Based on Deep Belief Network Intelligent Slag Carry-over Prediction Method

Abstract

According to working principle of continuous casting, this paper proposed an intelligent prediction model based on Deep Belief Network (DBN). The method predicts according to the data detected by the existing continuous casting production, and does not need to change the ladle structure. In order to extract features in the data more efficiently and predict the time series, DBN was introduced. First, the DBN model is constructed to predict the time series. The collected time series is used to train the network model layer by layer to predict the value of the next time variable. Then, the prediction error is calculated by using the DBN network output and the true value, which is defined as a condition detection indicator reflecting whether there is slag carry-over. Due to the poor pouring environment, the collected data has large fluctuations, and the calculated detection indicators are always extremely distributed, which may lead to false positives. Therefore, an adaptive threshold determined by the extreme value theory is proposed and used as a rule for the determination of the slag. This method can realize the early warning of the slag. Finally, the effectiveness of the proposed method is verified by simulation, and the method can judge the slag more accurately and earlier than the shallow neural network.