Back-Propagation Network Model For Predicting The Change Of Eutrophication Of Qiandao Lake

Eutrophication has become one of the world's environmental problem. The world's fast economical development has resulted in more and more seriously environmental pollution problems such as water eutrophication of lakes and reservoir. In the areas of lots of human activities with the eutrophication is speeding up, the the water quality is getting so worse and worse that the development of society and economy is limited. Located in Chun'an County of Hangzhou , Qiandao Lake is a large reservior formed in the 1950s after the dam of the Xin'an River power plant was built up. It provides services of generating electricity, controlling flood, tourism, breeding aquatics, shipping, waterhead for human and agriculture, and so on.The further exploitation of Qiandao Lake has brought about more and more environmental problems that threatened the water quality. The water environment is very important in the whole Qiandao Lake valley. So the effective prediction for the change of water quality is the base for setting down the strategy of exploring the tourism in Qiandao Lake valley. On the base of summing up and referring to the existing research achievement, we mainly carrying out the following work:(1) The research development of eutrophication was systematically expatiated as well as the approaches to it.(2) The theory of artificial neural network(ANN) was introduced in details in this paper as well as its application in the research of eutrophication.(3) Among the 13 testing spots of Qiandao Lake, we chose the spot of "Jiekou" as the research object. By means of "Factor Analysis", five water factors of Tw,pH,Chla,SD,TN were chose as the inputs of BP network model from 8 factors of Tw,pH,Chla,SD,TN,COD_Mn,TP,DO.(4) 24 groups of original data from January of 1999 to December of 2000 were adopted in the study. The original data were not enough for establishing a good BP network. Then the method of inserting new data between the original data were put forward to get enough training data for the BP network. 93 groups of training data were got finnaly.(5) Taking Tw,pH,Chla,SD,TN as the inputs and Chla as the output, five BP network models were established. Among the 92 groups of data, the former 89 groups were the for training and the last 3 groups were for validating. Each model was trained for 2000 epochs and the error reached 10^-3.(6) The result showed that among the five models that we constructed, the second one was the best one. It could predict the content of Chla in a relatively accurate extent. That is to say, the BP network model with Tw,pH,Chla,SD as the input and Chla as the output could successfully predict the change of water quality in Qiandao Lake in a short term. It's relative error was within 20%. According to the characteristic of eutrophication, we think this error was permitted. So we chose model 2 to predict the content of Chla at Jiekou.It was showed in this paper that the BP network model can effectively simulate the non-linear act as eutrophication. This BP model could successfully predict the content of Chla thus provide scientific evidence for constructing the administration measurement for Qiandao Lake.