基于神经网络再建模的模糊PID控制器精简化研究

针对模糊控制算法的计算复杂性和实时性能差的问题,以模糊PID控制器为研究对象,利用神经网络的万能函数逼近能力,通过神经网络二次建模,精确的逼近已知的模糊PID控制器,从而减少运算量,实现实时控制.然后,给定不同的输入信号,分别用模糊控制器和等效神经网络模型控制同一个被控对象.结果表明,控制效果非常相似.因此,用精简的神经网络模型来代替模糊控制器,可减少计算的复杂性,避免维度灾难,提高实时性能.

Simplifying Fuzzy PID Controller Based on Remodeling with NN

Aiming at computational complexity and poor real-time performance in fuzzy control algorithm,selecting fuzzy PID controller as study plant,an equivalent NN model with universal function approximating ability was utilized to accurately approach a known fuzzy PID controller. After that,the same plant model which was controlled by the fuzzy PID controller and the equivalent NN model were simulated with different reference inputs,respectively. Results demonstrate that control qualities from two different controllers are extremely similar. Therefore,the fuzzy PID controller can be replaced by an equivalent NN model in order to reduce the computational complexity,avoid the dimensional curse and improve the real-time performance.