节点文献
基于神经网络的水质评价与预测的探索
Exploration on Neural Networks for Water Quality Assessment and Prediction
【作者】 杜伟；
【导师】 孙宝盛；
【作者基本信息】 天津大学 ， 环境工程， 2007， 硕士
【摘要】 人工神经网络(ANN)是复杂非线性科学和人工智能科学的前沿,在水污染控制领域的应用在国内外尚处于起步阶段。本文在全面分析评述水质评价和水质预测的研究现状,分析阐述了人工神经网络基本原理、算法和模糊数学基本理论的后,将模糊神经网络(FNN)和径向基神经网络(RBFNN)引入水污染控制领域,主要在水质评价和水质预测方面进行了一些探索性的研究工作,为提高水质评价和水质预测的智能化水平做出了努力。本文介绍了FNN的原理、算法和模式特征。FNN不再是一黑箱,其所有节点与参数都具有物理意义,并克服了ANN结构的选择缺乏充分理论分析的缺点。FNN模型既能直接表达人们惯用的逻辑含义,又兼具ANN的自适应学习功能和非纯属表述能力等优点。将FNN应用于水质评价是本文的初探,通过实例研究证明,学习五类水质标准后的FNN能够正确评价其它的水质样本,具有较好的客观性、可靠性和可解释性。在充分研究RBFNN机理的基础上,将RBFNN应用到水质预测中,并以深圳河在线监测数据为训练样本,构建了RBFNN水质预测模型。应用该模型对深圳河2006年11月19日至2006年11月29日10天的水质进行预测,并以该时间段的真实监测数据验证预测结果的准确性,验证结果表明该模型预测结果误差较小、拟合性好。为了比较RBFNN与BPNN(反向传播神经网络)的预测性能,本文还以相同的监测数据建立了BPNN水质预测模型,将其预测结果与RBFNN水质预测模型的预测结果相比较,比较结果表明,RBFNN的预测结果明显优于BPNN。而且在两个模型建模的过程中,RBFNN无论在收敛速度,还是输出结果的稳定性,均好于BP神经网络。本文研究表明:用FNN和RBFNN进行水质评价和水质预测在理论上可行,在实践上有继续深入研究开发的价值,具有良好的应用前景。
【Abstract】 Artificial neural network(ANN) plays a leading role in the sciences for complex non-linear phenomena and artificial intelligence. Researches on its application in water pollution control are still in the preliminary stage in the world. On the basis of a conprehensive Analysis and evaluation of the present situation of the researches in water quality assessment and prediction, and on the bsis of a careful exposition of basic principles and the optimal algorithm of ANN and the basic theory of fuzzy mathematics, this dissertation gives an application of fuzzy neural netowrk (FNN) and radial basis fuction neural network (RBFNN) approach in the sheme of water pollution control, and the main research of this article is to do explore some new approach for water quality assessment and prediction.This paper emphasizes the principles, the algorithm and the pattern features of FNN model that is composed by ANN and Fuzzy System according to learning integrated. The FNN model is not a black box any more and its all nodes and parameters have physical meaning, and it overcomes the disadvantage which choosing ANN configuration is short of sufficient theoretical analysis. FNN model can not only direct expresses the logic meaning of peaple’s customs but also have the merits of ANN self-adaptation learning and non-linear expression. Researches on FNN application in the water quality evaluation are preliminary exploration of this article’s. Case studies show that FNNis abble to correctly evaluate other samples besides the training samples after learning, thus has better objectiveness, reliability and expression.Based on the algorithm of RBFNN study, this article applies it to water prediction that sets up a RBFNN(back propagation neural network) water predicton model on the training data from Shenzhen river on-line water monitoring station, uses the well trained model to predict the water quality of Shenzhen river from 2006-11-19 to 2006-11-29, and assesses the prediction precision by the monitoring data. The case study shows that the prediction results of RBFNN model have high precision. In order to compare the prediction performance of RBFNN’s with BPNN’s, this paper sets up a BP-NN water prediction model by the same data, and compares its prediciton result with the RBFNN’s. The comparing results show that the predicton outcomes of RBFNN’are evidently preciser than the BP-NN’s, and in the process of establishing the both models, the author finds that the RBFNN is also better than BPNN both in convergence speed and outcome stability.This research demonstrates that theirs theoretical feasibility and great practical utility, FNN water quality assessment and RBFNN water quality prediction have good prospects for further development and application.