Prediction Of Effluent Total Nitrogen In Wastewater Treatment Plant Based On BP Neural Network Model

Treatment of wastewater plants involves multi-domain processes and monitoring of dynamic water quality data is a crucial issue.TN is the main pollution indicator in wastewater treatment,measuring using manual experiments takes time and effort;on-line automatic monitors are expensive and require maintenance.The neural network model can be based on the optimal criteria,according to the relevant variables to infer the dominant variables,the model can be used to predict the outflow TN.Wastewater treatment plant data in the model input,need to be pre-processed and in-depth analysis before modeling to solve the model due to data clutter,redundancy and lack of predictability caused by the model's poor performance.In this paper,the operating data of the A~2/O process wastewater treatment plant with a processing capacity of 60,000 tons of water is used to perform preprocessing through error processing,null filling and smooth denoising.Using PCA method and rough set theory method,using MATLAB and ROSETTA software to simplify and refine the data,using MATLAB software to build BP neural network model,select the appropriate node transfer function and training function to determine a reasonable number of principal components to obtain the prediction accuracy Higher effluent TN prediction model for wastewater treatment plant,and based on the simulation results,proposed optimization adjustment suggestions for the operation parameters of the wastewater treatment plant.The main conclusions of this paper are as follows:(1)By comparing different neural network structures,the BP neural network model with one hidden layer and seven neurons has the advantages of convenient use and accurate prediction,and selects better functions and parameters for the adoption of a wastewater treatment plant in Wuhan City.After six months of production data training,the prediction accuracy can reach 98%.(2)The principal component analysis method combined with neural network can improve the accuracy of model prediction by reduce the dimensionality,it is recommended to use.However,the denoising method has no obvious effect on improving the accuracy of the model,and it is easy to lose the feature data.(3)According to model prediction results,the model was applied to the adjustment of the parameters of the wastewater treatment plant,and the dissolved oxygen value of the A~2/O aerobic section and the MLSS value of the biological pool were appropriately increased,and the removal rate of TN was increased.