迟滞混沌神经网络及应用

Neural Network with Hysteretic and Chaotic Characteristics and Its Applications

提出了一种同时具有迟滞和混沌特性的神经元模型,并利用该模型构造出神经网络,用于求解优化计算等问题。通过在神经元中引入自反馈,使得神经元具有混沌特性。将神经元的激励函数改为具有上升分支和下降分支的迟滞函数,从而将迟滞特性引入神经元和神经网络中。结合模拟退火机制,在优化计算初期,利用混沌特性可提高网络的遍历寻优能力,利用迟滞特性可在一定程度上克服假饱和现象,提高网络的寻优速度。在优化计算末期,网络蜕变为普通的Hopfiled型神经网络,按照梯度寻优方式收敛到某局部最优解。可通过构造能量函数的方法,将图像识别中的特征点匹配等问题转化为优化计算问题,从而可采用该神经网络进行问题求解。仿真结果验证了该方法的有效性。

A neuron model with chaotic and hysteretic characteristic is proposed.Neural network coupled by such neurons is applied to the optimization problem.The self-feedback is introduced into neurons,which makes the neurons have the chaotic characteristic.The activation function has ascending segment and descending segment,which makes the neurons have the hysteresis characteristic.In the initial process of optimization,the ability of ergodicity optimization is enhanced by the chaotic characteristic,and the searching speed is rasied by the hysteresis characteristic.In the terminal process,using the simulated annealing method,the network decay the conventional Hopfield neural network,and convergent to some local optimal solution according to gradient optimization.An appropriate energy function is constructed,and the feature points matching problem can be transformed into optimization problems which can be solved by the neural network.Simulation results show the proposed method effectiveness.