Study And Application Of Deep Learning For Palte Ultra Fast Cooling System

Steel plates are one of the important raw materials for the infrastructure construction in our country.Due to the improvement of market demand and increasingly fierce competition in the steel industry,problems and difficulties of steel plate production will be more outstanding.In terms of controlling cooling,the accuracy and stability of the temperature control model currently used have limitations.The traditional self-learning model also has limitations such as low precision and long learning time.In the steel plate production and control cooling system,there are also some problems,mainly in the aspects of low data utilization level,insufficient optimization of process parameters,and dependence on manual experience excessivly.To solve the above-mentioned problems of controlled cooling after plate rolling,because the limitations of the self-learning model in the traditional post-rolling controlled cooling system,this paper uses the big data information mining ability of deep learning,based on the original post-rolling controlled cooling system.The original system VSG self-learning model complements the deep neural network model,realizes the parallel control architecture of the double self-learning model,and increases the neural network model that can update the weights online,realizing the online synchronization of the process prediction and network training.Explores the migration learning strategy for the non-universality of the DNN model.The main research work of this paper includes:(1)Based on the review of the research progress a related literature of control rolling and control cooling technology domestic and abroad.This paper summarizes the development of temperature control model and self-learning model of cold control system,expounds the development of deep learning and its mathematical theoretical basis in detail,systematically studies the influence of network parameters on its accuracy,and analyzes the common framework of deep learning and how to construct deep neural network.(2)Developed a new type of on-line controlled cooling system for steel plates.In order to meet the requirements of high precision and robustness of the cooling control system in field production,the parallel operation of "VSG+DNN" dual self-learning model was proposed and established.Based on the original plate cold rolling control system,the original system VSG self-learning model is optimized to complement the deep neural network model,and the parallel control architecture of the dual self-learning model is realized.(3)Realized the first industrial application of the new cooling conrtroll with double selflearning model,and participated in the establishment of the post-rolling cooling controll system with "VSG+DNN" double self-learning model.In the production process,the online update function of the deep neural network is realized.The new cold control system developed based on the optimized supplementary deep learning model effectively improves the accuracy and stability in cooling temperature.(4)Exploratory research on the application of transfer learning in the cooling controll system.The data field of the deep neural network model is not universal,and use the transfer learning to train the production datas of 3500mm steel plate line.The scheme can greatly reduce the training time of the model and realize the gradual accumulation of high-quality data knowledge.