【作者】 魏恒；

【导师】 李伟光；

【作者基本信息】 哈尔滨工业大学 ， 市政工程， 2008， 硕士

【摘要】 近年来随着原水污染的日益严重,建立原水水质预警系统迫在眉睫。本研究采用大庆水库水自动监测站的水质监测数据,主要研究BP神经网络模型对大庆水库水质的预测性能,重点探讨了水库冰封期对水源水质预测的影响作用,并与其他常用模型作了对比,为该市水源水质预警系统的建立提供技术支持,同时也为其它寒冷地区水源地水质预测提供参考。本文确定了BP神经网络作为本次水质预测所用的模型;根据当地水厂的运行情况选择了水质预测参数,作为BP神经网络的网络输出;对原始数据进行灰色关联分析,确定了BP神经网络的输入变量;通过不同参数之间的对比确定了BP神经网络的训练及运行参数。通过对网络预测效果的分析得出BP神经网络能够较好地对该水库水质进行预测,但是数据规律的变化及数据样本数量的不足对预测结果会产生一定的影响。大庆水库处于我国东北寒冷地区,每年有大约半年时间处于冰封期。本文根据大庆水库的这一特点,将原始的数据分为非冰封期与冰封期两个部分进行分析,分别建立BP神经网络,最后得出结论:对于进入冰封期后数据的变化规律有了改变的水质参数,将数据分类后再建立模型能够提升BP网络的预测效果,而对于其他的水质参数,盲目分类只会减少训练样本的数量,使网络训练不充分,造成不能对水质指标进行准确的预测。本文还建立了自回归模型(AR)与灰色预测模型这两种常见的水质预测模型,并与BP神经网络模型做了对比。AR模型在处理复杂曲线的能力上不如BP神经网络,且模型阶数一般较高,但是当曲线的变化较为平缓时AR模型的预测效果更好;灰色预测模型则并不适用于变化幅度较大的水质预测。更多还原

【Abstract】 The raw water warning system is thought to be more and more important these days because of the raw water pollution. This paper analyse the data from Daqin water automatic monitoring station and research the effect of forecasting water quality using BP neural network especially in frozen period. The model is also compared with other common models. This research provide technique supports for setting up a raw water warning system in the city and a reference for how to forecast water quality in cold areas.This paper introduces BP neural network and applies it to forecast water quality of Daqin reservior. According to the technological processes of local waterworks, network outputs are decided. The author also analyzes grey association between different water parameters and choose the inputs of the BP nerual network. The best strutrure of network is established through vast comparisons. From the result we can see that BP neural network is a suitable model for water quality predicton, and the change of water quality and lack of training samples will influence the effect.Daqin reservior locates in the Northeast China, where the frozen period persists a long time every year. This paper divides the data into two parts throughout the water temperature. And set up a BP neural network for each part. To the water parameters whose rule changes in the frozen period, seting up a network after classification can enhance the effect of forecasting. To the other water parameters, the result is even worse because of lack of samples.This paper set up other two common models: AR model and grey model. And make a comparison between these two models and BP neural network. Only when the curve is very smooth, AR model which has a high order is better than BP neural network. Grey model is not suitable for water quality when the water parameter changes fiercely.更多还原