动态模糊神经网络研究

针对静态网络无法处理暂态问题，对具有递归环节的动态模糊神经网络进行了研究。通过在网络第二层中加入内部反馈连接，使其具有动态映射能力，从而对动态系统有更好的响应。网络使用遗传算法与反向传播BP(BackPropagation)算法相结合来训练，避免陷入局部最优解。采用时序预测和动态非线性系统进行了仿真研究，结果表明，动态模糊神经网络较之普通模糊神经网络在收敛速度、预测精度和网络规模等方面都有较大的改善，并具有更好的动态系统处理能力。

On dynamic fuzzy neural network

Since a static fuzzy neural network cannot deal with the temporal problem, a dynamic fuzzy neural network (DFNN) with recurrent units is proposed. DFNN is constructed by adding feedback connections in the second layer of the fuzzy neural network, so it has the ability of dynamic mapping and can cope with the dynamic systems. The network is trained using the genetic algorithm and BP algorithm to avoid being trapped in the local convergence. The DFNN was applied to the time series prediction and the identification dynamic nonlinear system. The results show that DFNN has a better performance in converging speed and has a small network size, it also has the stronger ability of handling the dynamic systems.