基于增量式朴素贝叶斯分类方法的电梯交通模式识别方法的研究
|
摘要
在电梯群控系统中,能否有效的分析和处理电梯交通流数据是影响电梯群控系统性能的一个重要因素。因此对建筑物内部的交通状况进行准确分类,在不同的交通状况下采用不同的电梯群控策略,可以有效地提高电梯的服务质量和各项性能指标。目前最主要的是采用模糊神经网络来进行电梯交通模式识别,但是该方法由于算法训练耗时大、网络结构具有不可确定性、而且对训练数据的依赖性较大、泛化能力差,使得电梯交通模式识别的准确率不高。朴素贝叶斯因其条件属性和决策类别间关系清晰、分类速度快,并且具有良好的健壮性,已被成功应用到许多领域。当获得大量带有类别标注的样本代价较高时,与增量学习理论结合是解决问题的有效途径。因此,如何实现朴素贝叶斯分类与增量学习算法相结合应用于电梯交通模式识别是一个值得研究和解决的课题。
论文主要包括以下工作内容:
①在分析贝叶斯网络结构,特点和应用的基础上建立了一个朴素贝叶斯分类器模型。
②针对传统的增量算法重点介绍了一种新的增量序列学习算法,该算法引入一个分类损失权重系数λ,用于计算分类损失的大小,引入该系数的作用在于:充分利用先验知识对分类器进行优化;通过选择合理的学习序列强化了较完备数据对分类的积极影响,从而可以提高分类精度。
③将朴素贝叶斯分类器模型与改进后的增量序列学习算法相结合从而建立了一个基于增量式朴素贝叶斯分类模型。最后,在认真分析了电梯交通流的特点和规律的基础上将该模型运用于电梯交通的模式识别中,通过对电梯交通流数据进行采集分析和特征的提取,利用MATLAB进行了模拟测试,并对实验结果进行了比较分析,测试结果表明该方法对电梯交通模式识别的准确率为92.3%,相比于模糊神经网络的识别准确率90.6%有所提高,因此其分类性能更加令人满意。
通过定义并建立一种基于朴素贝叶斯分类器和增量序列学习算法相结合的分类模型,为实现电梯交通模式识别提供了一种有效的解决办法。由于办公大楼的客流规律比较明显,因此课题只是针对一般性的办公大楼的电梯交通流进行的研究,而对于像商场,普通居民住宅区那些客流规律并不是非常明显的建筑物该模式识别方法是否适用,有待进一步研究。
In the elevator group control system, the effectiveness of the analysis and processing elevator traffic flow data is an important factor that affects the performance of elevator group control system. Therefore, to carry out the accurate classification for the traffic situation inside the building and to adopt diverse elevator group control strategies in different traffic conditions can effectively improve the quality of service and performance index of elevators. At present, the most important method is the use of fuzzy neural networks for pattern recognition elevator traffic; however, because of large time-consuming in algorithm training, uncertain network structure, dependence on training data and poor generalization ability, this method makes the low accuracy in elevator traffic pattern recognition. Because of clear relationship between its condition attributes and decision-making in categories, higher speed of classification and stronger robustness, Naive Bayes has been successfully applied to many areas. When tagging a large number of categories with high price of the sample, combined with the incremental learning theory is an effective way to solve the problem. Therefore, how to integrate the Naive Bayesian classification and incremental learning algorithm and how to used the integrate model in an elevator traffic pattern recognition are subjects to be resolved.
The paper includes the following work:
①Analyzed the Bayesian network structure, characteristics and applications and builded a Naive Bayes classifier model.
②Introduced an incremental classification model of Naive Bayes then put forward a new sequence learning algorithm to make up the deficiency of such models, this algorithm introduced a classifying loss weight coefficientλfor each training instance in order to calculate the total classifying loss. After introducing the coefficient, the classifier was optimized by fully utilizing the prior knowledge; by means of choosing reasonable learning sequence, positive influence of the mature data on classification was strengthened, classification precision was improved.
③Naive Bayesian classifier model and improved incremental learning algorithm combining sequence and thus the establishment of a based on the incremental model of Naive Bayesian Classifier. In a careful analyzed of elevator traffic flow characteristics and patterns of the model based on the pattern recognition used in elevator traffic, through the collection of Elevator traffic flow data analyzed and feature extraction, used of MATLAB to carry out a simulation test, compared and analyzed the experimental results, the test results showed that the method of elevator traffic pattern recognition accuracy rate of 92.3%, compared to the identification of fuzzy neural network 90.6% had increased the accuracy, therefore the classification performance was more satisfactory.
Through the definition and establishment of a Naive Bayesian Classifier-based and incremental learning algorithm for sequence classification integration model, this provides an effective solution for the elevator traffic pattern recognition. Naive Bayesian Classifier model feature vectors among the various components are relatively independent compared with the decision-making variable, which means the class variables are the only parent node for each of the attribute variables, this network structure reduce the complexity of constructing compared with the five-layer structure of fuzzy neural network, which makes the classification speed and classification performance excelled the fuzzy neural network. Because the passenger traffic of the office block is more obvious, the subject only studied the general office of the elevator traffic flow, such as shopping malls and uptown, which are not very obvious of passenger traffic ,this method whether suitable or not still need to be solved in future.
论文主要包括以下工作内容:
①在分析贝叶斯网络结构,特点和应用的基础上建立了一个朴素贝叶斯分类器模型。
②针对传统的增量算法重点介绍了一种新的增量序列学习算法,该算法引入一个分类损失权重系数λ,用于计算分类损失的大小,引入该系数的作用在于:充分利用先验知识对分类器进行优化;通过选择合理的学习序列强化了较完备数据对分类的积极影响,从而可以提高分类精度。
③将朴素贝叶斯分类器模型与改进后的增量序列学习算法相结合从而建立了一个基于增量式朴素贝叶斯分类模型。最后,在认真分析了电梯交通流的特点和规律的基础上将该模型运用于电梯交通的模式识别中,通过对电梯交通流数据进行采集分析和特征的提取,利用MATLAB进行了模拟测试,并对实验结果进行了比较分析,测试结果表明该方法对电梯交通模式识别的准确率为92.3%,相比于模糊神经网络的识别准确率90.6%有所提高,因此其分类性能更加令人满意。
通过定义并建立一种基于朴素贝叶斯分类器和增量序列学习算法相结合的分类模型,为实现电梯交通模式识别提供了一种有效的解决办法。由于办公大楼的客流规律比较明显,因此课题只是针对一般性的办公大楼的电梯交通流进行的研究,而对于像商场,普通居民住宅区那些客流规律并不是非常明显的建筑物该模式识别方法是否适用,有待进一步研究。
In the elevator group control system, the effectiveness of the analysis and processing elevator traffic flow data is an important factor that affects the performance of elevator group control system. Therefore, to carry out the accurate classification for the traffic situation inside the building and to adopt diverse elevator group control strategies in different traffic conditions can effectively improve the quality of service and performance index of elevators. At present, the most important method is the use of fuzzy neural networks for pattern recognition elevator traffic; however, because of large time-consuming in algorithm training, uncertain network structure, dependence on training data and poor generalization ability, this method makes the low accuracy in elevator traffic pattern recognition. Because of clear relationship between its condition attributes and decision-making in categories, higher speed of classification and stronger robustness, Naive Bayes has been successfully applied to many areas. When tagging a large number of categories with high price of the sample, combined with the incremental learning theory is an effective way to solve the problem. Therefore, how to integrate the Naive Bayesian classification and incremental learning algorithm and how to used the integrate model in an elevator traffic pattern recognition are subjects to be resolved.
The paper includes the following work:
①Analyzed the Bayesian network structure, characteristics and applications and builded a Naive Bayes classifier model.
②Introduced an incremental classification model of Naive Bayes then put forward a new sequence learning algorithm to make up the deficiency of such models, this algorithm introduced a classifying loss weight coefficientλfor each training instance in order to calculate the total classifying loss. After introducing the coefficient, the classifier was optimized by fully utilizing the prior knowledge; by means of choosing reasonable learning sequence, positive influence of the mature data on classification was strengthened, classification precision was improved.
③Naive Bayesian classifier model and improved incremental learning algorithm combining sequence and thus the establishment of a based on the incremental model of Naive Bayesian Classifier. In a careful analyzed of elevator traffic flow characteristics and patterns of the model based on the pattern recognition used in elevator traffic, through the collection of Elevator traffic flow data analyzed and feature extraction, used of MATLAB to carry out a simulation test, compared and analyzed the experimental results, the test results showed that the method of elevator traffic pattern recognition accuracy rate of 92.3%, compared to the identification of fuzzy neural network 90.6% had increased the accuracy, therefore the classification performance was more satisfactory.
Through the definition and establishment of a Naive Bayesian Classifier-based and incremental learning algorithm for sequence classification integration model, this provides an effective solution for the elevator traffic pattern recognition. Naive Bayesian Classifier model feature vectors among the various components are relatively independent compared with the decision-making variable, which means the class variables are the only parent node for each of the attribute variables, this network structure reduce the complexity of constructing compared with the five-layer structure of fuzzy neural network, which makes the classification speed and classification performance excelled the fuzzy neural network. Because the passenger traffic of the office block is more obvious, the subject only studied the general office of the elevator traffic flow, such as shopping malls and uptown, which are not very obvious of passenger traffic ,this method whether suitable or not still need to be solved in future.
引文
[1] Schofield.A.J, Stonham.T.J, Mehta.P.A, Automated people counting to aid lift control, Automation in Construction [J], 1997, 5-6:437-445.
[2] Peter RD, Mehta P.,Haddon J, Lift passenger traffic patterns: applications, current knowledge and measurement",Elevator World [J], 2000, 48:87-94.
[3]陈海霞,苑淼淼.面向数据挖掘的分类器集成研究[D].吉林大学,2006.
[4] ALBERT T P, BEEBE J R, CHAN W L, et al. Elevator traffic pattern Recognition by artificial neural network [C] Proc of 1995 Elevation 1995 Elevator Engineers Conference. Hong Kong: Int Association of Elevator Engineers, 1995:122 - 131.
[5]宗群,尚晓光,岳有军.电梯群控系统的交通模式识别[J].控制与决策,2001,16:163-166.
[6]许玉格,罗飞.新型电梯群控系统交通模式识别方法[J].华南理工大学,2005.
[7]杨广全,朱昌明.基于粒子群K均值聚类算法的电梯的交通模式识别[J].上海交通大学,2007.
[8] Cooper G F, Herskovits E.A Bayesian method for the induction of probabilistic networks from data [J].Machine Learning, 1992, 9:309-347.
[9] Tom M.Mithell.机器学习[M].机械工业出版社,2005.9:112-136.
[10]史忠植著.知识发现[M].北京:清华大学出版社,2002.
[11]姚宏亮.贝叶斯网络结构学习机器多Agent系统模型研究[D].合肥工业大学硕士学位论文.合肥.2003.
[12] Judea Pearl, Bayesian Networks [M], Los Angeles: Congitive Systems Laboratory Computer Science Department University, 1994:168-184.
[13]胡振宇等.贝叶斯学习中基于贝叶斯判别分析的先验分布选取[J].计算机科学, 2003,8:32-36.
[14] Schlimmer, J.C., Fisher, D.A case study of incremental concept induction [C].In: Proceedings of the 5th National Conference on Artificial Intelligence, Menlo Park, CA: AAAI Press, 1986:496-501.
[15] Utgoff, P.E. Incremental Induction of decision trees [J].Machine Learning, 1989, 4:161-186.
[16] Crawford, S, L.Extensions to the CART algorithm [J].International Journal of Man-Machine Studies, 1989, 31:197-217.
[17] McSherry D.Strategic induction of decision trees [J]. Knowledge-Based Systems, 1999, 12:269-275.
[18] Garofalakis, M., Gehrke, J., Rastogi, R.Querying and mining data streams: you only get onelook [C].In VLDB02, Hong Kong, China, Aug.2002.
[19] Fern A and Givan R.Online ensemble learning: an empirical study [J].Machine LEARNING, 2003: 71-109.
[20] Hulten, G., Spencer, L., Domingos, P., Mining Time-Changing Data Stream [C].Proceedings of the Seventh International Conference on Knowledge Discovery and Data Mining.2001:97-106.
[21] Wang, E.H.and Kuh, A.,“A smart algorithm for incremental learning,”in Proc [C].Int.Joint Conf.Neural Netw., 1992:121–126.
[22] Higgins, C.H.and Goodman, and R.M.: Incremental learning with rulebased neural networks [C].in Proc.Int.Joint Conf.Neural Netw., 1991:875-880.
[23] Platt J.A resource allocating network for function interpolation [J].Neural Computation, 1991, 3(2): 213-225.
[24] Kadirkamanathan V., Nirajan M.A function estimation approach to sequential learning with Neural Networks [J].Neural Compuation, 1993, 5:954-975.
[25] Yamakawa H.Masumoto D, Kimoto T.Nakaga S.Active data selection and subsequent revision for sequential learning [C].IEICE Technical Report, 1993, NC92-99, 33-40.
[26] Zhang BT, An incremental learning algorithm that optimizes network size and sample size in one trial [C]. Proceedings of International Conference on Neural Networks.Orlando, USA, 1994:215-220.
[27] Williamson, J.R., Gaussian ARTMAP: A neural network for fast incremental learning of noisy multidimensional maps [J]. Neural Netw. 1996, 5: 881–897.
[28] Widmer, G., Kubat, M.: Effective learning in dynamic environments by explicit context tracking [C]. Proc.6th European Conf.on Machine Learning, LNCS667, 1993:227-243.
[29] McSherry D.Strategic induction of decision trees [J].Knowledge-Based Systems, 1999, 12(5-6): 269-275.
[30] SAMUEL K.Modern Bayesian statistics [M].George Washington University Press.2000:109.
[31]应晓槟,吴炜,滕奇志,杨晓敏,朱强军.基于Dirichlet分布的贝叶斯分类算法的手写数字字符识别.电子测量技术[J]. 2007,2.
[32]宫秀军,刘少辉,史忠植.一种增量贝叶斯分类模型[J].计算机学报, 2002,25(6):645-655
[33]曹燕飞.电梯群控系统中智能交通分析方法的研究[D].天津:天津大学电气自动化与能源工程学院,1999.
[34]朱德文.最新的电梯群控系统[J].中国电梯,1996,16(2):54-57.
[35] Qun Zong, Yiju Cheng, Junyuan Song, Yu Cai. Simulating traffic flow generator based on elevator traffic probability simulation model [J]. Elevator World, 2004, 52(4):130-13.
[36]林都,郭秀成.用最大隶属原则识别电梯客流模式[J].太原机械学院学报,1990,11(2):33-39.
[2] Peter RD, Mehta P.,Haddon J, Lift passenger traffic patterns: applications, current knowledge and measurement",Elevator World [J], 2000, 48:87-94.
[3]陈海霞,苑淼淼.面向数据挖掘的分类器集成研究[D].吉林大学,2006.
[4] ALBERT T P, BEEBE J R, CHAN W L, et al. Elevator traffic pattern Recognition by artificial neural network [C] Proc of 1995 Elevation 1995 Elevator Engineers Conference. Hong Kong: Int Association of Elevator Engineers, 1995:122 - 131.
[5]宗群,尚晓光,岳有军.电梯群控系统的交通模式识别[J].控制与决策,2001,16:163-166.
[6]许玉格,罗飞.新型电梯群控系统交通模式识别方法[J].华南理工大学,2005.
[7]杨广全,朱昌明.基于粒子群K均值聚类算法的电梯的交通模式识别[J].上海交通大学,2007.
[8] Cooper G F, Herskovits E.A Bayesian method for the induction of probabilistic networks from data [J].Machine Learning, 1992, 9:309-347.
[9] Tom M.Mithell.机器学习[M].机械工业出版社,2005.9:112-136.
[10]史忠植著.知识发现[M].北京:清华大学出版社,2002.
[11]姚宏亮.贝叶斯网络结构学习机器多Agent系统模型研究[D].合肥工业大学硕士学位论文.合肥.2003.
[12] Judea Pearl, Bayesian Networks [M], Los Angeles: Congitive Systems Laboratory Computer Science Department University, 1994:168-184.
[13]胡振宇等.贝叶斯学习中基于贝叶斯判别分析的先验分布选取[J].计算机科学, 2003,8:32-36.
[14] Schlimmer, J.C., Fisher, D.A case study of incremental concept induction [C].In: Proceedings of the 5th National Conference on Artificial Intelligence, Menlo Park, CA: AAAI Press, 1986:496-501.
[15] Utgoff, P.E. Incremental Induction of decision trees [J].Machine Learning, 1989, 4:161-186.
[16] Crawford, S, L.Extensions to the CART algorithm [J].International Journal of Man-Machine Studies, 1989, 31:197-217.
[17] McSherry D.Strategic induction of decision trees [J]. Knowledge-Based Systems, 1999, 12:269-275.
[18] Garofalakis, M., Gehrke, J., Rastogi, R.Querying and mining data streams: you only get onelook [C].In VLDB02, Hong Kong, China, Aug.2002.
[19] Fern A and Givan R.Online ensemble learning: an empirical study [J].Machine LEARNING, 2003: 71-109.
[20] Hulten, G., Spencer, L., Domingos, P., Mining Time-Changing Data Stream [C].Proceedings of the Seventh International Conference on Knowledge Discovery and Data Mining.2001:97-106.
[21] Wang, E.H.and Kuh, A.,“A smart algorithm for incremental learning,”in Proc [C].Int.Joint Conf.Neural Netw., 1992:121–126.
[22] Higgins, C.H.and Goodman, and R.M.: Incremental learning with rulebased neural networks [C].in Proc.Int.Joint Conf.Neural Netw., 1991:875-880.
[23] Platt J.A resource allocating network for function interpolation [J].Neural Computation, 1991, 3(2): 213-225.
[24] Kadirkamanathan V., Nirajan M.A function estimation approach to sequential learning with Neural Networks [J].Neural Compuation, 1993, 5:954-975.
[25] Yamakawa H.Masumoto D, Kimoto T.Nakaga S.Active data selection and subsequent revision for sequential learning [C].IEICE Technical Report, 1993, NC92-99, 33-40.
[26] Zhang BT, An incremental learning algorithm that optimizes network size and sample size in one trial [C]. Proceedings of International Conference on Neural Networks.Orlando, USA, 1994:215-220.
[27] Williamson, J.R., Gaussian ARTMAP: A neural network for fast incremental learning of noisy multidimensional maps [J]. Neural Netw. 1996, 5: 881–897.
[28] Widmer, G., Kubat, M.: Effective learning in dynamic environments by explicit context tracking [C]. Proc.6th European Conf.on Machine Learning, LNCS667, 1993:227-243.
[29] McSherry D.Strategic induction of decision trees [J].Knowledge-Based Systems, 1999, 12(5-6): 269-275.
[30] SAMUEL K.Modern Bayesian statistics [M].George Washington University Press.2000:109.
[31]应晓槟,吴炜,滕奇志,杨晓敏,朱强军.基于Dirichlet分布的贝叶斯分类算法的手写数字字符识别.电子测量技术[J]. 2007,2.
[32]宫秀军,刘少辉,史忠植.一种增量贝叶斯分类模型[J].计算机学报, 2002,25(6):645-655
[33]曹燕飞.电梯群控系统中智能交通分析方法的研究[D].天津:天津大学电气自动化与能源工程学院,1999.
[34]朱德文.最新的电梯群控系统[J].中国电梯,1996,16(2):54-57.
[35] Qun Zong, Yiju Cheng, Junyuan Song, Yu Cai. Simulating traffic flow generator based on elevator traffic probability simulation model [J]. Elevator World, 2004, 52(4):130-13.
[36]林都,郭秀成.用最大隶属原则识别电梯客流模式[J].太原机械学院学报,1990,11(2):33-39.