基于主成分分析和神经网络的事故预测方法及应用

An Accident Forecasting Method of ANN Based on PCA and Its Application

为解决人工神经网络(ANN)对复杂系统进行事故预测建模时,易导致网络复杂,降低网络性能和增大预测误差的问题,提出一种基于主成分分析法(PCA)的ANN事故预测方法。介绍PCA法和ANN的基本理论,阐述基于PCA法的ANN事故预测模型及其预测步骤,即在利用ANN预测之前,先用PCA法分析事故影响指标,将多个指标转化为少数几个能反映原始信息的互不相关的综合变量(主成分),然后以这些变量作为输入进行ANN建模,从而达到简化模型,提高网络性能和计算精度的目的。以煤矿事故预测为例,进行应用和对比研究。结果表明:基于PCA的ANN事故预测相对误差小于3%,而直接运用ANN方法预测的相对误差达到5%。这说明,对复杂安全系统进行事故预测时,基于PCA法的ANN预测方法是更可行的。

In complex system,there are numerous factors affecting accidents that are complicated and correlated.If all the factors are taken as input parameters of ANN to conduct accident prediction,they will lead to the complicatedness of ANN structure,the reduction of ANN performance and an increase in prediction errors.So a new accident forecasting method based on PCA and ANN was worked out.In this method,PCA is first employed to analyze the numerous safety factors so that they can be transformed into several integrated variables（principal components） that do not influence each other and reflect the information on the original safety factors.Then,ANN was employed to build an accident prediction model with the integrated variables as the input parameters of ANN,so that the neural networks are simplified,and the performance and calculation precision of ANN are improved.Last,the new method was illustrated and tested taking a certain coal mine in China as an example.The results show that the accident relative error predicted by the new method is less than 3%,but the one by ANN is about 5%,which indicates that the new method presented in this paper is more reasonable and feasible.