摘要： 为加快自组织映射的学习速度，提出一个改进的自组织映射(SOM)算法。该算法将类似模拟退火过程应用于SOM学习算法中，动态调整学习参数来优化神经元的运动，并且在损耗值达到一定阈值的情况下提前停止自组织映射聚类，保证输入数据与映射规则的快速学习与较好性能。在提高学习速度的前提下，达到输入到输出的图形一致性。在不同大容量数据集的测试结果表明，该算法与原始SOM算法及其改进算法相比，在收敛速度上可以提高一倍左右，精度上较标准SOM提高50%左右。

关键词: 自组织映射, 模拟退火, 量化误差, 损耗值, 神经元, 增益函数

Abstract: To speed up Self Organizing Mapping(SOM) learning, this paper proposes an improved SOM algorithm, which uses simulated annealing procedure to monitor every epoch of SOM training process, adjusts the parameter dynamic to optimize the movement of neuron, and stops the training when the quantitative error reaches the threshold. An advantage of the proposed algorithm is that it preserves the simplicity of the basic algorithm, obtains fast learning and better performance in terms of matching of input data and regularity of the obtained map. Test compares the proposed algorithm with the original SOM to demonstrate the effectiveness of the new algorithm, the convergent speed can be increased by a factor, and the resolution can be improved by half a factor.

Key words: Self Organizing Mapping(SOM), simulated annealing, quantitative error, loss value, neuron, gain function

中图分类号: 

• TP391