Copper Alloy Process Model Based On Machine Learning

Accurate prediction of rolling force affects the whole hot rolling process and the performance of hot rolled products when copper alloy is hot rolled.The precision of hot rolling force setting determines the quality accuracy of the products such as thickness,conductivity and tensile strength after hot rolling of copper alloy slab,and can reduce the length of the rolling process by accurately predicting the hot rolling force.There are many factors in the process of copper alloy hot rolling,and the changes among variables can’t be expressed by formulas.If the traditional rolling mechanism model is used to model and predict the rolling force,it is easy to produce large errors due to the few applicable conditions,which can’t meet the needs of intelligent and accurate production for multi-specification products in today’s production.Starting from the improvement of the hot rolling mechanism model,the prediction model of hot rolling force is built by using machine learning method,which improves the accuracy and generalization ability of hot rolling force prediction.In order to improve the performance of the rolling force prediction model,the following work has been done in this paper:First,through the analysis and comparison of existing rolling force prediction algorithms,the application of current mechanism model and machine learning algorithm in actual production data is deeply analyzed.This paper puts forward the construction of rolling force prediction model using support vector regression,and discusses the basic theory of support vector machine and support vector regression.Secondly,after analyzing and discussing the parameters of machine learning algorithm that affect the performance of support vector regression,this paper presents a particle swarm optimization algorithm to optimize the parameters of support vector regression algorithm,which enhances the global optimization ability of support vector regression algorithm.Neighborhood partitioning is then added to the particle swarm algorithm to change the global optimal value to the optimal value in the neighborhood,which increases the convergence and local search ability of the network.The experimental results show that the model is superior to the genetic algorithm optimization model in prediction precision and model generalization ability and can meet the requirements in the actual production environment.Thirdly,by analyzing the production data produced during the actual copper alloy hot rolling process,this paper puts forward that the rolling force training sample set for the copper alloy hot rolling process is established by adding two features,rolling time and rolling pass,into the existing rolling training sample set.Thus,the hot rolling force of the slab in the whole process of hot rolling is predicted.Fourth,because the improved training sample set has more features and more data samples,it is difficult to fit the training data well only by using the support vector regression model.This paper presents a rolling force prediction model based on deep belief network,which combines deep belief network with support vector regression.The deep belief network is used to analyze the interaction between the features of training sample set,to provide deep features after dimension reduction for support vector regression,and to increase the efficiency and accuracy of support vector regression.The experimental results show that compared with other neural network algorithms,the prediction model built with deep belief network support vector regression reduces the overall operation time by 40%,and the prediction accuracy is also greatly improved.