Research On Fault Diagnosis Of Process Industry Based On Neural Network

The fault diagnosis of process industry plays a significant role in ensuring the safety of industrial production and improving the quality of products.Based on the research of neural network,this paper studies the method of fault diagnosis for process industry,which has the characteristics of complex data distribution and high dimension.The main work includes:(1)High dimensional and related characteristics of process industrial data will affect the accuracy of diagnostic results.So we use principal component analysis reduce the dimension of data to remove correlation,then input the processed data into learning vector quantization neural network for fault diagnosis,The Tennessee-Eastman process simulations show that this algorithm has a higher recognition rate.(2)This paper uses particle swarm optimization algorithm to determine the smoothing parameter of probabilistic neural network;according to the standard PSO algorithm is easy to fall into local extremum,proposes a nonlinear change of inertia weight instead of linear change of inertia weight to improved particle swarm optimization algorithm(IIWPSO),and finally applied to probabilistic neural network for fault diagnosis.Numerical and TE process simulations identify its effectiveness.(3)In practical applications,operators usually the ordinary technical workers and lack of neural network computing experience,it is difficult to guarantee the actual results,therefore,this paper used the neural network ensemble algorithm.First,the Bagging algorithm is used to generate multiple individual networks,then the IIWPSO-PNN algorithm is used to train the network and generate multiple classifiers,Finally,the result is generated by the relative majority voting method.Simulations of UCI standard machine learning library and TE process show the effectiveness of the proposed method.