The Research And Application Of Imbalance Classification Algorithm Based On Kernel Strategy And Deep Learning

Working condition identification is the key to realize stable control of production process,improve product quality,save energy and reduce consumption.In the industrial site,the probability of abnormal working conditions is low,and the phenomenon of category imbalance is widespread.At the same time,thermal time series data has the characteristics of multivariable,large lag,and strong coupling,and the task of identifying working conditions under unbalanced data is very difficult.Aiming at the above problems,this paper forms an imbalanced classification algorithm based on nuclear strategy and deep learning,which can effectively solve the problem of imbalanced multivariate time series classification in the identification of complex industrial process conditions.The algorithm in this paper is mainly divided into two parts: the first part is the discriminator part,which is mainly composed of the Classification Algorithm for Highly Imbalanced Data based on Data Distribution in Kernel Space(KSDD);the second part It is the feature extraction part,which mainly uses the fusion multivariate attention mechanism cyclic neural network and the fully convolutional neural network to extract the features of the time series data.The main work of the thesis is as follows:(1)Aiming at highly imbalanced data samples,this paper proposes the KSDD algorithm.Among them,a calculation method of nuclear space data distribution is used,which replaces some previous methods that use support vectors and sample interval information to reflect data distribution.This calculation method ensures the accuracy of subsequent under-sampling and nuclear correction operations.The algorithm combines under-sampling and kernel correction by improving the optimal interval distribution machine.The algorithm can effectively improve the classification accuracy of the minority class while avoiding the loss of too much information in the majority class.Because under-sampling reduces the amount of data for most classes and does not require pre-training classifiers,the training time of the algorithm can be greatly reduced.Finally,the experimental results on the standard data set show that the algorithm in this paper is significantly better than other compared algorithms.(2)Aiming at the imbalanced time series classification problem in complex industrial processes,this paper adopts an imbalanced classification algorithm based on nuclear strategy and deep learning.The feature extraction part combines the recurrent neural network and the fully convolutional neural network,which can not only obtain long-distance dependence information in the time series,but also obtain the local detailed feature information of the time series through the convolutional layer.Through end-to-end learning,the hidden features in the time series are extracted.The discriminator part uses the KSDD algorithm to replace the fully connected layer for category judgment,which can effectively solve the problem of category imbalance.(3)In order to verify the effectiveness of the algorithm,this paper verifies the imbalanced classification algorithm based on nuclear strategy and deep learning in the actual complex industrial site.Experiments were conducted on the data set collected on the alumina rotary kiln.The experimental results show that the overall recognition accuracy of this method is 97.19%,the F1 index is 95.76%,and the G-means index is 95.90%.Compared with other methods,the recognition accuracy has been significantly improved.