In the process of electrochemical wastewater treatment, the removal rate of electrocoagulation reactor will be affected by various factors such as the pH value of wastewater solution, the current density, the wastewater flow rate and the initial concentration of heavy metal ions. Therefore, this study proposes a prediction method of the removal rate of the electrocoagulation reactor based on deep learning Long and Short-Term Memory (LSTM) network combined with the Autoregressive Integrated Moving Average Model (ARIMA) commonly used in engineering. Firstly, according to the concentration of heavy metal ions in the outlet and inlet solution of the reactor, the calculation formula for the removal rate of the reactor is defined. Secondly, in order to deepen the LSTM network model to analyze and learn the change trend of the historical removal rate data, the gradient value of the historical removal rate of the reactor before and after the change is extracted as its change feature value, and this feature value is taken as one of the input variables of the LSTM network model. Comprehensive analysis considered important factors such as the historical removal rate value of the reactor, the initial pH value of the wastewater solution, the voltage and current value, and the wastewater flow rate as the input variables of the LSTM deep learning network. The predicted value of the removal rate of electrocoagulation reactor is concluded by testing the combination of activation function and the number of fully connected layers, and the error compensation of the predicted value is carried out by using the ARIMA model. The effectiveness of the proposed method is verified by the industrial data collected from a wastewater treatment plant.

在电化学废水处理过程中，电凝反应器的去除率会受到废水溶液pH值、电流密度、废水流速和重金属离子初始浓度等多种因素的影响。因此，本研究提出了一种基于深度学习长短期记忆（LSTM）网络结合工程中常用的自回归综合移动平均模型（ARIMA）的电凝反应器去除率预测方法。首先，根据反应器出口和入口溶液中重金属离子的浓度，定义了反应器去除率的计算公式。第二，为了深化LSTM网络模型分析学习历史去除率数据的变化趋势，提取变化前后反应器历史去除率的梯度值作为其变化特征值，并将该特征值作为 LSTM 网络模型的输入变量之一。综合分析将反应器历史去除率值、废水溶液初始pH值、电压电流值、废水流量等重要因素作为LSTM深度学习网络的输入变量。通过测试激活函数和全连接层数的组合得出电凝反应器去除率的预测值，利用ARIMA模型对预测值进行误差补偿。从污水处理厂收集的工业数据验证了所提出方法的有效性。