基于反向传播神经网络的电化学强化厌氧膜生物反应器膜污染预测模型

Membrane Fouling Prediction Model of Electrochemical-enhanced Anaerobic Membrane Bioreactor Based on Back Propagation Neural Network

厌氧膜生物反应器(AnMBR)在高效处理污水的同时能够捕获污水中的能量,产生清洁能源甲烷,对实现“碳中和”目标具有重要意义。膜污染问题是制约AnMBR在市政污水中大规模工程应用的首要挑战。基于电化学调控的AnMBR是实现膜污染控制耦合高效产能的一种潜在途径。本文构建了电化学强化AnMBR反应体系,收集反应器连续运行试验数据,基于反向传播神经网络(BPNN)理论,建立单层多节点隐含层的BPNN模型。采用两种不同方式分割数据集,经过多次训练实现模型性能的优化,可将已有水质时间序列数据作为输入,对未来的膜污染时间序列进行预测。结果表明,跨膜压差(TMP)与pH值、氧化还原电位(ORP)未呈现出显性关联,但其本身表现出典型的时间序列数据特性。所构建的BPNN膜污染预测模型误差能够达到1e-10以下,预测精确度接近100%,可为电化学强化AnMBR系统的运行管理提供有力的支持。

Anaerobic membrane bioreactor(AnMBR)is of great significance in achieving the goal of"carbon neutrality"as it efficiently treats wastewater while capturing the energy in the wastewater to produce clean energy methane.The membrane fouling issue is the primary challenge restricting the large-scale application of AnMBR in municipal wastewater treatment.An electrochemical-regulated AnMBR is a potential approach to achieve membrane fouling control and high-efficiency production.An electrochemical-enhanced AnMBR reaction system was constructed,and continuous operation test data of the reactor were collected.Based on the theory of backpropagation neural network(BPNN),a BPNN model with a single layer and multiple node hidden layers was established.By dividing the dataset in two different ways and optimizing the model performance through multiple trainings,the BPNN model can predict the future membrane fouling-related time series based on the input of existing water quality time series data.The results show that the transmembrane pressure(TMP)does not show a dominant correlation with pH value and oxidation-reduction potential(ORP),but it exhibits typical time series data characteristics.The constructed BPNN membrane fouling prediction model can accurately predict the changes in membrane fouling degree,with an error below 1e-10 and a prediction accuracy close to 100%,providing strong support for the operation and management of electrochemical-enhanced AnMBR systems.