基于EMD的BP神经网络在凌河流域旱灾预测中的应用

摘要

为提高旱灾预测模型预测精度，利用EMD(经验模态分解法)处理非平稳信号的优势，将其应用到BP神经网络预测模型中，建立基于EMD的BP神经网络旱灾预测模型，对凌河流域44个观测站(小凌河流域11站、大凌河流域33站)共51年(1960~2010)的降水资料进行旱灾预测应用，同时将基于EMD的BP神经网络旱灾预测模型结果与BP神经网络预测模型结果进行对比。结果表明：小凌河流域基于EMD的BP神经网络预测模型、BP神经网络预测模型的年均降水量预测值均方误差(MSE)分别为0.0011和0.0076，决定系数(R2)分别为0.95和0.83；大凌河流域基于EMD的BP神经网络预测模型、BP神经网络模型的年均降水量预测值均方误差(MSE)分别为0.0032和0.0092，决定系数(R2)分别为0.93和0.79。基于EMD的BP神经网络预测值均方误差(MSE)较小且决定系数(R2)较高，均优于BP神经网络预测值，提高了BP神经网络旱灾预测模型预测精度，具有一定的可行性。%To improve the accuracy of drought prediction, the EMD (Empirical Mode Decomposition) in processing non-stationary single was used to establish BP neural network forecast model. A drought prediction model was established to conduct a drought prediction for precipitation data from 44 stations (11 stations in Xiaoling River Basin, 33 stations in Daling River Basin) in a total of 51 years (1960-2010) in Ling River Basin and compare the forecast results obtaining from BP neural network prediction model and results from EMD of BP neural network. Results showed that the annual average rainfall prediction mean square error (MSE) based on EMD of BP neural network prediction model and BP neural network prediction model in Xiaoling River Basin were 0.0011 and 0.0076, determination coefficient (R2) were 0.95 and 0.83, The annual average rainfall prediction mean square error(MSE) based on EMD of BP neural network prediction model and BP neural network prediction model in Daling River Basin were 0.0032 and 0.0092, determination coefficient (R2) were 0.93 and 0.79, The mean square error (MSE) of BP neural network prediction values based on EMD is smaller, coefficient of determination (R2) is higher and they are all better than the BP neural network and improve the prediction accuracy of BP neural network drought model. Thus, it has a certainly feasibility, could provide the basis for drought control and drought resistance in Ling River Basin and a new method for drought prediction.