基于机器学习理论的水质预测技术研究

发布时间：2018-03-07 01:17

  本文选题：水质预测　切入点：支持向量机回归　出处：《浙江师范大学》2015年硕士论文　论文类型：学位论文

【摘要】：水质预测是水资源管理和污染控制的基础性工作,准确预测水体中污染物浓度随时间发展变化的趋势至关重要。目前国内外有多种水质预测方法,但这些方法仍存在一些缺点。本文讨论了四种水质预测模型,分别为支持向量回归水质预测模型、关联向量机水质预测模型、极限学习机水质预测模型以及深度信念网络水质预测模型。本文在建立支持向量回归水质预测模型时,采用了生物地理学优化算法确定支持向量机的控制变量,并以该水质预测模型对PH值、溶解氧、高锰酸盐指数和氨氮四种重要水质指标进行预测。采用国家环保部发布的四川攀枝花龙洞水质监测时间序列数据进行实验,并与支持向量机的传统控制变量寻优方法进行比较,结果表明改进生物地理学寻优方法建立的水质预测模型效果较好。支持向量机水质预测模型中存在一些问题,如核函数必须满足Mercer条件,支持向量的个数会随着训练样本的增加呈线性增加,且只给出确定性的预测结果,没有概率输出,无法估计预测的不确定性。在此基础上本文提出了一种基于关联向量机回归的水质时间序列预测模型,并对该模型的有效性进行了验证；然后将关联向量机回归预测模型与支持向量机回归预测模型进行比较。为了比较不同核函数的预测效果,实验中预测模型的核函数分别采用了线性函数和高斯函数,并且在应用关联向量机回归预测模型时给出了置信度95%的置信区间。实验结果表明,关联向量机回归模型的预测效果不亚于支持向量机回归模型；且在给出预测值时,还能同时给出预测结果的可信程度。人工神经网络算法易出现过学习或欠学习、局部极小、网络结构难以确定、推广能力差等问题。针对水质指标在线监测的特点,提出了一种基于在线贯序极限学习机算法的水质时间序列预测模型,并以该模型对支持向量回归模型采用过的数据进行实验,对该模型的有效性进行了验证。然后将在线贯序极限学习机预测模型与人工神经网络预测模型进行比较。实验结果表明,在线贯序极限学习机预测模型的预测效果整体上优于人工神经网络,且预测精度高,训练时间短。此外本文还对基于深度信念网络的水质预测模型进行了初步探讨。
[Abstract]:Water quality prediction is the basic work of water resources management and pollution control. It is very important to accurately predict the trend of pollutant concentration in water body over time. At present, there are many water quality prediction methods at home and abroad. However, there are still some shortcomings in these methods. In this paper, four water quality prediction models, namely support vector regression water quality prediction model and correlation vector machine water quality prediction model, are discussed. The water quality prediction model of extreme learning machine and the water quality prediction model of depth belief network. In this paper, the control variables of support vector machine are determined by biogeographic optimization algorithm when establishing support vector regression water quality prediction model. The water quality prediction model was used to predict four important water quality indexes, such as PH value, dissolved oxygen, permanganate index and ammonia nitrogen. The experiment was carried out using the time series data of water quality monitoring in Panzhihua Longdong, Sichuan Province, issued by the Ministry of Environmental Protection. Compared with the traditional control variable optimization method of support vector machine, the results show that the water quality prediction model established by improved biogeographic optimization method is effective, and there are some problems in the water quality prediction model of support vector machine. If the kernel function must satisfy the Mercer condition, the number of support vectors will increase linearly with the increase of the training sample, and only the deterministic prediction results will be given, and there is no probability output. It is impossible to estimate the uncertainty of prediction. On this basis, a water quality time series prediction model based on correlation vector machine regression is proposed, and the validity of the model is verified. Then the correlation vector machine regression prediction model and support vector machine regression prediction model are compared. In order to compare the prediction effect of different kernel functions, the kernel function of the prediction model adopts linear function and Gao Si function, respectively. The confidence interval of confidence degree 95% is given when applying the regression prediction model of association vector machine. The experimental results show that the prediction effect of correlation vector machine regression model is no less than that of support vector machine regression model, and when the prediction value is given, At the same time, the reliability of the prediction results can be given. The artificial neural network algorithm is prone to some problems, such as learning or underlearning, local minima, hard to determine the network structure, poor generalization ability, etc. In view of the characteristics of on-line monitoring of water quality index, A water quality time series prediction model based on online sequential limit learning machine algorithm is proposed, and the model is used to test the data used in the support vector regression model. The validity of the model is verified. Then the on-line sequential learning machine prediction model is compared with the artificial neural network prediction model. The experimental results show that, The prediction effect of on-line sequential extreme learning machine is better than that of artificial neural network on the whole, and the prediction accuracy is high and the training time is short. In addition, the water quality prediction model based on depth belief network is discussed in this paper.
【学位授予单位】：浙江师范大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：X832;TP181

【共引文献】

中国期刊全文数据库 前2条

1 李夕兵;范昀;兰明;尚雪义;;基于博弈论的磷石膏充填水质物元评价[J];科技导报;2015年15期

2 张磊;;基于灰色动态预测模型的清河水库水质预测研究[J];吉林水利;2015年11期

，

本文编号：1577320