ABSTRACT

In this study, four kernels extreme learning machines (KELM): radial basis function (RBELM), polynomial (POELM), wavelet (WKELM) and linear (LNELM) extreme learning machines were compared for modelling monthly pan evaporation from Algerian dams reservoirs, according to three scenarios. In the first scenario, the model were developed using splitting ratio of 70/30%, for training and validation subset, respectively, and the POELM1 achieves better performances. For the second scenario, the best models were trained using validation dataset and tested with the training dataset. Results showed that, RBELM1 would appear to yield the most accurate results, across all four dam’s reservoirs, with R² between 0.852 and 0.949, and NSE between 0.846 and 0.946, respectively. For the third scenario, when the models were developed using pooled data and validated at each station separately, the R² and NSE values ranged from 0.815 to 0.937 and from 0.809 to 0.928, respectively. Generally speaking the results obtained were very encouraging. Our findings show that KELM are good and more consistent models, and can predict evaporation across large climatic zones. The findings suggest that the proposed KELM is useful to help establish more robust tools and further improve available machines learning approaches.

摘要

在这项研究中，四个内核极限学习机 (KELM)：径向基函数 (RBELM)、多项式 (POELM)、小波 (WKELM) 和线性 (LNELM) 极限学习机进行了比较，用于模拟阿尔及利亚大坝水库的月蒸发量，根据到三个场景。在第一个场景中，模型是使用 70/30% 的拆分比例开发的，分别用于训练和验证子集，POELM1 实现了更好的性能。对于第二种情况，使用验证数据集训练最佳模型并使用训练数据集进行测试。结果表明，RBELM1 似乎在所有四个大坝的水库中产生最准确的结果，R ²分别在 0.852 和 0.949 之间，NSE 分别在 0.846 和 0.946 之间。对于第三种情况，当模型使用汇总数据开发并在每个站点分别进行验证时，R ²和 NSE 值的范围分别为 0.815 至 0.937 和 0.809 至 0.928。总的来说，取得的结果是非常令人鼓舞的。我们的研究结果表明，KELM 是良好且更一致的模型，可以预测大气候带的蒸发。研究结果表明，提议的 KELM 有助于建立更强大的工具并进一步改进可用的机器学习方法。