Research On The Application Of Mixed Feature Selection Method In Dust Concentration Monitoring Model Of Power Plant

Dust,as well known,will not only pollute the environment,but also threat human's health.At present,with the increasing demand for air quality of our residents,the increase of the installed capacity of power generation,the higher requirements of the state of the emission requirements of the power plant emissions,more stringent dust control policy,the dust concentration of accurate measurement and monitoring is becoming more important.The timely monitoring of coal dust concentration in coal-fired power plants can make power plants and environmental protection departments in a timely manner to better grasp the dust concentration of the situation.Then on the dust emissions do not meet the limits of the power plant,they can require its production to adjust,and the production of dust and dust treatment.It plays a very positive and important role to improve the quality of the environment and protect personal safety.With the power installed capacity increased year by year,power plant dust concentration control is increasingly important.The introduction of soft-sensing technology can make up for the existing instrumentation and other hardware measurement of low reliability,high cost shortcomings.The establishment of dust concentration monitoring model based on the feature selection intelligen t algorithm is very important for the effective prediction and monitoring of dust emission concentration.In this paper,an F-SSFS hybrid feature selection method is used to select the feature.Based on the soft-sensing method,the prediction model of dust concentration is established,and the accuracy and generalization ability of this method are verified.This hybrid feature selection method combines the F-score and the support sequential forward search(SSFS).We want to combine the advantages of the filtering method and the packaging method and choose the best subset of feature parameters from the original set of features.The prediction results are compared with the classical feature selection method-mean influence method(MIV).The prediction results of the two methods in the SVM model are compared with those in the reverse propagation neural network(BPNN)model.According to the experimental results,it can be seen that the F-SSFS mixed feature selection method has higher accuracy than the average effect method(MIV),and the model established by SVM is better than BPNN used to predict the dust concentration.Therefore,the SVM model based on the F-SSFS hybrid feature selection method has the highest predictive accuracy and generalization performance level,which can provide a reference for the prediction model of thermodynamic system construction.