Prediction of Matte Grade in Copper Flash Smelting Process Based on LSTM and Mechanism Model

Copper flash smelting is the main process method for copper extracting. It is a complex and multi-phase physical and chemical change process with the characteristics of multiple variable, nonlinearity, delayed detection of product quality. Matte grade is a key indicator reflecting the stability of flash furnace smelting conditions, which directly affect the product quality and process level of the flash smelting. With the ‘four highs’ (high oxygen-enriched blast, high matte grade, high feed rate, high thermal strength) technical requirements, previous forecasting models are no longer applicable to the new conditions, so it is necessary to establish a forecasting model that can meet the requirements of the latest process. This paper uses matte grade as an example to estimate the calculation results through the compound and establish a mechanism model, analyze its feasibility. Based on the mechanism model, analyze the characteristics of long short-term memory network (LSTM) and choose to establish MSLSTM model (Semi-supervised LSTM based on mechanism model), and compare those methods, which shows that MSLSTM dynamically analyzes and predicts the non-linear relationship between the multi-parameter time series. The measured data from a smelter in Jiangxi, China are used for verification. The analysis results show that the predictive model proposed in this paper has high accuracy. MSLSTM model can be used for production operation guidance and optimal control of copper flash smelting process.