WPD-COA-EL M模型在汛期月降水量时间序列预测中的应用研究  

Application of WPD-COA-ELM Model in the Prediction of Time Series of Monthly Precipitation in Flood Season

作　　者：杨琼波 崔东文 YANG Qiongbo;CUI Dongwen(Yunnan Provincial Hydrology and Water Resources Bureau,Honghe Branch,Honghe 661100,China;Yunnan Province Wenshan Water Bureau,Wenshan 663000,China)

机构地区：[1]云南省水文水资源局红河分局,云南红河661100 [2]云南省文山州水务局,云南文山663000

出　　处：《水文》2023年第1期17-23,共7页Journal of China Hydrology

摘　　要：针对月降水量时间序列多尺度非平稳性特点,提出小波包分解(WPD)-白骨顶鸟优化算法(COA)-极限学习机(ELM)相融合的降水量预测模型。首先,利用WPD将非平稳月降水量时间序列分解为若干子序列分量;然后在不同维度条件下利用6个典型函数对COA进行仿真测试;利用COA优化ELM输入层权值和隐含层偏值,对每一个子序列分量分别建立COA-ELM模型进行预测,将预测结果叠加重构后即为最终预测结果;最后,以云南省龙潭站汛期和主汛期月降水量数据为例进行实验,并与WPD-COA-BP、WPD-ELM、WPD-BP预测模型进行比较。结果表明:COA在不同维度条件下均具有较好的寻优精度和全局搜索能力。WPD-COA-ELM模型对实例汛期、主汛期月降水量时间序列预测的平均绝对百分比误差分别为3.91%、3.59%,预测精度优于WPD-COA-BP模型,远优于WPD-ELM.WPD-BP模型。WPD能科学降低月降水时间序列数据的复杂性,提高预测效果;COA能有效优化ELM输入层权值和隐含层偏值,提高ELM网络性能。According to the multi-scale non-stationary characteristics of monthly precipitation time series,a precipitation prediction model integrating wavelet packet decomposition(WPD),coot optimization algorithms(COA)and extreme learning machine(ELM)is proposed.Firstly,the non-stationary monthly precipitation time series is decomposed into several subsequence components by WPD;Secondly,the principle of coa is briefly introduced,and six typical functions are used to simulate and test COA under different dimensions;The COA is used to optimize the elm input layer weight and hidden layer bias,and the COA-ELM model is established for each subsequence component for prediction.The final prediction result is the superposition and reconstruction of the prediction results;Finally,taking the monthly precipitation data of flood season and main flood season of Longtan station in Yunnan Province as an example,the experiment is compared with wpd-coa-bp,wpd-elm and wpd-bp prediction models.The results show that COA has good optimization accuracy and global search ability under different dimensions.The average absolute error of WPD-COA-ELM model in predicting the time series of monthly precipitation in case flood season and main flood season is only 3.91%and 3.59%respectively.The predicted value closely follows the change trend of monthly precipitation,and the prediction accuracy is better than WPD-COA-BP model,far better than WPD-ELM and WPD-BP model.WPD can scientifically reduce the complexity of monthly precipitation time series data and improve the prediction effect;COA can effectively optimize elm input layer weight and hidden layer bias,and improve elm network performance.

关 键 词：降水量预测 小波包分解 白骨顶鸟优化算法 极限学习机 仿真测试 

分 类 号：P457.6[天文地球—大气科学及气象学]