滑坡预测的改进前馈网络方法研究

作者提出了滑坡位移预测的一种改进前馈网络方法——目的规划法，与通常的前馈网络方法相比，该方法改进了网络的准则函数，降低了网络的灵敏度，改善了网络的泛化性能，提高了滑坡位移的预测精度。同时它是一种面向数据的方法，适合于不同地区不同条件下滑坡的预测。清江隔河岩库区滑坡和卧龙寺滑坡的实例研究表明了该方法的可行性及有效性。

MODIFIED NEURAL NETWORK FOR LANDSLIDE PREDICTION

Generalization refers to the ability of a neural network to process correctly input data which is not a part of the data used for training the network.One of the important characteristics of the feedforward neural networks is their ability to generalize the input/output behaviors of functions based on a set of training examples.However,many aspects of the problem of improving generalization of feedforward neural networks have not been studied well.In this paper,we address the importance of this problem to landslide forecasting and propose two enhancements to the feedforward neural network,i.e.:(1)modification of the objective function;(2)reduction of the sensitivity of output to a small change in input.Case studies show the feasibility and superiority of the modified neural network model over the normal feedforward neural networks.