The Quantification Study Of Non-Point Source Pollution Of Lake Ecological Security Based On The Method Of GA-BP

Water pollution is the key link which will influence the ecological security in this basin, and it also becomes the constraint factor of the sustainable development of the society and economy. Since non-point source pollution is an important constituent part of the water pollution, the study which focuses on the non-point source pollution of the drainage area becomes highly significant to the whole ecological security of the lake.A model for non-point source pollutants of lake ecological security has been established respectively by using two methods-neural network and genetic algorithm combined neural network. On the basis of the actual situation of Dianchi Lake as well as the correlative factors of non-point source pollution, indexes of input variables have been set up while COD, TN and TP have been defined as the indexes of output ones. The river discharge, suspended solid (SS), drainage area, slope, population and the type of land use were considered as the indexes of the input invariable for COD. Two methods can be utilized as the input of TN. The first one had the same input indexes with COD, and the second way was to include the index of COD based on the former method. The indexes of the second method for input of TN combined with the TN index were regarded as the indexes of TP input variable. First, a comprehensive value of these six input invariable indexes has been extracted and calculated by using ArcGIS software, and the author used logarithmic transformation to pretreat both the input and output variable indexes. Secondly, based on the data of pretreatment, parameters of neural networks and genetic algorithm together with neural networks were determined and the researcher learned those two algorithms, and then to train the samples. Finally, the model for non-point source pollution of lake ecological security was established.The verification results, taking the Dianchi drainage area as an example, show that:Firstly, from the test error of the non-point source pollutant indexes including chemical oxygen demand (COD), total nitrogen (TN) and total phosphorous (TP), we found that as the input factors of neural network model and genetic algorithm-combined neural network model to predict the non-point source pollutant, variables including river discharge, the amount of suspended solid, drainage area, slope, population and the type of land use were feasible and had certain representativeness. Secondly, the average test error of the genetic algorithm-combined BP neural network model was obviously less than it of the neural network model, which meant the genetic algorithm-combined BP neural network model was better than the later one to predict the non-point source pollutant. Thirdly, the second input method of TN was better than the first one, it was also a feasible method to improve the accuracy.