the advantages of the model put forward in this paper;
2. Based on SVR’s solid theoretical support and a strong ability of nonlinear regression fitting, the SVR model is used to forecast mobile traffic time series;
3. This paper analyzes and discusses the influence of the various nuclear parameters selection methods of SVR on prediction model as well as the importance of selecting the nuclear parameters. Then the common methods of nuclear parameters selection are compared in this paper. For the first time, the traffic time series are predicted by the SVR mode with simulated annealing selecting nuclear parameters, which makes a very good prediction results;
4. The importance of embedded dimension in the process of time series analysis and forecasting is discussed and analyzed through experiments in this paper, then the conclusions on the sequence’s embedding dimension are drawn based on the experimental results;
5. The Tariff-SASVR model for traffic forecasting is proposed for the first time in this paper. It takes good advantages of the relationship between tariff level and traffic and then extracts the general trend of traffic. The residual sequence is predicted by SASVR. The results show that the model has a pretty good forecasting effect.
1.分析并比较当前常用的各种时间序列分析和预测算法,用实验验证了本文采用模型的优势;
2.基于SVR坚实的理论支持和很强的非线性拟合回归能力,采用SVR模型对移动通信话务量时间序列进行预测;
3.分析讨论了在SVR预测模型中核参数选择问题的重要性以及各个核参数对模型的影响情况,比较了当前常用的几种SVR核参数选择方法,并且首次采用基于模拟退火算法进行核参数选择的SVR模型对话务量序列进行预测,取得了很好的预测效果;
4.通过实验讨论和分析了嵌入维数在时间序列分析预测过程中的重要性,并结合实验结果关于序列嵌入维数的选择得出结论;
5.首次提出基于资费-SASVR模型的话务量预测模型,它充分利用话务量数据与话务资费水平之间的关系提取出话务量序列的总趋势,对提取后的序列采用SASVR模型进行预测。实验结果表明此模型具有很好的预测效果。
The forecasting of mobile telephone traffic has always been the major mobile operators’focus, it is of directive meaning to master the system capacity and network expansion and so on. SA-SVR is utilized to forecast mobile traffic in this paper, what’s more, the tariff levels are combined in the model which improved the prediction accuracy. The contents are included in the article as follows:
1. The common algorithms of time series analysis and prediction are analyzed, and experiments prove the advantages of the model put forward in this paper;
2. Based on SVR’s solid theoretical support and a strong ability of nonlinear regression fitting, the SVR model is used to forecast mobile traffic time series;
3. This paper analyzes and discusses the influence of the various nuclear parameters selection methods of SVR on prediction model as well as the importance of selecting the nuclear parameters. Then the common methods of nuclear parameters selection are compared in this paper. For the first time, the traffic time series are predicted by the SVR mode with simulated annealing selecting nuclear parameters, which makes a very good prediction results;
4. The importance of embedded dimension in the process of time series analysis and forecasting is discussed and analyzed through experiments in this paper, then the conclusions on the sequence’s embedding dimension are drawn based on the experimental results;
5. The Tariff-SASVR model for traffic forecasting is proposed for the first time in this paper. It takes good advantages of the relationship between tariff level and traffic and then extracts the general trend of traffic. The residual sequence is predicted by SASVR. The results show that the model has a pretty good forecasting effect.
引文
[1]Box GEP, Jenkins GM. Time series analysis, forecasting and control[M]. San Francisco: Holden Day, 1970:2-95.
[2]Wang Jian-xin, Xiao Xue-feng,Ye Jin. An Analysis of Traffic Load Prediction Base on Auto Regressive Model in Small Time Granularity[C]. Changsha: Communications, Circuits and Systems Proceedings, 2006:1727-1731.
[3]Javier Contreras, Rosario Espínola, Francisco J. Nogales. ARIMA Models to Predict Next-day Electricity Prices[J]. IEEE Trans. on Power Systems.2003, 18(3): 1014-1020.
[4]池启水,刘晓雪. ARIMA模型在煤炭消费预测中的应用分析[J].能源研究与信息.2007, 23(2):117-123.
[5]HOPTROFF R G. The principles and practice of time series forecasting and business modeling using neural nets[J]. Neural Computing and Applications.1993, 1:59-66.
[6]王悦.人工神经网络在经济预测中的应用[J].Value Engineering.2007, 5:90-93.
[7]马千里,郑启伦,彭宏,钟谭卫.基于相空间重构理论与递归神经网络相结合的股票短期预测方法[J].计算机应用研究.2007, 24(4):239-245.
[8]王燚,郭伟.一种混沌神经网络的混沌时间序列预测[J].电子测量与仪器学报.2007, 21(6):5-9.
[9]陶乃勇,蒋建忠,昃草心.神经网络在移动话务量预测中的应用研究[J].山东通信技术.2008, 28(1):9-12.
[10]A.J.Wang, B.Ramsay. A neural network based estimator for electricity spot-pricing with particular reference to weekend and public holidays[J]. Neurocomputing.1998, 23:47-57.
[11]V.N.Vapnik. The Nature of Statistical Learning Theory[M].New York:Springer,1995.
[12]邓乃扬,田英杰.数据挖掘新方法-支持向量机[M].北京:北京科学出版社,2004.12-240.
[13]Alex J. Smola, Bernhard Scho¨lkopf. A Tutorial on Support Vector Regression[J]. Statistics and Computing.2004, 14(3):199-222.
[14]Chun-Hsin Wu,Jan-Ming Ho,D.T.Lee. Travel-time Prediction with Support VectorRegression[J].IEEE Trans. on Intelligent Transportation Systems.2004, 5(4): 276-281.
[15]辛治运,顾明.基于最小二乘支持向量机的复杂金融时间序列预测[J].清华大学学报(自然科学版).2008, 48(7):1147-1149.
[16]O. Chapelle,V. Vapnik,O. Bousquet, S. Mukherjee. Choosing multiple parameters for support vector machines[J].Mach. Learn. 2002, 46:131–159.
[17]L.J.Cao, F.E.H.Tay. Support vector machine with adaptive parameters in financial time series forecasting[J].IEEE Trans. Neural Networks.2003,14 (6):1506–1518.
[18]Alexander Popov,Alexander Sautin. Selection of Support Vector Machines Parameters for Regression using Nested Grids[C]. Novosibirsk:Strategic Technologies, 2008.329-331.
[19]Chen Peng-Wei,Wang Jung-Ying,Lee Hahn-Ming. Model Selection of SVMs using GA Approach[C]. Taipei:Neural Networks, Proceedings,2004.2035-2040.
[20]Shukuan Lin,Shaomin Zhang,Jianzhong Qiao,Hualei Liu,Ge Yu. A Parameter Choosing Method of SVR for Time Series Prediction[C].Shenyang:Young Computer Scientists, 2008:130-135.
[21]Masayuki Karasuyama, Ryohei Nakano .Optimizing SVR Hyperparameters via Fast Cross-Validation using AOSVR[C]. Orlando:Proceedings of International Joint Conference on Neural Networks, 2007.12-17.
[22]S.Sathiya Keerthi. Efficient Tuning of SVM Hyperparameters Using Radius/Margin Bound and Iterative Algorithms[J].IEEE TRANSACTIONS ON NEURAL NETWORKS. 2002, 13(5):1225-1229.
[23]Pai Ping-Feng,Hong Wei-Chiang.Support Vector Machines with Simulated Annealing Algorithms in Electricity Load Forecasting[J]. Energy Conversion and Management.2005, 46: 2669-2688.
[24]P.-F.Pai,W.-C.Hong.Software Reliability Forecasting by Support Vector Machines with Simulated Annealing Algorithms[J]. Syst.Softw.2006, 79:747–755.
[25]S.-W.Lin,Z.-J.Lee,S.-C.Chen,T.-Y.Tseng. Parameter Determination of Support Vector Machine and Feature Selection using Simulated Annealing Approach[J].Applied Soft Computing.2008,8:1505–1512.
[26]Jin-Tsong Jeng.Hybrid Approach of Selecting Hyperparameters of Support Vector Machine for Regression[J].IEEE TRANSACTIONS ON SYSTEMS,MAN,AND CYBERNETICS-PART B:CYBERNETICS. 2006, 36(3):699-709.
[27]Jie Li,Feng Gao,Xiaohong Guan,Hui Xu. SVR Kernel Parameters Selection Based on Steady-State Genetic Algorithm[C].Dalian: Proceedings of the 6th World Congress on Intelligent Control and Automation, 2006.21-23.
[28] Le-Peng Bi,Hao Huang,Zhi-Yun Zheng,Han-Tao Song. NEW HEURISTIC FOR DETERMINATION GAUSSIAN KERNEL'S PARAMETER[C].Guangzhou:Proceedings of the Fourth International Conference on Machine Learning and Cybernetics, 2005.18-21.
[29]Kaibo Duan, S.Sathiya Keerthi, Aun Neow Poo. Evaluation of simple performance measures for tuning SVM hyperparameters[J]. Neurocomputing.2003, 51:41–59.
[30] Wenjian Wang,Zongben Xu,Weizhen Lu,Xiaoyun Zhang.Determination of the spread parameter in the Gaussian kernel for classification and regression[J].Neurocomputing. 2003, 55:643–663.
[31]Yulin Dong,Zhonghang Xia,Manghui Tu,Guangming Xing.An Optimization Method for Selecting Parameters in Support Vector Machines[C]. Proceedings of the 6th International Conference on Machine Learning and Applications,2007.1-6.
[32]Holger Frohlich,Andreas Zell.Efficient Parameter Selection for Support Vector Machines in Classification and Regression via Model-Based Global Optimization[C]. Montreal:Proceedings of International Joint Conference on Neural Networks,2005.1431-1436.
[33]Ingo Steinwart.On the Optimal Parameter Choice forνSupport Vector Machines[J]. IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE.2003, 25(10):1274-1284.
[34]Fan Wang,Guozhen Tan,Chao Deng and Zhu Tian.Real-time Traffic Flow Forecasting Model and Parameter Selection based on-SVR[C]. Chongqing: Proceedings of the 7th World Congress on Intelligent Control and Automation, 2008.25-27.
[35]G.Peter Zhang.Time Series Forecasting using a Hybrid ARIMA and Neural Network Model[J]. Neurocomputing.2003, 50:159–175.
[36]Ping-Feng Pai, Chih-Sheng Lin.A hybrid ARIMA and support vector machines model in stock price forecasting[J].Omega. 2005,33: 497–505.
[37]DilliRAryal, Yao-wu Wang.Time-series analysis with a hybrid Box-Jenkins ARIMA and neural net work mode[J]. Journal of Harbin Institute of Technology (New Series).2004, 11(4):413-421.
[38]S.L.Hoa,M.Xieb,T.N.Goh .A comparative study of neural network and Box-Jenkins ARIMA modeling in time series prediction[J].Computers &Industrial Engineering.2002,42:371-375.
[39]李军,赵峰.基于支持向量回归神经网络的时间序列预测[J].系统仿真学报.2008, 20(15):4025-4030.
[40]N.Wagner, Z.Michalewicz, M.Khouja, R.R.McGregor. Time Series Forecasting for Dynamic Environments: The DyFor Genetic Program Model[J].IEEE TRANSACTIONS ON EVOLUTIONARY COMPUTATION.2007, 11(4):433-452.
[41]林耀进,周忠眉,吴顺祥.集成灰色支持向量机预测模型研究与应用[J].计算机应用. 2009, 29(12):3287-3289.
[42]王晓兰,王明伟.基于小波分解和最小二乘支持向量机的短期风速预测[J].电网技术. 2010, 34(1):179-184.
[43]王燕.应用时间序列分析[M].北京:中国人民大学出版社,2005.16-207.
[44]向炳新.针对电信行业自适应智能话务预测模型的研究与实现[D].成都:电子科技大学,2007.
[45]韩路跃,杜行检.基于MATLAB的时间序列建模与预测[J].计算机仿真. 2005, 22(4):105-182.
[46]段晓雷.数据挖掘技术在移动通信网话务分析中的应用研究[D].北京:北京工业大学,2007.
[47]周灿文.基于递归神经网络的时间序列话务量预测方法研究[D].广州:华南理工大学,2006.
[48]陈晓峰.电信话务量OLAP分析和神经网络预测模型[D].厦门:华侨大学,2004.
[49]肖建,于龙,白裔峰.支持向量回归中核函数和超参数选择方法综述[J].西南交通大学学报. 2008,Vol.43(3):297-303.
[50]刘斌.支持向量机及其在信号处理中的应用[D].大庆:大庆石油学院,2006.
[2]Wang Jian-xin, Xiao Xue-feng,Ye Jin. An Analysis of Traffic Load Prediction Base on Auto Regressive Model in Small Time Granularity[C]. Changsha: Communications, Circuits and Systems Proceedings, 2006:1727-1731.
[3]Javier Contreras, Rosario Espínola, Francisco J. Nogales. ARIMA Models to Predict Next-day Electricity Prices[J]. IEEE Trans. on Power Systems.2003, 18(3): 1014-1020.
[4]池启水,刘晓雪. ARIMA模型在煤炭消费预测中的应用分析[J].能源研究与信息.2007, 23(2):117-123.
[5]HOPTROFF R G. The principles and practice of time series forecasting and business modeling using neural nets[J]. Neural Computing and Applications.1993, 1:59-66.
[6]王悦.人工神经网络在经济预测中的应用[J].Value Engineering.2007, 5:90-93.
[7]马千里,郑启伦,彭宏,钟谭卫.基于相空间重构理论与递归神经网络相结合的股票短期预测方法[J].计算机应用研究.2007, 24(4):239-245.
[8]王燚,郭伟.一种混沌神经网络的混沌时间序列预测[J].电子测量与仪器学报.2007, 21(6):5-9.
[9]陶乃勇,蒋建忠,昃草心.神经网络在移动话务量预测中的应用研究[J].山东通信技术.2008, 28(1):9-12.
[10]A.J.Wang, B.Ramsay. A neural network based estimator for electricity spot-pricing with particular reference to weekend and public holidays[J]. Neurocomputing.1998, 23:47-57.
[11]V.N.Vapnik. The Nature of Statistical Learning Theory[M].New York:Springer,1995.
[12]邓乃扬,田英杰.数据挖掘新方法-支持向量机[M].北京:北京科学出版社,2004.12-240.
[13]Alex J. Smola, Bernhard Scho¨lkopf. A Tutorial on Support Vector Regression[J]. Statistics and Computing.2004, 14(3):199-222.
[14]Chun-Hsin Wu,Jan-Ming Ho,D.T.Lee. Travel-time Prediction with Support VectorRegression[J].IEEE Trans. on Intelligent Transportation Systems.2004, 5(4): 276-281.
[15]辛治运,顾明.基于最小二乘支持向量机的复杂金融时间序列预测[J].清华大学学报(自然科学版).2008, 48(7):1147-1149.
[16]O. Chapelle,V. Vapnik,O. Bousquet, S. Mukherjee. Choosing multiple parameters for support vector machines[J].Mach. Learn. 2002, 46:131–159.
[17]L.J.Cao, F.E.H.Tay. Support vector machine with adaptive parameters in financial time series forecasting[J].IEEE Trans. Neural Networks.2003,14 (6):1506–1518.
[18]Alexander Popov,Alexander Sautin. Selection of Support Vector Machines Parameters for Regression using Nested Grids[C]. Novosibirsk:Strategic Technologies, 2008.329-331.
[19]Chen Peng-Wei,Wang Jung-Ying,Lee Hahn-Ming. Model Selection of SVMs using GA Approach[C]. Taipei:Neural Networks, Proceedings,2004.2035-2040.
[20]Shukuan Lin,Shaomin Zhang,Jianzhong Qiao,Hualei Liu,Ge Yu. A Parameter Choosing Method of SVR for Time Series Prediction[C].Shenyang:Young Computer Scientists, 2008:130-135.
[21]Masayuki Karasuyama, Ryohei Nakano .Optimizing SVR Hyperparameters via Fast Cross-Validation using AOSVR[C]. Orlando:Proceedings of International Joint Conference on Neural Networks, 2007.12-17.
[22]S.Sathiya Keerthi. Efficient Tuning of SVM Hyperparameters Using Radius/Margin Bound and Iterative Algorithms[J].IEEE TRANSACTIONS ON NEURAL NETWORKS. 2002, 13(5):1225-1229.
[23]Pai Ping-Feng,Hong Wei-Chiang.Support Vector Machines with Simulated Annealing Algorithms in Electricity Load Forecasting[J]. Energy Conversion and Management.2005, 46: 2669-2688.
[24]P.-F.Pai,W.-C.Hong.Software Reliability Forecasting by Support Vector Machines with Simulated Annealing Algorithms[J]. Syst.Softw.2006, 79:747–755.
[25]S.-W.Lin,Z.-J.Lee,S.-C.Chen,T.-Y.Tseng. Parameter Determination of Support Vector Machine and Feature Selection using Simulated Annealing Approach[J].Applied Soft Computing.2008,8:1505–1512.
[26]Jin-Tsong Jeng.Hybrid Approach of Selecting Hyperparameters of Support Vector Machine for Regression[J].IEEE TRANSACTIONS ON SYSTEMS,MAN,AND CYBERNETICS-PART B:CYBERNETICS. 2006, 36(3):699-709.
[27]Jie Li,Feng Gao,Xiaohong Guan,Hui Xu. SVR Kernel Parameters Selection Based on Steady-State Genetic Algorithm[C].Dalian: Proceedings of the 6th World Congress on Intelligent Control and Automation, 2006.21-23.
[28] Le-Peng Bi,Hao Huang,Zhi-Yun Zheng,Han-Tao Song. NEW HEURISTIC FOR DETERMINATION GAUSSIAN KERNEL'S PARAMETER[C].Guangzhou:Proceedings of the Fourth International Conference on Machine Learning and Cybernetics, 2005.18-21.
[29]Kaibo Duan, S.Sathiya Keerthi, Aun Neow Poo. Evaluation of simple performance measures for tuning SVM hyperparameters[J]. Neurocomputing.2003, 51:41–59.
[30] Wenjian Wang,Zongben Xu,Weizhen Lu,Xiaoyun Zhang.Determination of the spread parameter in the Gaussian kernel for classification and regression[J].Neurocomputing. 2003, 55:643–663.
[31]Yulin Dong,Zhonghang Xia,Manghui Tu,Guangming Xing.An Optimization Method for Selecting Parameters in Support Vector Machines[C]. Proceedings of the 6th International Conference on Machine Learning and Applications,2007.1-6.
[32]Holger Frohlich,Andreas Zell.Efficient Parameter Selection for Support Vector Machines in Classification and Regression via Model-Based Global Optimization[C]. Montreal:Proceedings of International Joint Conference on Neural Networks,2005.1431-1436.
[33]Ingo Steinwart.On the Optimal Parameter Choice forνSupport Vector Machines[J]. IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE.2003, 25(10):1274-1284.
[34]Fan Wang,Guozhen Tan,Chao Deng and Zhu Tian.Real-time Traffic Flow Forecasting Model and Parameter Selection based on-SVR[C]. Chongqing: Proceedings of the 7th World Congress on Intelligent Control and Automation, 2008.25-27.
[35]G.Peter Zhang.Time Series Forecasting using a Hybrid ARIMA and Neural Network Model[J]. Neurocomputing.2003, 50:159–175.
[36]Ping-Feng Pai, Chih-Sheng Lin.A hybrid ARIMA and support vector machines model in stock price forecasting[J].Omega. 2005,33: 497–505.
[37]DilliRAryal, Yao-wu Wang.Time-series analysis with a hybrid Box-Jenkins ARIMA and neural net work mode[J]. Journal of Harbin Institute of Technology (New Series).2004, 11(4):413-421.
[38]S.L.Hoa,M.Xieb,T.N.Goh .A comparative study of neural network and Box-Jenkins ARIMA modeling in time series prediction[J].Computers &Industrial Engineering.2002,42:371-375.
[39]李军,赵峰.基于支持向量回归神经网络的时间序列预测[J].系统仿真学报.2008, 20(15):4025-4030.
[40]N.Wagner, Z.Michalewicz, M.Khouja, R.R.McGregor. Time Series Forecasting for Dynamic Environments: The DyFor Genetic Program Model[J].IEEE TRANSACTIONS ON EVOLUTIONARY COMPUTATION.2007, 11(4):433-452.
[41]林耀进,周忠眉,吴顺祥.集成灰色支持向量机预测模型研究与应用[J].计算机应用. 2009, 29(12):3287-3289.
[42]王晓兰,王明伟.基于小波分解和最小二乘支持向量机的短期风速预测[J].电网技术. 2010, 34(1):179-184.
[43]王燕.应用时间序列分析[M].北京:中国人民大学出版社,2005.16-207.
[44]向炳新.针对电信行业自适应智能话务预测模型的研究与实现[D].成都:电子科技大学,2007.
[45]韩路跃,杜行检.基于MATLAB的时间序列建模与预测[J].计算机仿真. 2005, 22(4):105-182.
[46]段晓雷.数据挖掘技术在移动通信网话务分析中的应用研究[D].北京:北京工业大学,2007.
[47]周灿文.基于递归神经网络的时间序列话务量预测方法研究[D].广州:华南理工大学,2006.
[48]陈晓峰.电信话务量OLAP分析和神经网络预测模型[D].厦门:华侨大学,2004.
[49]肖建,于龙,白裔峰.支持向量回归中核函数和超参数选择方法综述[J].西南交通大学学报. 2008,Vol.43(3):297-303.
[50]刘斌.支持向量机及其在信号处理中的应用[D].大庆:大庆石油学院,2006.