Predicting Mechanical Properties of Steel Based on Ensemble Learning and Bayesian Optimization

With the accelerated development of digital transformation and upgrading of the steel industry, artificial intelligence technology has been introduced in steel mills to predict the mechanical properties of steel products, thereby reducing the frequency of product quality inspection and providing guidance for product quality optimization. Unfortunately, the accuracy of the existing steel mechanical properties prediction model is not very high, and the prediction model only predicts the mechanical properties of part of the steel grades, and does not cover all the steel grades of the entire production line. Aiming at the above problems in the existing steel mechanical properties prediction models, a prediction model based on ensemble learning and Bayesian optimization is proposed to improve the prediction accuracy and cover all steel grades in the entire production line. Taking steel yield strength prediction as an example, random forest (RF) and CatBoost (CATB) as base learners and linear regression (LR) as meta learners are used to establish the yield strength prediction model based on the stacking strategy. The comprehensive experimental results indicate that our proposed ensemble learning prediction model outperforms mainstream single machine learning prediction models. The accuracy of this model reached 92.62% (±30MPa absolute error) on the independent test sets, which significantly improved the accuracy of a single machine learning prediction model. In general, the ensemble learning framework has better prediction accuracy, which has important popularization significance and application value.