| With the rapid development of the industrial era,China has provided more than fifty percent of the world’s total output of steel products for many years in the world,but under the general trend of global low-carbon development,affected by the decision-making requirements of carbon peaking,the current strip export thickness prediction level has reached the bottleneck,so accurate control of the quality of strip rolled products is a powerful measure to ensure that the production quantity of steel products in China ranks in the forefront of the world.Strip export thickness is the main basis for judging the quality of steel rolled products,and inaccurate export thickness prediction not only wastes production materials,but also prolongs the production period and causes serious economic losses.Previously,strip export thickness was mostly controlled by traditional machine learning models,which had many shortcomings.For example,factors that have little impact on the strip outlet thickness cannot be eliminated,and in-depth analysis of nonlinear data cannot be performed,resulting in unsatisfactory prediction accuracy.Trying to work something out,This paper devises and achieves a strip thickness prediction system to optimize the IMPA-BiGRU bidirectional gated recurrent unit neural network model with an improved marine predator algorithm.The value of the neurons number in hidden layer,batch size and learning rate often affects the prediction effect of BiGRU,and choosing right hyperparameter is the essential to raising the prediction accuracy of the model,therefore the parameters of this model are determined by using marine predator algorithm.In order to ensure that the best hyperparameters are found,the original marine predator algorithm is refined,and for balancing the algorithm in the search space Brownian motion and Levy motion strategies,enhance the global search and local search functions,and introduce adaptive weight factor to adjust the search mechanism during prey movement.At the same time,the average fitness value strategy was used to improve the quality of population workaround and enhance the optimization accuracy.The improved marine predator algorithm is used for BiGRU hyperparameter optimization to build a model for prediction.The data in this paper derives from the surveyed strip data in the manufacturing of a steel syndicate in China,and the system contains five mold pieces for design and exploitation,consisting incorporated and login,user administration,data handling,model building and strip thickness prediction module.Among them,in the part of data handling,the primitive data need to extract the features by using the gray correlation analysis technique,retain the feature columns with high influence on the strip thickness to construct a new data set,and then process the data by min-max normalization method.The model building module uses the improved marine predator algorithm to optimize BiGRU hyperparameters to build a predictive model;Finally,the test set data is input into the constructed model to realize the strip thickness prediction.Compared with other traditional model experiments,the IMPA-BiGRU model adopted in this paper has better prediction results,which verifies its superiority.The designed system is easy to operate and has perfect performance in terms of use,and can be used for strip thickness prediction problems in rolling industrial production. |
