曲线拟合预测模型及算法在水质远程监测系统中的研究

发布时间：2018-04-08 15:26

本文选题：水质远程在线监测　切入点：曲线拟合　出处：《浙江理工大学》2017年硕士论文

【摘要】：目前以水质分析仪为主的水质分析设备虽备受青睐,但在实现远程在线监测以及水样(污水)毒性物质成分与含量的分析方面仍缺失,同时,监测系统存在采集终端高并发的数据访问、数据标准异构导致难以实现数据流快速响应和实时处理等问题。本文在研制出的水质分析仪的基础上,搭建了水质远程在线监测系统,该系统以对水样中毒性物质成分和浓度的识别为核心目的,并根据数据流动过程分别研究数据采集、数据处理和数据存储相关设计和优化技术,实现对未知水样的远程实时监测的同时提高系统的通信性能与数据处理能力。本文主要完成了以下几项工作:1、水质远程在线监测系统的搭建。针对目前水质的远程在线监测系统以及对水质毒性物质成分和浓度预测方面的短缺问题,设计并搭建了一种水质远程在线监测系统,利用发光细菌发光原理,以明亮发光杆菌3变种作为毒性测试物种,实现了对水样中毒性物质成分和浓度的识别、水质分析仪的远程管理与在线监测。2、曲线拟合模型构建与特征提取。针对使用常用的曲线拟合函数对反应机理曲线进行拟合时,出现的变质现象及拟合精度不高等问题,提出了一种基于改进的B样条曲线拟合算法,解决了在模型构建时因追求拟合精度而违背毒性物质对发光细菌抑制性作用等问题,并将拟合后的模型参数结合毒性物质属性设定为特征向量。3、毒性物质成分和浓度的识别。由于数据的冗余性,对提取的特征向量使用PCA和LDA算法进行降维处理,并结合BP神经网络对未知样本中的毒性物质成分进行识别处理,后针对使用BP神经网络训练时所需要的迭代次数多、收敛速度慢以及容易出现在未达到训练目标时训练终止等问题,提出了基于改进的BP神经网络算法对毒性物质成分和浓度进行识别,有效地改善了算法的性能,且LDA与改进后的BP网络模型相结合对毒性物质成分的识别正确率达到了100%,浓度识别率达到了92%以上。4、对水质远程在线监测系统的优化与研究。针对系统实际应用中遇到的高并发访问请求、数据的实时性能、数据传输效率以及数据包的封装和解析速率低等问题,分别对数据采集、数据处理以及数据存储三个方面进行优化。面对高速率数据访问请求时,在IOCP模型的基础上,提出了基于对象池模式的自适应线程池技术,有效的解决了对共享资源的并发访问效率低的问题,提高了系统的通信效率。针对数据传输效率以及数据包的封装和解析速率,提出了基于JSON和TLV的消息格式优化。最后针对数据的实时性和系统的使用率等问题,提出了高效数据流处理算法,对数据库管理技术进行改进。5、将以上研究成果用VC++语言实现,按一定的逻辑集成为数据处理模块,加入到水质远程在线监测系统的服务器端,使采集到的数据得到了利用,最后通过系统的实现和实际应用验证了本文提出的曲线拟合构建、毒性物质成分和识别的方法。研究结果表明,本文提出的基于改进的B样条曲线拟合方法弥补了常见拟合函数的拟合精度不佳和“变质失性”问题,提高了曲线拟合效果,为后续的特征提取工作提供了有利的支撑;基于改进的BP神经网络算法,弥补了传统BP神经网络的不足,提高了毒性物质成分和浓度的识别正确率;基于完成端口模型的网络通信性能的优化,解决了在实际应用中遇到的并发处理能力不强、数据传输效率低以及数据包的封装和解析速率低等问题,使系统的稳定性、性能得到了改善。
[Abstract]:Water quality analysis equipment at present water quality analyzer mainly is favored, but in the realization of remote online monitoring and water (sewage) analysis of ingredients and contents of toxic substances are still missing, at the same time, there are high concurrent access to the data acquisition terminal monitoring system, which leads to the problem of heterogeneous data standards difficult to achieve fast response and real-time data stream processing. Based on the water quality analyzer developed on set up a remote on-line monitoring system, the system to identify the constituents and concentration of water poisoning is the core purpose, and according to the data flow process were studied in data acquisition, data processing and data storage design and optimization technology to improve system's communication performance and the data processing ability of unknown samples in the remote real-time monitoring at the same time. This paper mainly completes the following work: 1, remote online monitoring of water quality To build the test system. At present, remote online monitoring system of water quality and water quality of toxic substance composition and concentration prediction of shortage, designed and built a water quality remote monitoring system, using the principle of luminescent bacteria, to Photobacterium phosphoreum toxicity test as 3 varieties of species, the identification of the material composition and concentration for water poisoning, remote management and online monitoring of.2 water quality analyzer, construction and feature extraction of curve fitting model. According to the curve fitting function is used to fit the curve of reaction mechanism, metamorphic phenomenon and the fitting accuracy is not high, we propose an improved B algorithm based on spline curve fitting, solving the contrary problem of toxic substances on the luminescent bacteria inhibition effect due to the pursuit of the fitting accuracy in the model construction, and combining the model parameters after fitting Toxic property is set to feature vector.3, identification of material composition and concentration of toxicity. Because of the redundancy of data, the dimensionality reduction process using PCA and LDA algorithm for feature extraction, and combining with the BP neural network recognition processing of toxic substances in the composition of unknown samples, the number of iterations for the use of BP neural network training when needed, slow convergence and easy to appear in the training target training does not reach the termination problem, put forward the improved BP neural network algorithm based on the composition and concentration of toxic substances are identified, effectively improve the performance of the algorithm, and the BP LDA network model and the improved combination of identification of toxic substances composition of the correct rate reached 100%, the concentration of recognition rate reached more than 92%.4, optimization and Research on remote monitoring system of water quality. The high concurrency encountered in practical application systems The access request, the real-time performance of data, the efficiency of data transmission and data packet encapsulation and resolution rate low, respectively on the three aspects of data acquisition, data processing and data storage optimization. With high speed data access request, based on the IOCP model, proposed an adaptive thread pool technique based on Object Pool Pattern and effectively solves the problem of low efficiency of concurrent access to shared resources, improve the efficiency of communication system. The efficiency of data transmission and data packet encapsulation and resolution rate, put forward the optimization of message format JSON and based on TLV. Finally according to the real-time data and system usage, put forward the flow efficient data processing algorithm, improved.5 database management technology, the research results with the VC++ language, according to certain logic integrated data processing module is added to the water. Quality of remote online monitoring system server, the collected data are utilized, and finally through the system implementation and practical application verify the curve fitting of this construction method, toxic substance composition and identification. The results show that the proposed method improved the B spline curve fitting based on offset fitting accuracy the common fitting function is not good and the "lost" problem of deterioration, improve the effect of curve fitting, provide a favorable support for the subsequent feature extraction; improved BP neural network algorithm based on traditional BP neural network, improve the recognition of material composition and concentration of the toxicity of the correct rate of network performance optimization; communication completion port model based on solving the concurrent processing capability encountered in the practical application is not strong, low efficiency of data transmission and data packet encapsulation and resolution rate etc. The problem is that the stability of the system and the performance of the system have been improved.

【学位授予单位】：浙江理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R123.1;TP274

【参考文献】