基于自适应动量因子的区间神经网络建模方法

Interval Neural Network Modeling Method Based on Adaptive Momentum Factor

摘要:

区间神经网络建模是区间控制的核心部分，也是提高系统鲁棒性的重要方法。针对区间神经网络算法收敛速度慢的问题，提出一种自适应动量因子算法。算法利用区间运算建立输入与输出数据的映射模型，通过引入具有自适应特性的动量项，使用最速下降法对动量项进行自适应更新，在加快系统收敛速度的同时，克服系统稳态误差大和容易陷入局部最小值的弊端。典型算例实验表明:区间神经网络能够较为精确地建立区间网络模型，自适应动量因子算法提高了区间神经网络整体性能。

Abstract:

The modeling of interval neural network is not only a component of interval control, but also an important role to improve the robust of systems. An adaptive algorithm of momentum factor is proposed to solve the problem of slow convergence speed on the interval neural network. In this paper, interval calculation method is used to establish the mapping model of input and output variables. By introducing a momentum term with adaptive characteristics, the steepest descent algorithm is applied to update the adaptive momentum factor. Compared with the traditional method, this method not only accelerates the convergence speed, but also overcome the disadvantages of the system steady state error and easily to fall into local minimum. According to the nonlinear experiments, interval neural networks are able to establish the zone models, and the algorithm of adaptive momentum factor increase the overall performance of the network. Classic bench mark experiments show that our work can more accurate to establish interval network model, while introducing of adaptive momentum factor algorithm also can improve the overall performance of the interval neural network