Research And Application Of Neural Network Soft-sensing Model For Papermaking Wastewater Biochemical Treatment In A~2/O Process

With major adjustments of the world economic pattern and the markedacceleration of China's economic and social transformations, environmentalconstraints faced by paper industry in China are increasingly severe. In order tosuccessfully achieve the transformation of traditional paper industry into sustainabledevelopment modern paper industry and significantly reduce the emission of paperpollutants, National Development and Reform Commission of People's Republic ofChina proposed the project of online monitoring system renovation in wastewatertreatment plant to be a priority construction project during National Twelfth Five-YearPlan period. It is encouraged to develop and improve the monitoring system, and toadopt strengthen the function of nitrogen and phosphorus removal to meet theGB3544-2008discharge standard for papermaking wastewater treatment. Soft-sensingis a currently major research focus in the fields of process control and processdetection, and the soft-sensing system can either replace hardware sensors or be usedin parallel with them to ensure the accuracy of measurement. Soft-sensing canimprove the performance of process monitoring and control in papermakingwastewater treatment with the result that enhancing the effect of papermakingwastewater treatment and reducing pollutant emission.Based on the analysis of neural network soft-sensing model application status inwastewater treatment, and basic structures and algorithms of back-propagation neuralnetwork (BP-ANN), genetic neural network (GA-ANN),and adaptive network basedfuzzy inference system (ANFIS), ideas and methods of constructing the optimal ANNsoft-sensing model were studied systematically. Simultaneously, BP-ANNsoft-sensing model, GA-ANN soft-sensing model, and ANFIS soft-sensing modelwere employed to predict the effluent chemical oxygen demand (COD_eff) and effluentammonia nitrogen (NH₄⁺_eff) of papermaking wastewater treatment in ananaerobic/anoxic/oxic (A²/O) process. Some pioneering and exploratory work on thesoft-sensing model used in papermaking wastewater treatment, and the constructionand implementation scheme of the ANFIS-based dissolved oxygen intelligent optimal control system were studied. Simultaneously, the control system was used to adjustdissolved oxygen for papermaking wastewater treatment in an A²/O process under thelaboratory condition, which provided a reference to improve the level of intelligentcontrol in papermaking wastewater treatment. Main contents and results are asfollows:1. Based on the analysis of A²/O biological nutrient removal process andaccording to papermaking wastewater treatment requirements, equipment selection,hardware installation, PLC control program design, and industrial control softwareMCGS configuration were completed to construct an embedded papermakingwastewater treatment automatic control system.2. Based on the analysis of basic structures and algorithms of neural network andthe principle of genetic algorithm, ideas and methods of constructing the optimalBP-ANN soft-sensing model were studied systematically. Meanwhile, the optimalBP-ANN soft-sensing model and the optimal GA-ANN soft-sensing model wereconstructed to predict COD_effand NH₄⁺_effof papermaking wastewater treatment in anA²/O process. Compared with BP-ANN soft-sensing models, GA-ANN soft-sensingmodels presented better estimate performance.3. Based on the analysis of structures and algorithms of adaptive network basedfuzzy inference system and the principle of fuzzy c-means (FCM) clusteringalgorithm, a new validity function B (c) was introduced to FCM clustering algorithmto intelligently optimize fuzzy rules of the ANFIS soft-sensing model. Eventually,optimal ANFIS soft-sensing models with9fuzzy rules were constructed to predictCOD_effand NH₄⁺_eff. When predicting, the maximum relative error absolute valuesbetween the observed and predicted values of COD_effand NH₄⁺_effwere5.4014%and6.6513%, respectively; root mean square errors of1.6317and0.1291for COD_effandNH₄⁺_effcould be achieved; mean absolute percentage errors for COD_effand NH₄⁺_effwere1.8458%and2.8984%, respectively; correlation coefficient values of0.9928and0.9951for COD_effand NH₄⁺_effcould also be achieved. The results indicated thatcompared with GA-ANN soft-sensing models, ANFIS soft-sensing models presentedbetter estimate performance; reasonable monitoring A²/O process performance just using on-line monitoring parameters, namely hydraulic retention time (HRT), influentpH (pH), dissolved oxygen in the aerobic reactor (DO) and mixed-liquid return ratio(r), has been achieved through the ANFIS soft-sensing model.4. The ANFIS-based dissolved oxygen intelligent optimal control system wasconstructed in MATLAB software, it achieved to exchange data between MCGS andMATLAB using OPC communication technology with the result that the controlsystem can be performed in MCGS. Simultaneously, it was used to adjust dissolvedoxygen for papermaking wastewater treatment in an A²/O process. The resultsdemonstrated that the observed COD_effwas between60.86mg/L and78.77mg/L whenthe desired COD_effwas set to70mg/L. It indicated that the intelligent control systemcan dynamically optimize dissolved oxygen to consequently meet the dischargestandard and steady effluent quality.