Soft sensor method for endpoint carbon content and temperature of BOF based on multi-cluster dynamic adaptive selection ensemble learning

Abstract The accurate control of the endpoint in converter steelmaking is great significance and value for energy saving, emission reduction, steel quality improvement. key to lies accurately predicting carbon content temperature. Converter a dynamic process with large fluctuation samples, traditional ensemble learning methods ignore differences among query samples use all sub-models predict. different performances each sub-model lead performance degradation learning. To address this issue, we propose soft sensor method based on multi-cluster adaptive selection (MC-DAS) temperature prediction. First, ensure diversity base model, clustering algorithm data partition characteristics construct pool diverse models. Second, model strategy proposed, which involves constructing similarity regions individual assessing model’s these identify most suitable weight combination respective sample. Compared method, simulation results actual show that prediction accuracy within ±0.02% error range reaches 92.8%, ±10°C 91.6%.