Research On The Mechanical Properties Prediction Method Of Aluminum Alloy Thin Sheet Strip Based On Ensemble Learning

The performance prediction model can be used for product performance optimization,new aluminum alloy model design and quality dynamic control in the production process of aluminum alloy sheet and strip.With the development of aluminum plate and strip production equipment,a large number of sensors are deployed on the production equipment.Using big data of production process to establish more accurate performance prediction model has become a research hotspot in the field of soft sensor modeling.Ensemble learning can improve the prediction accuracy and generalization performance of models.In recent years,it has been widely used in industrial process modeling.However,the current research often ignores the influence of the diversity of sub-models on the generalization performance of ensemble learning models.In view of the above problems,the research objectives of this paper mainly include: Firstly,based on the diversity of sub-models and the accuracy of sub-models,the sub-models are selected to improve the generalization performance of ensemble learning.Secondly,an ensemble algorithm for distributed parallel computing optimization is proposed to improve the computational efficiency of the ensemble learning model.Thirdly,the selective ensemble modeling method is applied to the performance prediction of aluminum sheet metal strip.The specific research contents are as follows:To solve the problem of sub-model selection in ensemble learning,a selective ensemble modeling method is proposed which considers the diversity among sub-models and the performance of a single model.First,a static ensemble selection method based on game theory is proposed.Taking the diversity contribution rate of the sub-model to the ensemble model and the precision of the single sub-model as the two sides of the game,the game theory principle is used to find the optimal selection scheme for the accuracy and diversity of the ensemble model.Secondly,a dynamic ensemble selection method based on approximate linear dependence is proposed.The predicted value of the sub-model on the sliding window data is taken as the feature attribute,and the diversity among the sub-models is evaluated by ALD condition.By synthesizing the prediction performance and diversity of the sub-model in the current window,the sub-model is dynamically ensemble and decreased to adapt to the characteristics of the modeling object when the integration model runs online.Furthermore,a weighted fusion method based on the historical error of the sub-model is proposed to fuse the prediction results of the multi-sub-model and further improve the generalization performance of the ensemble model.Finally,the rationality and validity of the proposed algorithm are verified by the open data set and the actual industrial data.In order to solve the problem of low computational efficiency of ensemble learning training and application in big data environment,a parallel algorithm of ensemble model is proposed.This paper first describes the parallelization method of online sequential extreme learning machine,and then parallelizes the ensemble framework,the parallel computing flow of ensemble training stage and ensemble application stage is given.Finally,the effectiveness of the parallel algorithm is proved by experiments.The proposed method is applied to predict the mechanical properties of aluminum alloy sheet and strip,The production process of aluminum alloy sheet and strip is analyzed and the key variables affecting the quality index are selected,Preprocess industrial field data,PCA algorithm is used to reduce the dimension of high-dimensional industrial process data,Establish selective ensemble model to predict quality indexes,The basic ensemble algorithm is compared with the improved algorithm proposed in this paper,The results show that the proposed algorithm has better prediction performance.