重置变结构神经网络及其在风险投资项目评估中的应用

基于神经网络的结构直接影响到网络性能的优劣,进而影响其推广使用,尝试将重置算法应用于经典BP神经网络的结构优化,研究了重置算法中最佳重置时间的性质,提出一种重置变结构经典BP神经网络。通过实例,将重置变结构经典BP神经网络应用于风险投资的项目风险评估中,并与经典BP神经网络的实验结果进行对比。结果表明,重置算法的引入有效地解决了神经网络结构的优化问题。

The Revisable Structure of Neural Network Based on Early Restart Algorithm and Its Application in Risk Evaluation of Venture Investment Items

The structure of neural network, as a hotspot of research in ANN, will affect its performance and application directly. This article applies the early-restart algorithm to the optimizatioin of neural network's structure, studies the character of optimal restart time in the algorithm, and advances the revisable structure of classic BP neural network based on early-restart algorithm. We apply it in the risk evaluatioin of the venture capital investment tiems successfully. Comparing with classical BP neural network, we find that the early-restart algorithm can optimize the structure of neural network effectively, and has a significance meaning of being generalized.