Optimizing GBDT's Strip Coiling Temperature Prediction with the Evolutionary Algorithm

Considering the low hit rate of the coiling temperature prediction of the laminar cooling system of the 2 250 mm hot rolling production line in Steelworks B, the gradient boosting decision tree optimized with the diffe-rential evolution algorithm (DE-GBDT) is used to establish the strip coiling temperature prediction model. Five regression prediction models, including Support Vector Machine (DE-SVM) and Wavelet Neural Network (DE-WNN) optimized with the differential evolution algorithm and the three basic prediction models (Gradient Boosting Decision Tree (GBDT), Support Vector Machine (SVM) and Wavelet Neural Network (WNN)), are added to the experiment for comparison. The experimental results show that DE-GBDT prediction model can provide strong su-pport for improving the precision of strip coiling temperature control: (1)compared with DE-SVM and DE-WNN, the DE-GBDT prediction model has the smallest error indicators and the mean square error is 18.232; (2)compared with the three basic prediction models, the error indicators of the DE-GBDT prediction model are significantly smaller than those of the three basic prediction models; compared with the GBDT prediction model, the hit rate of the DE-GBDT prediction model has increased by 2.9% and the mean square error has been reduced by 40.294, indicating that the differential evolution algorithm can significantly improve the model performance.