Research On Prediction Modeling Method Of Key Indicators In Wastewater Treatment Process

The monitoring of effluent indicators and the level of dissolved oxygen(Dissolved Oxygen,DO)of sewage treatment plants are the two most critical components in the operation and management of sewage treatment plants.Mastering the changes in the above two types of key indicators can ensure the efficient and stable operation of the sewage treatment system on the one hand,and on the other hand,excessive energy loss can be avoided.In the Internet era,data-driven technologies have been widely used to explore the characteristics of data samples in various fields.The traditional management approach has begun to shift to a digital management model.In this process,machine learning intelligent prediction methods play a central role.However,it is difficult for the existing conventional intelligent management methods to adapt to the dynamic changes in the sewage treatment process in the actual situation.Based on the above problems,the real operation engineering data of a sewage treatment plant in Chongqing was collected,and the sewage treatment process(Anaerobic-Anoxic-Oxic,A~2/O)was used as the experimental scenario.For the key indicators of sewage effluent quality and DO in the sewage treatment process,two intelligent prediction methods based on parameter optimization were proposed,with a view to improving the prediction performance of the prediction method on the dynamic sewage treatment process.The work of this article can be summarized into the following three aspects:(1)A set of automated data cleaning and preprocessing methods was designed using computer programming,and the two types of key index data collected were sequentially removed from outlier and duplicate values,as well as normalized calculation processing and ratio calculation processing and other preprocessing operations.The method can not only quickly achieve the requirement of processing a large number of data samples in the preliminary data preparation,but also quickly migrate to other large-scale business data of similar scenarios,thus realizing the cleaning of massive data.(2)To address the problems that the sewage effluent quality cannot be measured in time and the machine learning trained by data samples lacks dynamic adaptiveness to scene changes,the abstract description of the sewage treatment process by neural network model was studied.The intelligent management model of the A~2/O sewage treatment process based on a neural network was constructed to improve sewage effluent quality prediction performance.Firstly,convolutional neural network and gated recurrent unit were used to extract data spatial features and temporal features,respectively,and a hybrid network model was successfully constructed to realize the prediction of key water discharge indexes.Secondly,the data update optimization method was introduced to retrain the original model and adjust the parameters using the updated data from the actual project operation for continuous optimization of the model,and then made the model cope with the dynamic sewage treatment process with long-term advantages.Finally,the training and test sets of historical and updated data were set,and the efficiency and stability of the prediction model under this model were verified by a series of experiments.(3)A DO prediction model based on support vector regression(Support Vector Regression,SVR)was constructed for the prediction of DO concentration in sewage treatment process.The model aims to achieve the prediction of the appropriate DO values under normal operation of the sewage treatment process,and thus realize the energy-efficient operation of the sewage treatment process.In addition,particle swarm optimization(Particle Swarm Optimization,PSO)algorithm was used to optimize the model parameters.First,the radial basis kernel function method of SVR was used to fit the nonlinear function relationship between DO value and relevant characteristic variables,so as to effectively predict DO value in A~2/O aerobic region,and the PSO algorithm was used to optimize the parameters of the constructed model.Furthermore,the experimental results confirmed that the prediction results of the model were in good agreement with the actual values by comparing with the baseline model and had the advantage of adaptability to dynamic scenarios.