基于深度学习的污水处理厂出水总磷预测方法

Prediction of total phosphorus in wastewater treatment plant effluent based on deep learning

水体中磷含量过高是造成水体富营养化的关键因素之一,故总磷是污水处理的一项重要水质控制参数,传统总磷测定方式无法实现对出水总磷的实时监测,不利于处理过程的智能化发展。使用BP神经网络(BPNN)、卷积神经网络(CNN)、长短期记忆递归神经网络(LSTM)、Informer模型建立了污水处理厂出水总磷预测模型,分析表明:BPNN模型的R2为0.459 7,模型预测结果平稳性差;CNN模型的各项评估指标都较差,不适用于对污水处理厂出水总磷的预测;LSTM模型的均方误差(MSE)、均方根误差(RMSE)、平均绝对误差(MAE)和R^(2)分别为0.008 2、0.090 5、0.068 4和0.606 8,模型预测精度较高;相较于LSTM模型,Informer模型的MSE、RMSE、MAE分别降低了21.95%、11.60%、28.65%,R2提高了19.94%,具有明显的预测优势。Informer模型预测精度高且泛用性强,预测结果平稳性好,能有效预测污水处理工艺的出水总磷,对于污水处理厂提高实时智能化水平、优化处理工艺、提高除磷效率、减少能耗和实现碳中和具有重要意义。

Excessive phosphorus in water is one of the key factors causing eutrophication in water bodies.There-fore,total phosphorus is an important water quality control parameter for sewage treatment.Traditional total phos-phorus testing methods can not realize real-time monitoring of effluent total phosphorus,which is not conducive to the intelligent development of treatment process.This paper used back propagation neural network(BPNN),convolu-tional neural network(CNN),long short-term memory recurrent neural network(LSTM),and Informer to establish a prediction model for total phosphorus in sewage treatment plant effluent.The analysis showed that the R2 of the BPNN model was 0.4597,and the prediction results of the model were poorly stationary.The evaluation indicators of the CNN model were poor,and it was not suitable for the prediction of total phosphorus in the sewage treatment plant effluent.The mean square error(MSE),root mean square error(RMSE),mean absolute error(MAE),and R^(2) of the LSTM model were 0.0082,0.0905,0.0684 and 0.6068 respectively,and the model prediction accuracy was high.Compared with the LSTM model,the MSE,RMSE,and MAE of the Informer model were reduced by 21.95%,11.60%,and 28.65%,respectively,and the R2 was increased by 19.94%,which had obvious prediction advantages.The Informer model had high prediction accuracy and strong universality,with good stability in prediction results.The Informer model could effectively predict the total phosphorus in wastewater treatment plant effluent,which was of great significance for improving real-time intelligence level,optimizing treatment process,improving phosphorus removal efficiency,reducing energy consumption,and achieving carbon neutrality in wastewater treatment plants.