Review on the accident-forecasting research methods

The present review is conducted in hoping to assist the researchers in the field to understand the general principles and methods of accident-forecasting and prevention in a systematic and comprehensive manner as well as provide hints to identify the research status and trends. As is known,accident-forecasting is increasingly becoming the hot topic in the whole society,which is likely to be of great significance for system safety and accident prevention in advance. Keeping these goals in mind,this paper intends to generalize the main commonly recognized methods into six major categories,including scenario analysis,regression method,time series method,Markov chain method,grey model as well as the nonlinear model. At the same time,it has also made a comparative analysis and generalization on them,as well as a study on the model selection and accuracy measure. Thus,it can be concluded that: 1) Most forecasting theories and methods are designated for certain particular application trades or fields. As the model complexity,accuracy and data requirement may differ from each other,a generally adopted method should be suitable and accurate to the specific modes and characteristic features of a certain accident; 2) In case a randomly chosen city and contingency were eliminated in order to gain more useful guide of modeling,stress should be put on analyzing the outlier in the given sample space and the heterogeneity of factors; 3) The system safety proves to be non-linear,dynamic and stochastic in nature,and,accordingly,it is better for realistic application to combine the quantitative and qualitative analysis,the methods both linear and nonlinear in nature,and static and dynamic methods in to an integrated entity; 4) The idea of prevention should be made the focus of the accident-forecasting research based on the state forecasting tasks in the future.