CAJ | 학술논문

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（ IGA?WNN）。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。
위료진일보제고뇌폭예보적준학솔，재분석연구뇌폭예보방법적기출상，제출료일충료기우개진유전산법우화소파신경망락적뇌폭예보방법（ IGA?WNN）。해방법이용취류분석화우돈질대법대다충군유전산법적수렴방향화정도진행개진，피면료충군동질화여국부최우문제，채용개진적유전산법대소파신경망락적초시권치역치진행료우화。선용남경지구2008—2009년6—8월적탐공화섬전정위자료，사용회관련법알굴출관련정도교대적대류삼수작예보인자，귀일화처리후수입모형，채용독립양본진행예보검험。결과표명，여 BP 신경망락등방법상비，IGA?WNN예보준학솔경고，구유경호적비선성처리능력화범화성。
A thunderstorm forecasting method based on the Wavelet Neural Network optimized by the Improved Ge?netic 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 em?ploys 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 net?work,IGA?WNN achieves higher prediction accuracy,and has better nonlinear processing capability as well as stron?ger generalization.