论文部分内容阅读
通过分析其他BP算法,从中了解这些改进算法的优劣性,提出改进算法。由于测试样本内在关联度的差异,算法的关键是综合考虑局部误差和全局误差2个因素,并且以全局误差值作为步长改变函数的因变量,进而动态地改变学习率,而且这种学习率的改变是有条件的改变。最后,在MATLAB 7.0基础上对XOR问题进行试验分析,综合比较分析证明这个改进算法的有效性。
By analyzing other BP algorithms, we know the advantages and disadvantages of these improved algorithms and propose improved algorithms. Due to the difference of the internal relevance of test samples, the key of the algorithm is to consider two factors of local error and global error synthetically, and take the global error value as the dependent variable to change the function, and then dynamically change the learning rate, and this learning rate The change is a conditional change. Finally, on the basis of MATLAB 7.0, the XOR problem is tested and analyzed, and the comprehensive comparative analysis proves the effectiveness of the improved algorithm.