Abstract

The amount of returned sludge is considered as one of the important and controllable parameters in the operation of wastewater treatment plants and play a vital role in process. There are different approaches to measure the rate of the returned sludge from secondary sedimentation to aeriation tank but all of them rely on the results of tests that are done after the process. Therefore, determining dynamic estimation methods is very important. In the present study, artificial neural network (ANN) models and adaptive fuzzy-neural inference system (ANFIS) were used to achieve this goal. First, different compositions according to the quality parameters of wastewater such as sewage inlet flow, BOD5, temperature, TDS, TS and returned sludge flow with time delay were considered as input, and the amount of returned sludge as network output. Then, by training the network and determining the desired structure based on the type, number of membership functions and related laws, and using MATLAB software, the most appropriate model were obtained based on statistical data, the mean squared error and the efficiency of the coefficient of determination model. As a result, the inputs were introduced as the most suitable model by combining a one-dimensional Sugeno inference system with relevant membership functions. The results of different methods were compared and finally, a Genfis2 model, which is moderation of ANFIS systems, with a training coefficient above 93% (MSE = 0.0081 and RMSE = 0.0898) and a validation coefficient above 91% (MSE = 0.0027 and RMSE = 0.0518) was selected and presented for accurate estimation of the amount of returned sludge up to the next 24 h. This study was done with comprehensiveness and practicality for the first time in Iran and could lead to prevention of polluting the receiving waters.

中文翻译：

---

使用人工神经网络和模糊推理系统估计回流污泥（案例研究：德黑兰 Shahrake-Gharb 废水处理厂）

摘要

污泥回流量被认为是污水处理厂运行中重要且可控的参数之一，在处理过程中起着至关重要的作用。有不同的方法可以测量从二沉池返回到曝气池的污泥速率，但它们都依赖于处理后的测试结果。因此，确定动态估计方法非常重要。在本研究中，人工神经网络（ANN）模型和自适应模糊神经推理系统（ANFIS）被用来实现这一目标。首先，根据污水进水流量、BOD5、温度、TDS、TS和延时回流污泥流量等废水质量参数的不同成分作为输入，回流污泥量作为网络输出。然后，通过训练网络并根据隶属函数的类型、数量和相关规律确定所需的结构，并使用MATLAB软件，根据统计数据、均方误差和确定系数的效率得到最合适的模型模型。结果，通过将一维 Sugeno 推理系统与相关隶属函数相结合，将输入作为最合适的模型引入。比较了不同方法的结果，最后得到了一个 Genfis2 模型，它是 ANFIS 系统的调节模型，训练系数在 93% 以上（MSE = 0.0081 和 RMSE = 0.0898），验证系数在 91% 以上（MSE = 0.0027 和 RMSE） = 0.0518) 被选择并呈现用于准确估计到下一个 24 小时的返回污泥量。