Aiming at forecasting typhoon tracks by using genetic algorithm and neural network approach, this paper presents a new prediction modeling scheme for selecting the structure of networks and determining the initial connection weight. The probability for obtaining a global optimal solution is raised by designing a strategy that the optimum individual for each generation is reserved after a certain number of generations in the computational process of genetic evolution. The case forecast results of the typhoon track over the South China sea area show that mean absolute error of the prediction during 1990-2003 is 150.0km form this new forecast model, and in comparison with the optimum individual form the last generation, under the conditions of the same predictors and period forecast error is 161.1km. Furthermore, it is also found that higher predictive accuracy form the forecast models using genetic algorithm and neural network approach, comparing the results to those form objective prediction technique of typhoon tracks and the climatology and persistence (CLIPER) methods
[1]
Meng Zhiyong,et al.
Recent progress on tropical cyclone research in China
,
2002
.
[2]
Konstantinos P. Ferentinos,et al.
Biological engineering applications of feedforward neural networks designed and parameterized by genetic algorithms
,
2005,
Neural Networks.
[3]
S. Aberson,et al.
On the Predictability of Tropical Cyclone Tracks in the Northwest Pacific Basin
,
2003
.
[4]
C. J. Neumann,et al.
Statistical Prediction of Tropical Cyclone Motion: An Analog-CLIPER Approach
,
2002
.
[5]
Qin,et al.
Study on the Overfitting of the Artificial Neural Network Forecasting Model
,
2004
.