预备知识：一Sigmoid:)exp(11xy)1())exp(1(1)exp(11))exp(1(1)exp(1))exp(())exp(1(1)exp())exp(1(1))(ex1(22221-yyxxxxxxdxxdxdxxpddxdy二BP推导1.符号定义：网络结构：输入层有I个神经元，隐层有J个神经元，输出层有K个神经元，权值为W。输入样本TIxxxx).....,(21，隐层输出TJhhhH).....,(21实际输出TKYyyY).....,(21教学值TKtttT).....,(21，net为神经元的输出。隐层：IllljjjjJjxWnetnetgh.....2,1;*),(输出层：JlllkkkkKkyWnetnetgy.....2,1;*),(误差和函数：KlllytE2)(21权值调整：wEmwmw)()1(2..输出层权值调整：。。。IllljjjjJjxWnetnetgh.....2,1;*),(JlllkkkkKkyWnetnetgy.....2,1;*),(jkW表示第j个神经元到第k个神经元之间的权值nethnethnethnety误差求导：jkkkjkkkkkjkkkjkWnetWnetnetyyEWnetnetEWE)1()(-kkkkkyyytjjkJlllkjkJlllkjkjkJlllkjkkhWhWjlWhWWjlWhWWnet时当的函数侧时不是当0!jkkkkjkkkjkhyyytWnetWE)1()(-3.隐层权值调整：。。。IllljjjjJjxWnetnetgh.....2,1;*),(ijW表示第i个神经元与第j个神经元之间的权值误差求导：ijjhijjjjjijjjijWnetWnetnethhEWnetnetEWExxxnethjjkljlljjjlKlljjjnethWnethnetnetnetEnethhEh)(jjnetgh)1(jjjjhhneth)1(hjjkljllhhWiijjxWnetxihhWyyytxihhWWinetWEjjkljllllljjkljlljjhij)1()1()()1(xihhWYYYTjjkljl)1()1()(
