Research On Meteorological Time Series Prediction Based On RBF Neural Network

Meteorological time series as the important information of national economy in all industries, if we can carry it on the reasonable utilization, it will provide the important theory and realistic basis for prevention of meteorological disaster and meeting the public meteorological service needs.According to the nonlinear and non-stationary characteristics of meteorological time series, Radial Basis Function (RBF) neural network has been selected. Because RBF network has some shortcomings like the performance is heavily dependent on the initial parameter, some effective improved strategies have introduced and put forward some optimization RBF network models. Prediction models are applied to the meteorological time series forecast, the results are satisfactory. This research work and innovations includes the following aspects:(1) As RBF neural network is sensitive to parameter settings, the particle swarm optimization (PSO) algorithm and quantum particle swarm optimization (QPSO) are introduced and optimization RBF network models are built. Then they are applied for monthly rainfall forecast in Zhanjiang station from2001to2003, the results verify the reliability and superiority of QPSO-RBF neural network model.(2) QPSO algorithm has some shortcomings such as decreasing diversity and easily falls into local convergence. In order to solve this problem, this paper put forward an improved QPSO algorithm based on genetic operations. By adopting genetic operation strategy, the population diversity has increased and the global search ability of QPSO algorithm has improved. Experimental results verify the effectiveness of the proposed algorithm.(3) In the temperature prediction application, GQPSO algorithm is applied to optimize RBF neural network, a GQPSO-RBF neural network model has been built to apply the temperature prediction, and compare with the QPSO-RBF neural network and RBF neural network. The results show that, the predicted temperature variation and phase by the GQPSO-RBF neural network model is more consistent, the model has highly precision and stability predicted ability.