Samantaray Sandeep, Sahoo Abinash, Agnihotri Ankita
Department of Civil Engineering, NIT Srinagar, India.
Department of Civil Engineering, NIT Silchar, India.
MethodsX. 2023 Feb 7;10:102060. doi: 10.1016/j.mex.2023.102060. eCollection 2023.
A crucial necessity in integrated water resource management is flood forecasting. Climate forecasts, specifically flood prediction, comprise multifaceted tasks as they are dependant on several parameters for predicting the dependant variable, which varies from time to time. Calculation of these parameters also changes with geographical location. From the time when Artificial Intelligence was first introduced to the field of hydrological modelling and prediction, it has produced enormous attention in research aspects for additional developments to hydrology. This study investigates the usability of support vector machine (SVM), back propagation neural network (BPNN), and integration of SVM with particle swarm optimization (PSO-SVM) models for flood forecasting. Performance of SVM solely depends on correct assortment of parameters. So, PSO method is employed in selecting SVM parameters. Monthly river flow discharge for a period of 1969 - 2018 of BP ghat and Fulertal gauging sites from Barak River flowing through Barak valley in Assam, India were used. For obtaining optimum results, different input combinations of Precipitation (P), temperature (T), solar radiation (Sr), humidity (H), evapotranspiration loss (E) were assessed. The model results were compared utilizing coefficient of determination (R) root mean squared error (RMSE), and Nash-Sutcliffe coefficient (N). The most important results are highlighted below.•First, the inclusion of five meteorological parameters improved the forecasting accuracy of the hybrid model.•Second, model comparison specifies that hybrid PSO-SVM model executed superior performance with RMSE- 0.04962 and NSE- 0.99334 compared to BPNN and SVM models for monthly flood discharge forecasting.•Third, applied optimization algorithm has easy implementation, simple theory, and high computational efficacy. Results revealed that PSO-SVM could be utilised as an improved alternate method for flood forecasting as it provided a higher degree of reliability and accurateness.
洪水预报是水资源综合管理的一项关键需求。气候预报，特别是洪水预测，包含多方面的任务，因为它们依赖于多个参数来预测因变量，而因变量会随时间变化。这些参数的计算也会因地理位置而异。自人工智能首次被引入水文建模与预测领域以来，它在水文领域的进一步发展研究方面引起了极大关注。本研究调查了支持向量机（SVM）、反向传播神经网络（BPNN）以及支持向量机与粒子群优化算法相结合（PSO - SVM）模型在洪水预报中的适用性。支持向量机的性能完全取决于参数的正确选择。因此，采用粒子群优化算法来选择支持向量机的参数。使用了印度阿萨姆邦巴拉克山谷中流经的巴拉克河上BP加特和富勒塔尔测量站1969年至2018年期间的月河流量数据。为了获得最佳结果，评估了降水（P）、温度（T）、太阳辐射（Sr）、湿度（H）、蒸发散损失（E）的不同输入组合。利用决定系数（R）、均方根误差（RMSE）和纳什 - 萨特克利夫系数（N）对模型结果进行了比较。最重要的结果如下：首先，纳入五个气象参数提高了混合模型的预测精度。其次，模型比较表明，与BPNN和SVM模型相比，混合PSO - SVM模型在月洪水流量预报方面表现更优，RMSE为0.04962，NSE为0.99334。第三，应用的优化算法易于实现、理论简单且计算效率高。结果表明，PSO - SVM可作为一种改进的洪水预报替代方法，因为它具有更高的可靠性和准确性。
