Research On Carbon Temperature And Soft Measurement Method At Blowing End Point Based On Data Integration Learning Of Converter Steelmaking Production Process

Basic oxygen furnace steelmaking technology is currently the most widely used steelmaking technology in the world,and its key is to achieve accurate control of the basic oxygen furnace steelmaking endpoint.However,the industrial site environment of basic oxygen furnace steelmaking is harsh and accompanied by complex physical and chemical reactions in the production process.In order to achieve timely and effective endpoint control and meet the requirements of steel tapping,it is extremely important to accurately predict the carbon content and temperature of molten steel in the molten pool.Among them,the basic oxygen furnace steelmaking production process data is closely related to the endpoint carbon content and temperature.Therefore,the method of generating a soft sensing model of endpoint carbon content and temperature based on the basic oxygen furnace steelmaking production process data came into being.However,in the process of basic oxygen furnace steelmaking,the changes in the carbon content and temperature of molten steel in the molten pool are complex,and there are many influencing factors.A single global model often does not perform well when targeting these data.In actual production,due to the instability of raw materials and differences in quality,the volatility of furnace samples is large,and the interpretation of issues such as high data dimensions is not strong enough,so it is based on soft sensor modeling method of ensemble learning provides a good idea for solving the above problems by integrating multiple single partial models to construct an integrated learner.This paper uses the idea of ensemble learning to conduct an in-depth study on the problem of constructing a prediction model for the carbon content and temperature of the basic oxygen furnace steelmaking endpoint.(1)Soft measurement method of endpoint carbon content and temperature of basic oxygen furnace steelmaking based on LR-SVGD ensemble learning.Aiming at the problem that the global single model will affect the integration effect and cannot accurately predict the endpoint carbon content and temperature caused by the large fluctuation of the heat sample caused by the difference in the quality of raw materials in actual production,an ensemble learning soft measurement method based on LR-SVGD model selection is proposed.First,peak density clustering algorithm is used to divide the visualized training data to form a partial sample subset,build a one-to-one correspondence between the subset and the original data to generate a generalized regression neural network sub-model.The probability of correlation between the test sample and different models is obtained through LR-SVGD.Secondly,the general regression neural network sub-model that is better associated with the test sample is selected as the local model through the correlation probability,and the LR-SVGD model selection method is proposed to obtain the ensemble output carbon content and temperature prediction results.According to simulation results of actual basic oxygen furnace steelmaking production process data,it is shown that prediction accuracy of carbon content within ±0.02 % error range 81%,temperature within ±10? error range reach 73.8%.(2)Soft measurement method of endpoint carbon content and temperature of basic oxygen furnace steelmaking based on LNN-DPC weighted ensemble learning.This chapter proposes a gray correlation weighted integrated soft sensor modeling method based on density clustering.Firstly,improved peak density clustering algorithm is applied to classify training data after dimensionality reduction to form local sample subset,than one-to-one correspondence between subset and original data is constructed to generate gaussian process regression sub-model,and the entropy-weighted subset“centroid” is obtained by measuring under original data subset;secondly,model with strong correlation degree of test samples is selected as local model by gray correlation analysis,and weighted ensemble strategy of correlation degree is proposed to output carbon content and temperature prediction results.According to simulation results of actual basic oxygen furnace steelmaking production process data,it is shown that prediction accuracy of carbon content within ±0.02 % error range 85.2%,temperature within ±10? error range reach 84.8%.(3)Soft measurement method of carbon content and temperature at the endpoint of basic oxygen furnace steelmaking based on autoencoder feature extraction.Aiming at the problem that the high dimensionality of the production process data in the actual production process is not conducive to the accurate prediction of the final carbon temperature of the basic oxygen furnace steelmaking,an ensemble learning basic oxygen furnace steelmaking endpoint carbon content and temperature soft measurement method based on autoencoder is adopted.First,the improved peak density clustering algorithm is used to divide the visualized training data to form a local sample subset,and form a high-dimensional data subset through the mapping relationship between the constructed local subset and the original data;secondly,autoencoder is used to perform feature reduction on each high-dimensional data subsets to form dimensionality-reduced data subsets to train different models,and select the best training sub-models under different models through self-learning;finally,the best training sub-model that has a strong correlation with the test sample is selected through gray correlation analysis,and the output carbon content and temperature prediction result of the gray correlation weighted integrated model is obtained.According to simulation results of actual basic oxygen furnace steelmaking production process data,it is shown that prediction accuracy of carbon content within ±0.02 % error range 90%,temperature within ±10? error range reach 87%.