Research On Controlled Cooling Temperature Prediction Model For Hot Rolled Sheet And Strip Based On Neural Network

With the development and improvement of China’s industrial system,industrial industries such as housing and construction,automobile production,shipbuilding,etc.,have an increasing demand for steel with good properties,and the quality,variety and performance requirements of hot-rolled sheet and strip continue to increase.Laminar cooling technology is widely used in hot-rolled strip production lines,and its final cooling temperature directly affects the organizational properties of the steel sheet,but cooling heat exchange is a non-linear process,which is difficult to be accurately described by mathematical models.With the increase of data volume and computer power in industrial production process,the superiority of deep learning models becomes more and more prominent.Among the deep learning algorithms,convolutional neural networks have been successful in many fields,with strong noise immunity,feature extraction capability,complex function expression capability,especially suitable for handling complex nonlinear processes,and stable performance when learning data is sufficient.In order to improve the prediction accuracy of the final cooling temperature of laminar cooling of medium-thick plate,this paper avoids the complex theoretical calculation of heat exchange coefficient and performs regression prediction of the final cooling temperature by neural network.Twenty-two variables,such as thickness of medium-thick plate,running speed,and collector flow rate measured in the actual production process,are used as the inputs of the neural network,and the final cooling temperature is predicted by neural network mapping.Firstly,a BP neural network based laminar cooling final cooling temperature prediction model was established and good prediction results were achieved.For the shortcomings of the fully connected structure of BP neural network,a one-dimensional convolutional neural network prediction model is used for temperature prediction.Based on this,the prediction model was further improved to a one-dimensional multi-scale convolutional neural network prediction model.The results show that the one-dimensional multiscale convolutional neural network prediction model has strong generalization and nonlinear fitting ability,which improves the prediction accuracy of final cooling temperature compared with BP neural network and traditional structure of convolutional neural network,and has theoretical reference and application value in practical industrial production.