摘要
针对过程神经元网络的训练问题,提出了一种基于离散Walsh变换的学习算法。对网络过程式输入及时变权向量,实施离散Walsh变换,用变换后的数据训练网络。在该算法下,可简化过程神经元的聚合运算,避免复杂的积分过程,使过程神经元网络的训练等同于普通网络的训练,即将泛函数逼近问题转化为函数优化计算问题。仿真实验证明了该算法的有效性。
In view of the training problems of process neuron networks, a new learning algorithm based on discrete Walsh conversion is proposed in this paper. This algorithm conducts discrete Walsh conversion to the procedure input of network and the weight vector quantity changing with time, then trains the network by the data having been converted. In this algorithm, the aggregation computation of process neuron can be simplified and the complex integration procedure can be avoided, and the training of process neuron networks can equate to the training of common networks, namely the arbitrary function approximating problems are converted to the function optimization problems. The availability of this algorithm is proved by the simulation test.
出处
《大庆石油学院学报》
CAS
北大核心
2003年第4期58-61,共4页
Journal of Daqing Petroleum Institute
基金
黑龙江省教育厅科学技术研究项目(10511119)
