开关电容神经网络的扩展拉格朗日优化设计

Augmented Lagrangian Optimization Design Method of Switched Capticitor Neural Networks

研究了扩展拉格朗日方法的人工神经网络模型，实现了其电路结构，并进行了仿真实验。基于扩展拉格朗日优化方法的ＳＣ电路具有神经网络的特征和大规模并行处理容量，其描述的函数值在很宽的范围内一般不受参数变化的影响，呈现出显著的坚固性。扩展拉格朗日优化方法具有较好的收敛性，且定位时间较短。

We have investigated artificial neural network models based on an augmented Lagrange method, and accomplished the circuit structure and the simulation experiments. SC circuits based on the augmented Lagrange method have features of neural networks and massively parallel processing capacity,the described function value generally is not affected by parameter variation in a wide range, and exhibits notable robustness. The results illustrate that applying augmented Lagrangian optimization methods have better convergence and the settling time is short.