【作者】 张莹；

【导师】 荆涛；

【作者基本信息】 北京交通大学 ， 通信与信息系统， 2017， 硕士

【摘要】 在经济高速发展的今天,工业现代化带来的严重的大气污染已经影响到人类的生产生活,因此,环境监测成为热点,而无线传感器网络的应用为污染气体排放源的监测提供了极大的发展空间,由此人们对污染气体排放的实时监测技术要求越来越高。在此基础上,本文给出了与污染源排放监测算法相关的研究。无线传感器网络中分布有多个传感器节点,各个传感器节点在感知污染源信息时,会在节点处产生对同一污染源和不同污染源的浓度监测值,为了提高源强计算的准确度,本文提出改进的BP神经网络对这两种监测数据进行分类,并在得到同一污染源监测数据的基础上,采用自适应模拟退火算法优化源强计算,实现对污染源源强的实时监测。简而言之,本文主要介绍了基于改进BP神经网络的数据分类模型和自适应模拟退火算法及其改进算法,并将其应用到污染源源强反算的过程中。针对BP神经网络在数据分类方面的应用,本文提出了基于选择更新布谷鸟搜索(SUCS)算法的BP神经网络数据分类模型,并与具有局部最优解检测预防功能的BP算法(LMDBP)以及基于布谷鸟搜索算法的改进BP神经网络模型(BPCS)进行对比,以Iris数据集为基础,文章通过对这三个模型的训练以及测试,分别得出基于MATLAB平台的仿真结果,并根据该结果分析这三种算法的性能以及数据分类能力,总结出基于SUCS算法的BP神经网络的优点和缺点。此外,本文还将基于SUCS算法的BP网络模型与支持向量机(SVM)在数据分类方面进行了比较,得出该模型的分类能力有待提高的结论。在传感器监测数据经过基于SUCS算法的BP神经网络的分类后,本文对此分类结果结合高斯扩散模型进行源强反算以实现实时监测。文章在源强计算过程中,得出自适应模拟退火(ASA)算法可以计算污染源的瞬时排放速率的结论。同时文章针对提高ASA算法收敛速度这一点,提出了改进自适应模拟退火算法,并证明出该算法在收敛速度上优于ASA算法,但是在计算污染源连续排放速率方面,由于数据采集、传输过程的噪声影响以及分类模型的输出误差等因素,改进ASA算法无法作出精确运算甚至无法收敛。更多还原

【Abstract】 In the rapid economic development today,the serious air pollution caused by industrial modernization has already affected human’s production and life.Therefore,the environmental monitoring becomes the hot spot,while the application of wireless sensor network has provided the great development space for the monitoring of emission sources of pollutant gases,thus people’s requirements on the real-time monitoring technology of pollutant gases emission are getting higher and higher.On this basis,this paper gives the research on monitoring algorithm of pollution sources emission.In the wireless sensor network,there are many sensor nodes.When the sensor nodes perceive the information of pollution source,they will produce the concentration monitoring value of the same pollution source and different pollution sources at the nodes.In order to improve the accuracy of the source intensity calculation,this paper proposed the improved BP neural network to classify these two kinds of monitoring data,and based on the same pollution source monitoring data,this paper used adaptive simulated annealing algorithm to optimize the source intensity calculation,and achieved the real-time monitoring of pollution sources’ intensity.In short,this paper mainly introduced the data classification model based on improved BP neural network and adaptive simulated annealing(ASA)algorithm and improved algorithm about ASA,and applied these to the process of back-calculation of pollution sources’ intensity.Based on the application of BP neural network in data classification,this paper proposed the data classification model of BP neural network based on the selectively update cuckoo search(SUCS)algorithm,which was compared with BP algorithm with local minimum detection algorithm(LMDBP)and the improved BP neural network model based on the cuckoo search algorithm(BPCS).Based on the Iris dataset,this paper has obtained the simulation results based on MATLAB platform by respectively training and testing these three models,and this paper analyzed the performance and data classification ability of these three algorithms according to this results and summarized the advantages and disadvantages of BP neural network based on SUCS algorithm.In addition,the paper also compared the BP network model based on SUCS algorithm with support vector machine(SVM)in data classification,and obtained the conclusion that the classification ability of this model needs to be improved.After the sensor monitoring data were classified by the BP neural network based on SUCS algorithm,in this paper,the classification results were combined with the Gaussian diffusion model to carry out the back calculation of sources’ intensity to achieve real-time monitoring.In the process of sources intensity calculation,this paper got the conclusion that the adaptive simulated annealing algorithm can compute the instantaneous emission rate of the pollution sources.And at the same time,in order to improve the convergence speed of the ASA algorithm,this paper proposed the improved adaptive simulated annealing algorithm and proved that the improved ASA algorithm was superior to the ASA algorithm in convergence speed.However,in terms of computing the continuous emission rate of pollution sources,the improved ASA algorithm can not make accurate operation and can not even converge due to the influence of noise of data collection and the transmission process and the output error of the classification model.更多还原