基于AP聚类的RBF神经网络研究及其在转炉炼钢中的应用

RBF Neural Network Base on Affinity Propagation Clustering and Its Application in BOF Steelmaking

针对RBF神经网络学习算法不能很好地确定其隐含层节点数、隐含层节点中心及其半径的问题,利用AP聚类算法无需事先确定聚类数的特点,提出了一种基于AP聚类的RBF神经网络算法。将该算法应用于120t转炉Q235B钢种冶炼过程的终点碳含量和温度预报,预测结果与实际结果比较,表明该算法具有预测精度高等优点,可为类似应用提供借鉴。

The number, center points and width of hidden layer units of RBF Neural Network are hard to be deter- mined. AP （Affinity Propagation） Clustering doesntt need to known the number of cluster first. Using this advan- tage of AP Clustering, an RBF neural network algorithm base on affinity propagation clustering was proposed to solve the problems above. The end-point temperature and carbon content of steel grade of Q235B of a 120t con- verter of a steel factory were predicted using the algorithm and the prediction value and actual value was compared. The result shows that this algorithm has a high precision on the condition of higher requirements of the target pa- rameter errors. And it can provide a reference for similar aoolications.