The quality of surface waters plays a key role in the sustainability of ecological systems. Measuring water quality parameters (WQPs) is of high importance in the management of surface water resources. In this paper, contemporary-developed regression analysis was proposed to estimate the hard-to-measure parameters from those that can be measured easily. To this end, we proposed a novel modification of support vector regression (SVR), known as multiple-kernel support vector regression (MKSVR) algorithm. The MKSVR learns an optimal data representation for regression analysis by either linear or nonlinear combination of some precomputed kernels. For solving the optimization problem of the MKSVR, the particle swarm optimization (PSO) algorithm was used. The proposed algorithm was assessed using WQPs taken from Karun River, Iran. MKSVR was used to estimate chemical oxygen demand (COD) and biochemical oxygen demand (BOD) using nine WQPs as the input variables, namely electrical conductivity, sodium, calcium, magnesium, phosphate, nitrite, nitrate nitrogen, turbidity, and pH. The results of the proposed MKSVR were compared with those obtained using the SVR and Random Forest regression (RFR). The results showed that the MKSVR algorithm (correlation coefficient [R] = 0.8 and root mean squared error [RMSE] = 4.76 mg/l) increased the accuracy level of BOD prediction when compared with SVR (R = 0.68 and RMSE = 5.15 mg/l) and RFR (R = 0.77 and RMSE = 5.15 mg/l). In the case of COD estimation, the performance of a developed support vector machine (SVM) technique was satisfying. Overall, the use of MKSVR along with the PSO algorithm could demonstrate the superiority of the newly developed SVM technique for the WQPs estimation in the natural streams.

地表水的质量在生态系统的可持续性中起着关键作用。测量水质参数 (WQP) 在地表水资源管理中非常重要。在本文中，提出了当代发展的回归分析，以从易于测量的参数中估计难以测量的参数。为此，我们提出了支持向量回归 (SVR) 的新修改，称为多核支持向量回归 (MKSVR) 算法。MKSVR 通过一些预先计算的内核的线性或非线性组合来学习回归分析的最佳数据表示。为了解决 MKSVR 的优化问题，使用了粒子群优化 (PSO) 算法。所提出的算法是使用取自伊朗卡伦河的 WQP 进行评估的。MKSVR 用于估计化学需氧量 (COD) 和生化需氧量 (BOD)，使用九个 WQP 作为输入变量，即电导率、钠、钙、镁、磷酸盐、亚硝酸盐、硝态氮、浊度和 pH。将提议的 MKSVR 的结果与使用 SVR 和随机森林回归 (RFR) 获得的结果进行比较。结果表明，与 SVR（R = 0.68 和 RMSE = 5.15 mg/l）相比，MKSVR 算法（相关系数 [R] = 0.8 和均方根误差 [RMSE] = 4.76 mg/l）提高了 BOD 预测的准确度水平。 l) 和 RFR（R = 0.77 和 RMSE = 5.15 mg/l）。在 COD 估计的情况下，开发的支持向量机 (SVM) 技术的性能令人满意。全面的，