模糊神经网络的优化及其应用

Optimization and Application of Fuzzy Neural Network

为了提高模糊神经网络的泛化能力和收敛速度,以及减小误差,提出了采用动量梯度下降算法和RMSprop优化算法模糊神经网络的方法,研究并设计神经网络的参数调整的自适应过程,在代价函数中加入正则项,实现参数的更新,同时保证模糊神经网络的收敛速度,对预测输出和实际输出的拟合效果和误差进行比较。将二种优化模糊神经网络的算法应用于非线性函数逼近、Mackey-Glass混沌时间序列和水质等级评价的输出预测中,实验结果表明,RMSprop的预测输出和实际输出的误差和拟合效果优于动量梯度下降算法。

In order to improve the generalization ability,convergence speed and reduce the error of the fuzzy neural network,the method of using momentum gradient descent algorithm and RMSprop optimization algorithm to optimize the fuzzy neural network is proposed.The adaptive process of parameter adjustment of the neural network is studied and designed.Regular terms are added to the cost function to update the parameters,and the convergence speed of the fuzzy neural network is guaranteed.The fitting effect and error between the predicted output and the actual output are compared.The two optimization algorithms of fuzzy neural network are applied to output prediction of non-linear function approximation,Mackey-Glass chaotic time series and water quality grade evaluation.The experimental results show that the error and fitting effect of RMSprop′s predicted output and actual output are better than that of momentum gradient descent algorithm.