一种基于聚类的交叉覆盖算法

A Kind of Alternative Covering Algorithm Based on Clustering

与传统的前向神经网络相比,覆盖算法具有运行速度快、精度高的特点,但覆盖算法的初始领域中心是随机选取的。实验表明网络性能与学习顺序有密切的关系。在前向神经网络交叉覆盖算法基础上提出了一种新型改进的交叉覆盖算法——基于聚类的交叉覆盖算法。该方法是一种根据聚类结果确定学习顺序的方法。实例表明这种改进的算法是确定性学习方法,可以有效减少覆盖数量,提高交叉覆盖算法的测试速度,减少拒识样本数,提高识别的精度。

Compared with traditional algorithm of forward neural network （FNN）,covering algorithm （CA） possesses some advantages, such as faster speed and higher precision. But the original centers of sphere domains are selected at random. Experiments show that the performance of network is related with the order of study closely. A new kind of algorithm named CABC, which combines covering algo- rithm and clustering is put forward. Instances show that this kind of algorithm is deterministic learning algorithm. It can reduce the number of sphere domains avail,ably, low down testing time, reduce the number of rejected samples,and improve the recognition precision.