A Novel Fault Detection Research Based On Improved AdaBoost With Double Feedback Elm For Industrial Process

In industrial process,production link is closely related.And the extent of harm will increase and cause significant economic losses and even casualties in the system.It is very important to adopt appropriate and efficient fault detection methods.With the development of industrial technology,a large number of process data relevant fault are collected.How to effectively analyze and apply the information contained in the process data to improve the accuracy of fault detection is important.Therefore,in the paper,the proposed method mainly improves from the following aspects.(1)Considering that AdaBoost is widely used in fault detection,in order to make it high classification accuracy,the proposed method of Double Feedback Extremely Learning Machine(DFELM)adds self-feedback layer and input-output feedback layer of the hidden layer on ELM.The output weight of the hidden layer is adjusted by using the trend feature factor of the self-feedback layer and the output weight of the self-feedback layer.The error coefficient and variation coefficient of the input-output feedback layer adjust the connection weight and bias between the input layer and the hidden layer.By adjusting the output weight,the connection weight and bias,the classification accuracy of DFELM is improved.(2)Considering that an effective method reduces the number of iterations and improving the classification accuracy for AdaBoost,an improved AdaBoost method based on local selective-ensemble is proposed.it can make full use of the effective information contained in the weak classifiers with the poor performance.It divides weak.classifiers into the good weak classifiers and the poor weak classifiers by the error rate.A new sample set with high similarity with specific samples is obtained by the nearest neighbor method and inputs them into the poor weak classifiers with good classification performance for specific samples.The strong classifiers are formed by selectively integrating the good weak classifiers and the poor weak classifiers tied with good local classification performance.The improved AdaBoost uses the effective information of the weak classifier.So it both improves the classification accuracy and decreases iterations times of the weak classifier.(3)Combined with Tennessee Eastmanprocess,the simulation experiment is used to verify the effect.The final experimental results show that the proposed method has high effect in improving the fault detection performance compared with other methods in fault detection.