基于粒子群优化径向基神经网络的烧结终点预测研究

Research on Burning Though Point Based on PSO-RBF Neural Network Method

针对烧结生产过程中多变量、强耦合的特点和RBF神经网络结构参数选取依据经验的问题,为提高烧结终点预报模型的精度,提出粒子群算法优化RBF神经网络的烧结终点预测方法。在标准PSO算法的基础上,优化RBF神经网络隐层节点中心和宽度2个结构参数,并建立烧结终点预测模型;在此基础上利用UCI数据库中的Computer Hardware和Concrete Slump Test标准数据,验证了方法的有效性,并以某钢厂265 m2烧结机的实际生产数据,建立烧结终点的预报模型。结果表明,与标准BP,RBF相比,基于PSO优化RBF的烧结终点预测模型精度高、泛化能力强。

Aimed at the problems of multi-variable, strong coupling characteristics in the process of sintering production and RBF neural network structure parameter selection based on experience, the burning through point （BTP） prediction method was proposed based on particle swarm optimizing RBF neural network to improve the accuracy of prediction model. The structural parameters of RBF neural network hidden node centers and width were optimized in the method based on the standard PSO algorithm, and the BTP prediction model was established. On the basis of this, the effectiveness of the method was verified by the standard data of Computer Hardware and Concrete Slump Test in UCI database. Finally, a BTP prediction model was built in a steel plant with the actual production data of a 265 m^2 sintering machine. The results show that, compared with standard BP and RBF, the BTP prediction model has a high accuracy and strong generalization ability.