基于遗传算法的水质数据挖掘与应用研究

发布时间：2018-03-16 23:11

本文选题：水质评价分析　切入点：数据挖掘　出处：《燕山大学》2016年博士论文　论文类型：学位论文

【摘要】：随着我国沿海地区经济的高速发展和人口的快速增长,海洋资源被过度开发利用,近岸海域的水体污染和生态破坏现象不断加剧。为有效解决海洋生态问题,保障海洋生态的健康有序发展,对海洋生态环境进行动态实时监测非常必要。通过监测平台可以获取海量的监测数据,而传统的数理统计和经验预测等方法依靠于专家经验,监测数据没有被充分挖掘利用,因此通过先进算法来充分挖掘监测数据中的有用信息、准确评价海洋水体质量,并对海洋水质建立有效的分析和预警模型,是当前亟需解决的重要问题。依托特种光纤与光纤传感重点实验室的光电传感网监测的水质数据,在总结当前水质评价技术和研究现状的基础上,理论分析研究了水质监测数据挖掘、规则更新和水质分析预警模型等问题,通过仿真实验验证了算法的有效性,并应用于水质监测数据的挖掘分析与预警中。本文的主要研究内容如下:首先,为了充分挖掘利用海量的水质监测数据,研究了一种基于自适应免疫遗传算法的关联规则挖掘算法。将免疫算法引入经典遗传算法,设计适应度函数,改进免疫遗传算法的交叉变异概率,使之在整个执行过程中能够随着个体适应度连续变化,并将该算法应用到关联规则挖掘中,该方法在保证挖掘规则精度的同时,能够缩短挖掘时间。其次,为解决水质监测数据库随时间不断积累新数据而动态变化引起的规则维护问题,在自适应免疫遗传关联规则挖掘算法的基础上,研究了一种规则更新挖掘算法。针对属性数据库的规则更新挖掘,通过将部分原始数据和新增数据组成一个新的数据库来进行。计算原始和新增数据的差异度,从原始数据库中抽取不同比例的数据组建新的数据库,对新建数据库进行免疫遗传关联规则的挖掘。该算法不仅可以保留原有支持度高的规则,还能够发现新增规则。然后,针对水质监测参数的变化问题,为使变化的参数同样适用于BP神经网络分析预警模型,研究了一种基于免疫递阶遗传算法的BP神经网络优化算法。水质监测参数在实际的监测过程中会发生变化,而变化的参数影响BP神经网络的结构。通过引入免疫算法来提高算法处理速度,设计适合的适应度函数,利用递阶遗传算法的染色体分层特点,把染色体分为参数基因和控制基因,在传统遗传算法中加入解码操作,从而分别优化神经网络的结构和权值,调整遗传算子的交叉变异概率,使其在整个执行过程中能够自适应连续调整。该算法能够根据水质监测参数的动态变化而及时调整BP神经网络结构,使得分析预警模型更适用于当前参数的学习与分析并及时作出预警。最后,为了提高水质监测分析预警模型的精度,将基于自适应免疫递阶遗传算法的BP神经网络优化算法与贝叶斯正则化的LM-BP神经网络数值优化算法结合,研究了一种水质监测数据评价预警模型。通过免疫递阶遗传算法优化神经网络的参数和结构,再通过贝叶斯正则化LM-BP神经网络提高评价模型的精度。该模型适用于多参数大样本训练。
[Abstract]:With the rapid growth of the rapid development of economy in China's coastal areas and population, over exploitation of marine resources, coastal water pollution and ecological destruction phenomenon intensified. In order to effectively solve the problem of marine ecology, marine ecological protection of the healthy and orderly development, it is very necessary to carry out dynamic real-time monitoring of the marine environment monitoring data through. The monitoring platform can get mass, while the traditional mathematical statistics and empirical forecasting methods rely on expert experience, the monitoring data are not fully exploit, so through the advanced algorithm to dig out useful information in the monitoring data, the accurate evaluation of marine water quality, and establish effective analysis and early warning model of marine water quality, is an important the problem to be solved. Based on the water quality data of photoelectric sensor network monitoring special optical fiber and optical fiber sensing key laboratory, in summary Based on the current water quality evaluation technology and the research status, theoretical analysis on the water quality monitoring data mining, rule update and analysis of water quality early warning system model, through simulation experiments verify the effectiveness of the algorithm, mining analysis and early warning and applied to water quality monitoring data. The main contents of this paper are as follows: firstly, in order to fully the water quality monitoring data mining research, an association rule mining algorithm based on adaptive immune genetic algorithm. The immune algorithm is introduced into the classical genetic algorithm, fitness function, crossover and mutation probability improved immune genetic algorithm, so that the entire implementation process with individual fitness and the continuous change. The algorithm is applied to mining association rules, the method of mining rules while ensuring the accuracy, which can shorten the time of mining. Secondly, in order to solve the water quality monitoring data With the accumulation of new data base and dynamic change rules of maintenance of mining association rules in adaptive immune genetic algorithm based on the research, a rule updating algorithm. According to the attribute database update rule mining, for some of the original data and new data to form a new database. The difference calculation the original and new data, from the original database extraction in different proportions to form a new data mining database, immune genetic association rules of the new database. The algorithm can not only retain the original support rules of high, can also find new rules. Then, aiming at the problem of monitoring water quality parameters, as the parameters the change also applies to BP neural network analysis of early warning model, a BP neural network optimization algorithm immune genetic algorithm based on hierarchical water quality. The monitoring parameters will change in the monitoring of the actual process, and changes of parameters affecting the structure of BP neural network. To improve the algorithm processing speed by using the immune algorithm, suitable for the design of fitness function, using the layered characteristics of hierarchical genetic algorithm, the chromosome is divided into parameter gene and control gene, join in the traditional decoding operation the genetic algorithm, which optimize the structure and weights of neural network, adjusting crossover and mutation probability of genetic operator, which can adaptively adjust continuously throughout the entire implementation process. This algorithm can change dynamically according to the parameters of water quality monitoring and timely adjustment of the structure of BP neural network, learning and analysis which makes the analysis of early warning model is more suitable for the current parameter and timely warning. Finally, in order to improve the water quality monitoring early warning analysis of the accuracy of the model, based on hierarchical adaptive immune genetic algorithm BP neural network optimization algorithm and Bias regularization LM-BP neural network optimization algorithm combined with numerical method, studied a kind of monitoring data of water quality evaluation model. Through the structure and parameters of genetic algorithm to optimize neural network immune hierarchical, by Bias regularization LM-BP neural network to improve the accuracy of the evaluation model. The model is suitable for multi parameter large sample training.

【学位授予单位】：燕山大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：X832;TP18;TP311.13

【相似文献】