Industrial Process Fault Classification Based On Stacked Extreme Learning Machine

In today's era,the progress of distributed computer control technology has greatly promoted the expansion of industrial process production scale.For the complex and huge industrial production process,efficient fault diagnosis technology is of great significance to ensure the safe operation of equipment.As an important branch of fault diagnosis technology,fault classification can effectively identify the specific fault types.Stacked extreme learning machine(SELM)is a classification algorithm based on deep learning thought,which can effectively solve the problem of fault classification in industrial processes.Based on the stacked extreme learning machine classification algorithm,an improved stacked extreme learning machine algorithm is presented and this further solves the problems of ignoring variable influence difference and unbalanced data classification in industrial process.The main work of this thesis is as follows:Firstly,aiming at the problem that the multi-layer structure of traditional stacked extreme learning machine retains limited information when passing information between layers,an improved stacked extreme learning machine based on mutual information(MI-SELM)fault classification method is studied.The traditional stacked extreme learning machine reserves limited node information when using the principal component analysis method to reduce the dimension of the hidden layer output matrix iterate to the next layer.It is impossible to iterate more effective information to the next layer.and by calculating the mutual information of the hidden layer output matrix can reserve more node information and iterate to the next layer.The simulation results on the typical industrial process dataset show that the improved method has better fault classification performance than the traditional method.Secondly,the traditional stacked extreme learning machine ignores the problem of variable influence difference,and proposes a stacked extreme learning machine based on the partial F-valued(MI-FSELM)fault classification method.The stacked extreme learning machine algorithm treats all process variables the same and does not further mine the information existing in the data.Stacked extreme learning machine based on partial F value calculates partial F value of each process variable and assigns certain weight information to the variable through partial F value.The simulation results of the method on the data set of typical industrial process conditions show that the method can fully mine the information contained in the variables and further improve the accuracy of fault classification.Finally,aiming at the fault data imbalance characteristics in industrial process,a fault classification method based on triple weighted stacked extreme learning machine(MI-TWSELM)is established.The method designs the weights from three aspects: the number of samples,the spatial distribution of the samples and the critical point data.The number of samples of different categories can be used to assign the first weight.By considering the spatial distribution of the sample,the sample can be given the second weight,further considering the impact of the critical point,the sample can be given the third weight.The simulation results on typical industrial process datasets and standard datasets show that the proposed method can better eliminate the influence of unbalanced data on the classifier and have better classification effect on the unbalanced dataset.