Research On Optimization Of Neural Network Model By Improved PSO Algorithm And Its Application

In the field of hydrological subject, accurate prediction is one of the important research topics preventing disaster. In recent years, because of the impact of global climate change and human activities, the national meteorological disasters get more and more serious, and last longer and the influence scope is larger, thus, the development of national economy in China has been seriously affected. It is very necessary and important to analyze and research the existing hydrologic data in order to find the law of the change. In recent years, researchers at home and abroad have been successful applied neural network (Neural, Networks, NNS) method in various predictive modelling of atmospheric science. But there is no quantitative method determining the parameters used for the optimization of neural network, among which BP neural network is used the most frequently. But there are many disadvantages for complex and multidimensional training data or the parameters of different settings such as overfitting, low convergence speed, into a local optimum and bad effect of forecasting. This greatly limits the application of neural network in real-time rainfall and runoff forecasting model. In recent years, the group of intelligent optimization of artificial neural network has gradually become a hot issue in the field of modern optimization. The particle swarm optimization (PSO) algorithm is a hot research topic in the field of the current optimization algorithm.This paper proposes an improved PSO (particle swarm optimization algorithm), based on the problem that it is difficult for artificial neural network to determine the network structure and optimization of network parameters in the process of forecasting model. Then, it applies each parameter of neural network optimization in the rainfall and runoff forecast model in Guangxi. First of all, it improves the particle swarm algorithm, uses the method of random distribution to obtain the inertia weight to keep the diversity of population and improve the searching ability, and uses strategy of asynchronous change to change the learning factor value to enhance the learning ability of particle and speed up the convergence to the global optimal solution. Secondly, it uses the improved particle swarm to optimize feedforward neural networks. Finally, in the view of the complex and multidimensional factors, it uses a variety of dimension reduction processing method to get the main factors of relevant information. And it designs all data as the training sample set and testing sample set in order to establish rainfall and runoff forecasting model based on neural network of the improved particle swarm optimization. The experiment results show that this model is better than the runoff forecasting model of BP neural network in local and global searching ability, speed and accuracy.