Endpoint carbon content and temperature prediction model in BOF steelmaking based on posterior probability and intra-cluster feature weight online dynamic feature selection

Abstract

A posterior probability and intra-cluster feature weight online dynamic feature selection algorithm is proposed to address the issues of high dimensionality and high volatility of data in the basic oxygen furnace (BOF) steelmaking production process. First, a genetic algorithm with fixed feature space dimensions is introduced, which narrows the solution space by predefining the number of selected features, thereby enhancing the stability of feature selection. Second, the posterior probability of samples and intra-cluster feature weights are used to weigh and calculate the feature importance of the current sample, obtaining the optimal features that align with the current operating conditions. Finally, the dynamically selected features are used in a regression model to predict the carbon content and temperature of the BOF steelmaking process data. Simulations of actual BOF steelmaking process data showed that the prediction accuracy was 86% within a carbon content error range of 0.02, and 88% within a temperature error range of 10°C.

Keywords

• bof steelmaking
• soft sensor
• online dynamic feature selection algorithm
• genetic algorithm