神经网络空间映射结构的研究与改进

Research and improvement of neuro-space mapping structure

针对粗模型与器件的直流特性差异较大而交流特性相似时建模过程复杂的问题,对已有神经网络空间映射(Neuro-SM)结构进行了改进。改进的模型在Neuro-SM结构基础上,增加电容和电感,使映射网络仅调整输入信号中的直流分量,不影响交流分量。在不改变粗模型交流特性的情况下改进直流特性,用少量的优化变量和简单的映射关系即可达到模型匹配的效果。通过仿真实验表明,改进后的Neuro-SM模型充分利用粗模型与器件非线性响应相似的特点,既保持了模型的精度又简化了建模过程。

In some cases, the difference of DC responses between the coarse model and devices is large, however the nonlinear responses are similar. Concerning the complex modeling process, an improved Neuro-Space Mapping （Neuro-SM） structure was proposed. The capacitors and induetors were added on the traditional Neuro-SM model to constitute a new Neuro- SM model. The DC component of the input signal was adjusted by the mapping network, but the AC component is independent on the mapping network. The new model can improve the DC feature without changing AC characteristic and match the device with a few optimization variables and simple mapping relationship. The simulation experimental results demonstrate that the enhanced Neuro-SM model can make full use of the similar nonlinear responses between the coarse model and devices, maintaining the accuracy of the model as well as simplifying the modeling process.