MODELING AND SIMULATION FOR PREDICTING WATER-FLOWER BASED ON RBF NEURAL NETWORK

The research on ecological models of water flower is difficult and complicated, because of the complexity of its mechanism and the effect of human beings. Main factors which make water bloom break out in river and lakes is analyzed, and the modeling method of short-time prediction for water bloom based on RBF neural network, including supervise learning method for the center, width and weight of base function in RBF neural network, error-correction algorithm based on gradient descent of RBF, is proposed. The effect which hidden layer of RBF brings to network performance is compared, and fitting capacity between RBF's width and generalization capability of network is discussed. According to the results of network training and water bloom predict, RBF neural network can be used to predict the change of Chi_a (Chlorophyll a) in short term. Because of the strong generalization capability, high predict precision and good fitting performance, the model has established a solid foundation for further research on short-term predict methods of water bloom in river and lakes and the simulation result showed that the method is very practice and useful.