Use an Improved BP Network to Simulate Lake Eutrophication

Along with its rapid economy expanding, China is also facing the increasingly serious problem of cultural eutrophication. As part of the eutrophication research activities, applications of mathematical models are also widely used in China. In recent yearspsila researches, there is a growing tendency to use artificial neural networks (ANNs) to complement or even replace empirical models, among which BP networks are the most popular ones. However, these works still have some limitations in topology design of their networks, it is because the networkspsilatopology fails to embody important characteristics of aquatic ecosystem that the effectiveness of their models seems remain to be improved. In this paper, with a study on eutrophication of Shenzhen Reservoir, we apply a BP network to simulate and forecast a lake aquatic ecosystem.The simulation result demonstrates the satisfying performance of BP networks in eutrophication modeling. Meanwhile, comparing three kinds of network topology with each other, this paper further proves that, by way of dividing factors into press and response ones to define inputs and outputs respectively and building up feedback routs between inputs and outputs, the performance of BP networks to simulate eutrophication could be improved effectively. Moreover, the trained network is also used to optimize countermeasure combination, with the aim to control eutrophication of Shenzhen Reservoir effectively and economically.