摘要
神经网络的学习速度是影响其在实时控制中应用的重要原因之一，在文中提出了一种基于局部调整方法的模糊神经网络快速学习算法．该算法通过采用对输入数据进行判别的方法来选择每次学习时所需调整的有效规则，大大减少了学习中调整的规则数，从而加快了模糊神经网络的学习速度．同时，通过这一判别还可进一步确定是否需增加新规则以及增加的规则数，因此该算法不仅能够进行模糊神经网络的参数调整，还能实现神经网络的结构自适应调整功能．随着神经网络的输入维数以及初始规则数目的增加，算法的上述优点更加明显．最后采用快速算法与普通算法分别对单输入及多输入系统进行了辨识，仿真结果证明了上述结论：在初始规则数较少，普通算法无法收敛时，应用快速算法则可以收敛；随着规则数目与输入维数的增加。
This paper presents a novel fast learning algorithm based on local adjustment for fuzzyneural network. Only the effective rules judged by the input data are adjusted, so the number of the adjusted rules decreases greatly and the learning time is saved. Furthermore, it can be judged whether new rules should be added and what is the number of the added rules. So not only the parameters of the fuzzyneural network can be optimized, but also the structure of the neural network can be adjusted adaptively. The advantage becomes more obvious as the input dimension and initial rule number increase. Finally, a comparison is made between the fast algorithm and the common algorithm. Two different cases are simulated: singleinput and multiinput system have been identified by the presented algorithm and the common algorithm respectively. The results show that when the initial rule number is small, the identification is convergent by the presented algorithm while it is unstable by the common learning algorithm. In other cases, the presented algorithm is better than the common one as the input dimension and initial rule number increase.
出处
《上海交通大学学报》
EI
CAS
CSCD
北大核心
1998年第8期58-62,共5页
Journal of Shanghai Jiaotong University
关键词
模糊神经网络
自学习
规则调整
BP算法
fuzzyneural network
selflearning
rule adjustment
BP algorithm
