基于改进遗传小波神经网络的雷暴预报方法

Thunderstorm forecasting method based on improved genetic wavelet neural network

为了进一步提高雷暴预报的准确率,在分析研究雷暴预报方法的基础上,提出了一种了基于改进遗传算法优化小波神经网络的雷暴预报方法(IGAWNN).该方法利用聚类分析和牛顿迭代法对多种群遗传算法的收敛方向和精度进行改进,避免了种群同质化与局部最优问题,采用改进的遗传算法对小波神经网络的初始权值阈值进行了优化.选用南京地区2008—2009年6—8月的探空和闪电定位资料,使用灰关联法挖掘出关联程度较大的对流参数作预报因子,归一化处理后输入模型,采用独立样本进行预报检验.结果表明,与BP神经网络等方法相比,IGA-WNN预报准确率更高,具有更好的非线性处理能力和泛化性.

A thunderstorm forecasting method based on the Wavelet Neural Network optimized by the Improved Genetic Algorithm （IGA-WNN） is put forward in order to improve the accuracy of thunderstorm potential prediction. This method takes use of Cluster Analysis and Newton Iteration Method to improve the convergence direction and precision of multiple population genetic algorithm, thus avoids population homogeneity and local optimum;and employs improved Genetic Algorithm to optimize the initial weights of the threshold of wavelet neural network. The sounding data and lightning location data in Nanjing area from June to August during 2008 and 2009 were used for thunderstorm forecasting, and the convective parameters with higher degree of association, which were selected by grey correlation method, were normalized and put into the proposed model. Independent data are used to verify the forecast result.The forecasting and verification result indicate that, compared to other methods like BP neural network, IGA-WNN achieves higher prediction accuracy, and has better nonlinear processing capability as well as stronger generalization.