训练前向神经网络的全局优化新算法及其应用

A New Global Optimization Algorithm for Training FeedforwardNeural Networks and Its Application

把填充函数法与 BP算法相结合 ,提出一种训练前向神经网络的混合型全局优化新算法 .该算法首先由 BP算法得到一个局部极小点 ,然后利用填充函数使 BP算法跳出局部最优 ,得到一个更低的极小点 .重复此过程最终求得全局最优解 .最后给出一个应用实例 .

This paper proposes a new global optimization technique in which combines the filled function method and BP algorithm for Training feedforward neural networks. In this algorithm, the BP algorithm finds one of local minimal points first, the filled function method finds the point that is lower than the minimal point previously found. By repeating these processes, a global minimal point can be obtained at last. Practical examples indicate that the method works well in avoiding sticking in local minima. Compared with usual BP training algorithm, this new global optimization algorithm is more efficient and has a higher accuracy in application to establishing production quality model.