Study Of Soft Sensor Modeling For Wastewater Treatment Process Base On Relevance Vector Machine

Wastewater treatment is becoming more and more complex and parameters have time-varying, multivariable, strong coupling, nonlinear, big lag and other characteristics. In face of these characteristics, the traditional sensor cannot be applied effectively, so important variables cannot be quickly and accurately measured. Soft measurement technology is a new type of intelligent detection technology which has a very broad application prospects in industrial process and control system. Based on the wastewater treatment process and the biochemical and process knowledge, this paper studied some modeling methods of soft measurement. In order to solve the problem of low accuracy, weak robustness and small scope of application existed in the traditional soft measurement modeling method, this paper studied a kind of soft measurement of modeling method based on the relevance vector machine regression, and successfully established the soft measurement model of effluent quality, simulation results verify the effectiveness of the proposed method.First, the basic principle of soft measurement technology and a literature review on the application of machine learning algorithm in the soft measurement and analysis in the field of sewage at home and abroad research status are introduced. Then the paper points out the main problems existed in the process of soft measurement modeling.Secondly, the principle of relevance vector machine and its soft measurement model of regression are introduced. Due to the reason that data dimension of wastewater is high, building the soft measurement model is very difficult, this paper adopted a kind of attribute reduction method of fuzzy rough monotone increasing, and then obtaining several kinds of input attributes which has a greater impact on the sewage effluent BOD5. On this basis, a soft measurement model predicting effluent BOD5 is proposed based on RVM fuzzy rough monotone increasing method.Then, in order to improve the prediction accuracy of the model, the kernel function method is studied. Combining global kernel function and local kernel function, a relevance vector machine regression model based on hybrid kernel function is proposed. To solve the problem of parameters optimization, the artificial immune optimization algorithm is applied to the model, giving the realization process in detail. The simulation results show that the proposed hybrid kernel function model prediction accuracy is significantly higher than the single kernel function model, and the artificial immune optimization algorithm has a very fast convergence speed and strong global search ability.Finally, according to the needs of working real-time in wastewater treatment, the online soft measurement model based on relevance vector machine is proposed. First, with the fast marginal likelihood algorithm to quickly determine the parameters in the model, the algorithm implementation steps are obtained, and test its performance. The fast relevance vector machine is used in sewage on-line soft measurement model. The simulation results show that the model improves the prediction accuracy, and at the same time it reduces the time of model updating, well meets the requirement of the real-time performance.