多脉冲控制的暂态混沌神经网络及其应用

Multi-pulse controlled transiently chaotic neural network and its application to optimization

针对多极值情况下暂态混沌神经网络(TCNN)只进行一次混沌动力施加而导致的部分极值点漏检问题,发展研究了一种名为脉冲暂态混沌神经网络(PICNN)的新型神经网络。PICNN通过对混沌动力控制因子z_i(t)和神经元输出陡度ε_i(l)两个参数的调制,将混沌动力以脉冲方式进行控制,施加于神经网络上,形成具有多频次跳出局部极小点功能的脉冲暂态混沌神经网络。PTCNN包含若干可调参数,可以控制整个系统呈现丰富多样的动力特性。随着混沌动力脉冲的间歇加入,系统交替进入混沌状态和稳定状态,因此既可以不断地跳出极小点的局域范围,又可以在局部区域内向此区域的极小点不断靠近,进而稳定到此极小点,使系统可以更有效地进行问题的全局寻优。算例表明,PTCNN比TCNN在全局寻优方面更具优势。PTCNN实际是TCNN的推广和延伸,比TCNN更具有一般性和更强的优化搜索能力,因此应用空间更为广泛。

Transiently chaotic neural network (TCNN) with only once chaos programming will leave out part of optima when the cost function has many optima. So a new type of NN named pulse transiently chaotic neural network (PTCNN) is researched in this paper. PTCNN can leap from the local optima many times by adapting chaos drive control parameter zi( t) and gradient parameter of neural cell εi( l).The chaos drive is brought to bear on the NN in a pulse manner. Then the system will fall into chaos phase and stabilization phase by turns. In the programming, the system can dap out of the optima, and also constantly approach the optima in its range. The system includes many parameters that make PTCNN present many abundance dynamic characteristics. It is showed by an example that PTCNN can find all optima including the partial optima and the global optima. In conclusion, PTCNN is more advanced than TCNN. In fact, PTCNN extends TCNN, and it is more common and better than TCNN. Therefore, PTCNN can be used in a larger range than TCNN.