使用多重包含模型和大型气候指数进行降雨预测。
Rainfall prediction using multiple inclusive models and large climate indices.
作者信息
Mohamadi Sedigheh, Sheikh Khozani Zohreh, Ehteram Mohammad, Ahmed Ali Najah, El-Shafie Ahmed
机构信息
Department of Ecology, Institute of Science and High Technology and Environmental Sciences, Graduate University of Advanced Technology, Kerman, Iran.
Institute of Structural Mechanics, Bauhaus Universität Weimar, 99423, Weimar, Germany.
出版信息
Environ Sci Pollut Res Int. 2022 Dec;29(56):85312-85349. doi: 10.1007/s11356-022-21727-4. Epub 2022 Jul 6.
Rainfall prediction is vital for the management of available water resources. Accordingly, this study used large lagged climate indices to predict rainfall in Iran's Sefidrood basin. A radial basis function neural network (RBFNN) and a multilayer perceptron (MLP) network were used to predict monthly rainfall. The models were trained using the naked mole rat (NMR) algorithm, firefly algorithm (FFA), genetic algorithm (GA), and particle swarm optimization (PSO) algorithm. Large lagged climate indices, as well as three hybrid models, i.e., inclusive multiple model (IMM)-MLP, IMM-RBFNN, and the simple average method (SAM), were then employed to predict rainfall. This paper aims to predict rainfall using large climate indices, ensemble models, and optimized artificial neural network models. Also, the paper considers the uncertainty resources in the modeling process. The inputs were selected using a new input selection method, namely a hybrid gamma test (GT). The GT was integrated with the NMR algorithm to create a new test for determining the best input scenario. Therefore, the main innovations of this study were the introduction of the new ensemble and the new hybrid GT, as well as the new MLP and RBFNN models. The introduced ensemble models of the current study are not only useful for rainfall prediction but also can be used to predict other metrological parameters. The uncertainty of the model parameters and input data were also analysed. It was found that the IMM-MLP model reduced the root mean square error (RMSE) of the IMM-RBFNN, SAM, MLP-NMR, RBFNN-NMR, MLP-FFA, RBFNN-FFA, MLP-PSO, RBFNN-PSO, MLP-GA, and RBFNN-GA, MLP, and RBFNN models by 12%, 25%, 31%, 55%, 60%, 62%, 66%, 69%, 70%, 71%, 72%, and 72%, respectively. The IMMs, such as the IMM-MLP, IMM-RBFNN, and SAM, outperformed standalone models. The uncertainty bound of the multiple inclusive models was narrower than that of the standalone MLP and RBFNN models. The MLP-NMR model decreased the RMSE of the RBFNN-NMR, RBFNN-FFA, RBFNN-PSO, and RBFNN models by 15%, 26%, 37%, 42%, and 45%, respectively. The proposed ensemble models were robust tools for combining standalone models to predict hydrological variables.
降雨预测对于可用水资源的管理至关重要。因此，本研究使用大滞后气候指数来预测伊朗塞菲德罗德河流域的降雨量。采用径向基函数神经网络（RBFNN）和多层感知器（MLP）网络来预测月降雨量。使用裸鼹鼠（NMR）算法、萤火虫算法（FFA）、遗传算法（GA）和粒子群优化（PSO）算法对模型进行训练。然后采用大滞后气候指数以及三种混合模型，即包含性多模型（IMM）-MLP、IMM-RBFNN和简单平均法（SAM）来预测降雨量。本文旨在利用大气候指数、集成模型和优化的人工神经网络模型来预测降雨量。此外，本文还考虑了建模过程中的不确定性因素。采用一种新的输入选择方法，即混合伽马检验（GT）来选择输入。将GT与NMR算法相结合，创建了一种用于确定最佳输入方案的新检验方法。因此，本研究的主要创新点在于引入了新的集成模型和新的混合GT，以及新的MLP和RBFNN模型。本研究引入的集成模型不仅对降雨预测有用，还可用于预测其他气象参数。还分析了模型参数和输入数据的不确定性。结果发现，IMM-MLP模型分别将IMM-RBFNN、SAM、MLP-NMR、RBFNN-NMR、MLP-FFA、RBFNN-FFA、MLP-PSO、RBFNN-PSO、MLP-GA和RBFNN-GA、MLP以及RBFNN模型的均方根误差（RMSE）降低了12%、25%、31%、55%、60%、62%、66%、69%、70%、71%和72%。IMM-MLP、IMM-RBFNN和SAM等IMM模型的表现优于独立模型。多个包含性模型的不确定性范围比独立的MLP和RBFNN模型更窄。MLP-NMR模型分别将RBFNN-NMR、RBFNN-FFA、RBFNN-PSO和RBFNN模型的RMSE降低了15%、26%、37%、42%和45%。所提出的集成模型是用于组合独立模型以预测水文变量的强大工具。
Zotero 插件