Data-Driven Based Condition Classification And Furnace Temperature Real-Time Prediction For Aluminum Smelting Process

Aluminum is one of the most widely used metals today,and as the most important part of aluminum smelting,the operation of the aluminum smelting process is critical to the yield,quality and energy consumption of aluminum.In the actual production process of aluminum melting,the aluminum liquid temperature is usually used to guide the production,but due to the high temperature and corrosive nature of the aluminum liquid,the thermocouple sensor needs to be protected with a protective jacket,which makes it difficult to obtain the aluminum liquid temperature data and increases the production cost.It is found that the smelting furnace temperature can replace the liquid aluminum temperature to guide the aluminum smelting production,and the furnace temperature data is relatively easy to obtain,which is more conducive to modeling.However,the variable composition and nature of the raw materials added in the aluminum smelting process lead to frequent adjustments of the smelting furnace temperature,which makes it difficult to predict the furnace temperature.To address the above problems,a data-driven working condition classification model and a real-time furnace temperature prediction model for aluminum smelting process are studied and proposed in this paper.In this paper,with the production process of regenerative aluminum smelting furnace in a largescale aluminum plant as the background,we have studied the trend of furnace temperature change in the production process of aluminum smelting furnace and made some results on the prediction of furnace temperature.The main contents and innovative results of the paper are as follows:(1)Firstly,the whole process of production of recycled aluminum is introduced,focusing on the analysis of the recycled aluminum smelting process.Subsequently,after a mechanistic analysis of the process of aluminum smelting process,the furnace temperature and related process parameters that affect the smelting state were analyzed,and the relevant variables that have an important effect on the furnace temperature were obtained.(2)In view of the variable composition and nature of the raw materials added in the aluminum melting process and the frequent changes of the furnace temperature,a dynamic time-regularized fuzzy clustering method based on the optimization of the sparrow search algorithm to cluster the furnace temperature data series after the sliding window is proposed.This clustering method can accurately cluster the furnace temperature sequence according to the change trend of furnace temperature.At the same time,the one-dimensional convolutional neural network is used to predict the category of working conditions in the smelting process at the next moment according to the relevant variables affecting the furnace temperature change at the current moment,realizing the clustering and classification of working conditions in the aluminum smelting process.Finally,the effectiveness of the working condition classification and clustering model is verified with the actual production data of an aluminum plant.(3)To address the problem of time-varying in aluminum smelting process and the difficulty of smelting furnace temperature prediction,a method to establish a soft sensor of the furnace temperature by combining the above work condition classification model and a long and short-term memory neural network model based on time-series correlation weighting is proposed,which achieves the prediction of the furnace temperature.Firstly,a just-in-time learning strategy based on time-series correlation weighting.This just-in-time learning strategy uses the inter-sequence distance under the dynamic time warping algorithm as a similarity index,and selects modeling samples based on that similarity between query sample and the historical samples when the query samples are available.Then,based on the above just-in-time learning strategy and the long short-term memory neural network model,a soft sensor of long short-term memory neural network based on time-series correlation weighting to achieve the prediction of furnace temperature this paper established,which improves the ability of long short-term memory neural network to handle nonlinearity and time variability.Thirdly,in order to reduce the time consumed for modeling and prediction of the above soft sensor,the aforementioned working condition classification model is applied to the soft sensor.Finally,the validity of the proposed soft sensor is verified by the actual production data of an aluminum smelting plant.