一种基于Legendre正交基变换的过程神经元网络训练

为解决过程神经元网络训练涉及的时域聚合运算问题,提出了过程神经元网络的一种学习算法。算法在网络的输入函数空间引入Legendre正交函数基,将输入函数和网络连接权函数表示为该组正交基的有限项展开形式,利用Legendre函数基的正交性,避免复杂的积分过程,降低过程神经元在时间聚合运算中的复杂性,提高学习效率。仿真实验结果证明了算法的有效性。

Training of Process Neural Networks Based on Legendre Orthogonal basis Conversion

Aiming at process neural network aggregation operation problem in time field, a training method of PNN is shown. Legendre function orthogonal base is introduced into input function space of network, and both input function and weight function are denoted limited items of function orthogonal base. Using the orthogonality of function base, the complexity of PNN in time aggregation computing is simplified, the learning efficiency of the network is improved. The algorithm of general realization is presented, and the simulation experience proved availability of the algorithm.