Research And Application Of Soft Sensor Modeling Based On Online Incremental LSSVM Sewage

Today’s society,the shortage of fresh water resources and the pollution of the water environment remind us of the need for sewage treatment.Sewage treatment is a complex modern industrial process with the characteristics of strong non-linearity,large hysteresis and large inertia.The process of complex and diverse makes it difficult to measure some important process variables which can reflect the water quality index in the sewage and predict the output according to these process variables,those current existing testing devices or because of the cost,or because of precision,can not give a very reliable guarantee.However,as the extension and development of traditional detected technology,soft sensor technology applied to the actual sewage treatment process will have a very important practical significance.In this paper,as the research object,the biological aerated filter,established a soft soil water measurement model to achieve the effluent COD(chemical oxygen demand)concentration forecast,to determine the quality of water and make the appropriate measures to provide advanced criteria.The main contents are as follows:First of all,based on the advantages of Least square support vector machine(LSSVM)compared with traditional support vector machine(SVM),this paper chooses LSSVM as the basic algorithm of this model,and improves it from two aspects: model kernel function and parameters,on the one hand,the linear weighted mixed kernel function is used to replace the single kernel function.On the other hand,the improved intelligent optimization algorithm is used to optimize the parameter combination.Simulation comparison analysis,the two aspects of the improvement of the model accuracy has a certain degree of improvement.Then,in view of the fact that the overall predictive effect will deteriorate when the offline model has confronted with large-scale real-time updates,in this paper,we use an improved online incremental LSSVM algorithm,which uses predictive error for selective incremental learning and matching pruning algorithms.When the new sample is added,it is first determined whether the incremental learning is needed according to the prediction error threshold.Incremental learning for samples that exceed the threshold,updating the support vector in a recursive way,not only avoids the standard inversion operation,but also shortens the computation time and improves the efficiency of the forecast.At the same time,when the size of the sample reaches acertain scale,the pruning operation of the earliest sample is chosen so that the sample length is always maintained at a certain scale.The main contributions and innovations of this topic are as follows: this paper chooses to improve the LSSVM soft-sensing model from both accuracy and on-line performance.The combination of a linear weighted mixed kernel function and an improved intelligent optimization algorithm resulting in a significant degree of improvement in model accuracy;through clever iterations,threshold selection,time window setting,the improved online incremental LSSVM algorithm has made a certain degree of improvement of the rapidity and on-line performance of the model.Both improvements have achieved an on-line estimate of effluent COD precision.