小波神经网络在降水量预测中的应用研究

Application of WNN in Precipitation Forecast

根据吉林西部的自然条件,运用小波神经网络方法,分别对乾安、通榆、扶余三个地区1957-2010年的月降水量资料进行分析,利用多组输入输出数据对模型进行训练,最终分别确定上述三地区任35a的相应降水量值与之后10a每年月降水量最大值、最小值及平均值之间的映射关系,并预测这三个地区2011-2020年的降水量值。与前人利用一组输入输出数据对模型进行训练不同,本文利用多组输入输出样本对模型进行训练。使模型更可靠,更贴近实际,进一步提高了模型的精确性。结果表明:采用小波神经网络方法对降水量进行预测,方法可行、精度较高、结果可靠;上述三地区2011-2020年平均降水量较1957-2010年平均降水量偏少,处于降水量变化周期的枯水期。降水量预测可为当地的防洪、抗旱工作提供依据。

According to the western Jilin natural conditions,using wavelet neural network method,the monthly precipitation data of three areas,including Qianan,Tongyu,Fuyu,in 1957-2010 are analyzed,and the quantity of annual monthly rainfall of the three areas in 2011-2020 are forecasted.The model is trained by using multiple sets of input and output data to finally determine the mapping relationship between the precipitation values of the past 35 years and the corresponding precipitation value of the next 10 years.Different from the former use of a set of input and output data to train model,this paper uses multiple sets of input and output samples to train the model in.So the model is more reliable,more close to reality and more accurate.The results show that using wavelet neural network method to predict precipitation is feasible,of high precision and reliable results.The average precipitation of the three areas during 2011-2020 is less than that during 1957-2010,and it would get into the dry season.The precipitation prediction can provide a basis for the local flood control,drought relief work.