Research On BP Neural Network Breakout Prediction System Based On Ant Colony Algorithm Optimization

With the continuous development of modern technology,the requirements for continuous casting technology are getting higher and higher,so the development of high efficiency continuous casting technology is very fast.The thin slab continuous casting technology in the high efficiency continuous casting technology is working faster and the friction between the billet and the crystallizer is increasing,which will cause the problem of poor lubrication and thus the breakout accident,which will not only cause the damage to the related equipment,but also cause the financial loss,thus reducing the productivity and efficiency of the enterprise.In order to understand the internal state of the crystallizer in a timely manner,this paper develops a visual breakout prediction system for accurate prediction of breakout accidents,which focuses on the following four aspects.Firstly,the paper introduces the types and detection methods of sticker breakout,and takes sticker breakout as an example,discusses and analyzes its formation process,and gives corresponding preventive measures for its common triggering factors;then selects the thermocouple temperature measurement method to detect the breakout,and gives the prediction principle of sticker breakout based on its formation process.Secondly,K-type thermocouple is selected as the temperature data acquisition device,and its burial method and arrangement order on the copper plate are given.The collected temperature signal is then converted into temperature data,and operations such as screening and analysis are carried out,and the temperature data is then normalized after its change pattern is analyzed.Then,the ACO algorithm is used to find the optimal value of the weight threshold of the BP neural network and establish the single-even time series network model to identify the temperature time series change of a single thermocouple and judge whether it conforms to the sticker characteristics;then the group-couple spatial network model is established according to the propagation characteristics at the sticker crack,and when the single-even time series network model is judged to be the sticker state,it is further judged in the spatial structure whether it conforms to the sticker characteristics,and the model is tested offline using the pre-processed historical data to compare the performance of the model before and after optimization.Finally,the ANSYS ADPL software was used to build a model of the crystallizer copper plate,solve its steady-state temperature field,and develop a visual breakout prediction system based on the Windows platform,using the secondary development function of ANSYS ADPL connected with Python,and import the temperature data when the breakout prediction model is judged to be the state of breakout into the interface.The test results show that the visualization interface can accurately and quickly display the temperature cloud map of the current state of the crystallizer copper plate hot surface,which can be more directly used to judge the state of breakout through the cloud map.