基于广义模拟退火的人工神经元网络学习方法

Neural-network learning method based on generalized simulated annealing

本文利用广义模拟退火法（简称GSA，下同）能将非线性多极值目标函数较快收敛于全局极值的特点，替换人工神经元网络学习过程中基于梯度下降原理的误差反向传播算法（简称BP，下同）。该方法将由神经元网络的学习输出与期待输出之差的平方和构成的目标函数视为一整体能量系统，模拟热物理学中金属退火处理过程，调整网络中的连接权值，使系统能量尽可能收敛于全局极小。与BP法相比，本方法无需计算梯度，输出响应可采用不可微分的激励函数；另外，无需作误差反向传播计算，因而在神经元网络学习中可使用局部反馈连接的网络结构。该方法的应用为神经元网络学习提供了一种新途径。

Generalized simulated annealing algorithm can make nonlinear multi-extremumobject function converge to global extremum property, thus replacing error backpropagation algorithm based on gradient decline principle in learning of neural network. In the method described here,the object function which was constructed byusing square sum of the differences between learning outputs and expected outputsof neural network is considered as a whole energy system to simu1ate metal annealing processing and to regulate joint weight values in network,so that energy in thesystem converges to global minimum. It is better than BP algorithm. This methodneeds no gradient computation,offers the output response in the form of nondifferentiable excitation function, and per forms no error backpropagation computation;therefore,local feedback-jointed network structure can be introduced in the learningof neural network. This method is a new way for neural network learning.