Abstract:

This study is aimed at addressing the challenges posed by the intricate internal mechanisms of wastewater treatment and the difficulty in achieving effective real-time control of effluent quality through online monitoring. A hybrid prediction model for effluent quality is proposed, combining data decomposition with an improved dung beetle optimizer (DBO)-optimized TCN-BiGRU/BiLSTM architecture. The correlation analysis method is used to select the variables with strong correlation with the outflow index from the inflow variables as the auxiliary input features of the prediction model. The effluent quality time series were decomposed and simplified into several subsequences using variational mode decomposition (VMD). The sample entropy of each subsequence was calculated, and the subsequences were classified into high and low complexity levels based on this measure. Accordingly, two hybrid prediction models, TCN-BiLSTM and TCN-BiGRU, were constructed. An enhanced DBO algorithm incorporating Tent chaotic mapping and Cauchy mutation strategies was introduced to optimize the combined model. Comparative experimental results demonstrate that in predicting effluent total nitrogen (TN) and chemical oxygen demand (COD), the proposed model outperformed the CNN-LSTM, VMD-TCN-BiGRU, VMD-TCN-BiLSTM, and VMD-TCN-BiGRU/BiLSTM models. It achieved reductions in average RMSE and MAE ranging from 35.22% to 52.41% and 39.38% to 55.53%, respectively, and increased the average R² by 2.91% to 7.55%. The prediction accuracy of the model is significantly improved, and it performs well for nonlinear complexity problems in measured data, and has good engineering application value.

Key words: algorithm, optimization, prediction, wastewater treatment

摘要：

针对污水处理内部机理错综复杂，出水水质难以实时检测和有效控制的问题，提出了一种基于数据分解与改进蜣螂优化（DBO）TCN-BiGRU/BiLSTM的出水水质组合预测模型。采用相关性分析法在进水变量中选出与出水指标强相关的变量，作为预测模型的辅助输入特征；通过变分模态分解（VMD）对出水水质序列进行分解，简化为若干子序列，并计算每个子序列的样本熵，将其按照复杂度划分为高、低两类，据此构建出TCN-BiLSTM和TCN-BiGRU两种混合预测模型；引入Tent混沌映射和柯西变异策略改进的DBO算法对组合模型进行优化。对比实验结果表明，在出水总氮（TN）和化学需氧量（COD）的预测中，相较于CNN-LSTM、VMD-TCN-BiGRU、VMD-TCN-BiLSTM和VMD-TCN-BiGRU/BiLSTM模型，所提出的模型平均RMSE和MAE分别降低35.22%~52.41%和39.38%~55.53%，平均R²提高2.91%~7.55%，模型预测精度明显提高，且对于实测数据中的非线性复杂性问题表现出色，具有良好的工程应用价值。

关键词: 算法, 优化, 预测, 污水处理