Research On COD Soft Sensing Technology In Livestock And Poultry Wastewater

In the industrial process,due to the influence of cost and detection technology,many variables to be measured cannot be accurately detected in real time by on-site equipment,and laboratory detection also faces the problems of high cost and high lag.It is a process of multi-condition characteristics of livestock and poultry breeding industry and sewage treatment industry.As an important indicator of water quality parameters,chemical oxygen demand(COD)reflects the degree of organic pollution in water samples.Measurement,the traditional detection method is obviously not conducive to the development of the enterprise and the protection of the environment.The soft measurement method not only replaces the hardware conditions of chemical monitoring,but also greatly reduces the material cost and time cost,which is important for the sewage treatment process and other fields.meaning and value.The main research problem of this paper is to establish a soft-sensor model that can resist the disturbance of the open air environment of livestock and poultry farms and sewage treatment plants.First,after analyzing the characteristics of livestock and poultry breeding wastewater and the sewage treatment process,based on the activated sludge method to treat livestock and poultry breeding wastewater,the auxiliary variables required by the soft sensor model were determined,and the data used in the biochemical tank for sewage treatment were installed.Collect sensors,and perform outlier processing,normalization processing and correlation analysis on the collected data,as the data set of the soft sensor model.Secondly,the partial least squares regression,BP neural network and radial basis function neural network among the commonly used data-driven soft-sensing detection models for sewage treatment are studied as comparative models,providing basic theoretical introduction and reference for subsequent models.Shallow feedforward neural network is currently the most popular method used in soft measurement models,but in the application of sewage treatment,due to the complex interference environment of weather,the data set is more complex,and the model is often affected in actual prediction.Deviation from the actual value,so this paper proposes a soft-sensor model based on a stacked self-encoding deep network,adjusts the network to a sewage treatment-oriented structure,and applies a layer-by-layer greedy pre-training algorithm to solve the problem of difficult parameter training of deep neural networks.The unsupervised pre-training of input reconstruction and the supervised training of data set cascade fine-tuning are used to complete the parameter training of the deep neural network.Compared with the three commonly used data-driven soft-sensor models,their errors and decision coefficients are better than those of the commonly used models.The prediction effect on the interference data is good.At the same time,for the problem that the number of nodes in the hidden layer of the deep neural network is difficult to determine,the genetic algorithm is used to search for nodes.The validity of the soft-sensor model in the open air environment is verified by the data of a sewage treatment plant in Fujian.