Accurate prediction of final sulfur contentSulfur content in molten iron pretreatment is crucial for precise control in desulfurization processesProcess. This study proposes a prediction modelPrediction model that optimizes hyperparameters of the extreme learning machine (ELMExtreme Learning Machine (ELM)) using particle swarm optimization (PSOParticle Swarm Optimization (PSO)). The PSOParticle Swarm Optimization (PSO) algorithm globally optimizes ELMExtreme Learning Machine (ELM)'s input weights and hidden layer thresholds, effectively addressing the common issue of local optimization in traditional ELMExtreme Learning Machine (ELM) models, thereby improving sulfur contentSulfur content prediction accuracy. Based on big data from a specific industrial facility, we developed the PSOParticle Swarm Optimization (PSO)-ELMExtreme Learning Machine (ELM) model and conducted comparative experiments with ELMExtreme Learning Machine (ELM), support vector machine (SVM), and other models. Results demonstrate that the PSOParticle Swarm Optimization (PSO)-ELMExtreme Learning Machine (ELM) model achieves a 94.2% hit rate within a ± 0.0003% sulfur contentSulfur content deviation range, showing significant improvement over ELMExtreme Learning Machine (ELM) and SVM. This research provides an effective solution for precise sulfur contentSulfur content control during molten iron pretreatment in desulfurization processesProcess.