膜生物反应器(Membrane Bio-Reactor, MBR)处理污水是一个复杂的动态过程，难以用数学模型直接建模。针对该问题，本文利用差分进化算法(Differential Evolution Algorithm, DE)优化的循环神经网络(Recurrent Neural Network, RNN)对污水处理过程的膜通量进行预测。首先运用主成分分析法确定影响膜通量的相关过程变量；然后用DE算法优化RNN的初始权值和阈值；最后用训练好的DE-RNN模型进行预测并与样本数据对比。结果显示，该模型对膜通量地预测有着较高的准确率，具有很好的自适应性，达到了预期目标。 Wastewater treatment by membrane bio-reactor (MBR) is a complex dynamic process, which is difficult to be modeled directly by mathematical model. To solve this problem, this paper uses the recurrent neural network (RNN) based on differential evolution algorithm (DE) to predict membrane flux in wastewater treatment. First, authors used principal component analysis to determine the process variables affecting membrane fouling; and then used differential evolution algorithm to optimize initial value of RNN model; finally, the trained DE-RNN model was used to predict and compare with the sample data. The results show that the model has high accuracy, good adaptability and achieves the expected goal.
展开▼
机译：膜生物反应器(Membrane Bio-Reactor, MBR)处理污水是一个复杂的动态过程，难以用数学模型直接建模。针对该问题，本文利用差分进化算法(Differential Evolution Algorithm, DE)优化的循环神经网络(Recurrent Neural Network, RNN)对污水处理过程的膜通量进行预测。首先运用主成分分析法确定影响膜通量的相关过程变量;然后用DE算法优化RNN的初始权值和阈值;最后用训练好的DE-RNN模型进行预测并与样本数据对比。结果显示，该模型对膜通量地预测有着较高的准确率，具有很好的自适应性，达到了预期目标。 Wastewater treatment by membrane bio-reactor (MBR) is a complex dynamic process, which is difficult to be modeled directly by mathematical model. To solve this problem, this paper uses the recurrent neural network (RNN) based on differential evolution algorithm (DE) to predict membrane flux in wastewater treatment. First, authors used principal component analysis to determine the process variables affecting membrane fouling; and then used differential evolution algorithm to optimize initial value of RNN model; finally, the trained DE-RNN model was used to predict and compare with the sample data. The results show that the model has high accuracy, good adaptability and achieves the expected goal.
展开▼
