Artificial neural network (ANN) mathematical models, such as the radial basis function neural network (RBFNN), have been used successfully in different environmental engineering applications to provide a reasonable match between the measured and predicted concentrations of certain important parameters. In the current study, two RBFNNs (one conventional and one based on particle swarm optimization (PSO)) are employed to accurately predict the removal of chemical oxygen demand (COD) from polluted water streams using submerged biofilter media (plastic and gravel) under the influence of different variables such as temperature (18.00-28.50 °C), flow rate (272.16-768.96 m3/day), and influent COD (55.50-148.90 ppm). The results of the experimental study showed that the COD removal ratio had the highest value (65%) when two plastic biofilter media were used at the minimum flow rate (272.16 m3/day). The mathematical model results showed that the closeness between the measured and obtained COD removal ratios using the RBFNN indicates that the neural network model is valid and accurate. Additionally, the proposed RBFNN trained with the PSO method helped to reduce the difference between the measured and network outputs, leading to a very small relative error compared with that using the conventional RBFNN. The deviation error between the measured value and the output of the conventional RBFNN varied between + 0.20 and - 0.31. However, using PSO, the deviation error varied between + 0.058 and - 0.070. Consequently, the performance of the proposed PSO model is better than that of the conventional RBFNN model, and it is able to reduce the number of iterations and reach the optimum solution in a shorter time. Thus, the proposed PSO model performed well in predicting the removal ratio of COD to improve the drain water quality. Improving drain water quality could help in reducing the contamination of groundwater which could help in protecting water resources in countries suffering from water scarcity such as Egypt.

中文翻译：

---

人工神经网络在使用浸没式生物滤池改善农业排水水质中的应用。

人工神经网络（ANN）数学模型，例如径向基函数神经网络（RBFNN），已成功用于不同的环境工程应用中，以提供某些重要参数的测量浓度和预测浓度之间的合理匹配。在当前的研究中，采用了两种RBFNN（一种是常规的，一种是基于粒子群优化（PSO）的）来精确预测在污染环境下使用浸没式生物滤池介质（塑料和砾石）从污水中去除化学需氧量（COD）的情况。温度（18.00-28.50°C），流速（272.16-768.96 m3 /天）和进水COD（55.50-148.90 ppm）等不同变量的影响。实验研究结果表明，当使用两种塑料生物滤池以最小流速（272.16立方米/天）使用时，COD去除率最高（65％）。数学模型结果表明，使用RBFNN测得的COD和获得的COD去除率之间的接近度表明，该神经网络模型有效且准确。此外，提出的采用PSO方法训练的RBFNN有助于减小测量结果与网络输出之间的差异，与使用常规RBFNN相比，导致相对误差非常小。测量值和常规RBFNN输出之间的偏差误差在+ 0.20和-0.31之间变化。但是，使用PSO，偏差误差在+ 0.058和-0.070之间变化。所以，提出的PSO模型的性能优于传统的RBFNN模型，并且能够减少迭代次数并在更短的时间内达到最佳解决方案。因此，提出的PSO模型在预测COD去除率以改善排水水质方面表现良好。改善排水水质可以帮助减少地下水的污染，这可以帮助保护水资源短缺国家（如埃及）的水资源。