基于DBN-LSSVM的热连轧带钢厚度预测

为提高热连轧带钢精轧厚度预测精度,建立了通过深度置信网络(deep belief network,简称DBN)提取特征的最小二乘支持向量机回归模型(DBN-LSSVM),并且利用粒子群优化算法(particle swarm optimization,简称PSO)优化最小二乘支持向量机(least squares support vector machine,简称LSSVM)相关超参数。通过采集某热连轧带钢生产线实时现场数据,对所提出厚度预测模型进行训练和离线仿真。结果表明,基于DBN-LSSVM的预测模型具有良好的学习能力和泛化性,DBN-LSSVM模型的预测精度较传统BP算法和DBN-BP算法有显著提高,该厚度预测模型在生产实践中具有很好的应用前景。

Hot strip thickness prediction based on least squares support vector machine with feature extraction by deep belief network

In order to improve the accuracy of thickness prediction for hot strip mill,a least squares support vector machine( LSSVM) model was established with extracting features through deep belief network( DBN). The particle swarm optimization( PSO) algorithm was used to optimize the LSSVMrelated hyperparameter. By collecting real-time field data of a hot strip production line,the proposed thickness prediction model was trained and simulated offline. The simulation results showed that the prediction model based on DBN-LSSVM had good learning ability and generalization. The prediction accuracy of DBN-LSSVM model was significantly higher than that of traditional BP algorithm and DBN-BP algorithm. The thickness prediction model had a good application prospect in production practice.