Abstract:The determination of the endpoint of converter blowing is crucial in the entire pyrometallurgical process. This process is generally judged by operators based on experience or fixed factors monitored by instruments. However, due to personal operational differences and other factors, the accuracy of blowing cannot be fundamentally guaranteed. This paper designed an intelligent prediction system based on feature significance analysis, and constructed an endpoint prediction model using Resnet network. It integrated three different dimensional features of flame, continuous time PbS/PbO concentration ratio, and SO2 concentration to achieve endpoint prediction during slag and copper making periods. The experimental results showed that the average prediction error during the slag making period was 2.12%, the average prediction error during the copper making period was 2.06%, and the accuracy of endpoint prediction reached 93.66%. The high precision characteristics were mainly attributed to the significance index design based on traditional characteristics, the selected flame and smoke characteristics could accurately predict the end point judgment in a high dimension, and the model trained by introducing a large number of historical data could significantly quantify the prediction process.