Mutual Information Based RBF Neural Network Model And Its Application In Short-term Runoff Forecasting

With the development of our society and economy, water resources, as an essential part of human survival and social and economic development, has an irreplaceable place. It has always been the hot point and difficulty of reasonable development and utilization in and abroad, hydrologic forecasting is the precondition of reasonable exploitation and utilization and the key to the question of reasonable planning and construction of water conservancy project. The hydrologic forecasting develops up to now, it has developed from traditional method such as hydrology statistical method and factor analytical method to fuzzy mathematics, neural network method, grey system theory, multi hierarchic method and optimal combined forecasting model. Each of these method has its own characteristics and superiority.This paper expounds the fundamental of mutual information and the nearest neighbor clustering RBF neural network, it sucks up the variable correlativity representational capacity of mutual information and the nonlinear approximation function of RBF neural network and establishes the mutual information based modified RBF neural network forecasting model. Aiming at the shortage of the nearest neighbor RBF neural network, this arithmetic filtrates the input layer and simplifies the neural network structure and improves the computation speed and prediction accuracy. On this basis, this paper analyzes the forecasting accuracy and error characteristics and take use of the limited memory least square method to proceed real-time correction, which can improves the forecasting precision. Taking the Luning hydrometric station located in La-lung river as example and analyzes the application of forecasting model in short-term runoff forecasting, it verifies the reasonability and effectiveness of the model presented in this paper. The result shows that the mutual information based modified RBF neural network forecasting model can improve the forecasting precision, and a better result can be obtained after real-time correction. It can provides a more reliable result for the operation of Guandi hydropower station.