Research Of Accident Forecasting Based On Artificial Neural Network

Safety in Production is a huge project. The research work of safety science is one of the important tache in this huge project. Various types of particularly serious accidents have frequently occurred in recent years which have had a major negative impact on China's national economic development. It also brings serious influence on safety in production. Therefore, scientifically studied each kind of casualties change tendency of our country enterprise to production will be able to provide the reference for our country safety in production to constituting reasonable solving measure.Because artificial neural network technology has highly nonlinear mapping capability, self-organizing, self-adaptive and self-learning ability, particularly suited to address the complex causal relationship between non-deterministic reasoning, judgment prediction and classification problems. Time series forecast, then is from analyzing time sequence change characteristic information and so on, choosing the suitable model and the parameter, establishing forecast model, and according to the inertia principle, assuming the forecast object former change tendency will be able to continue the future, thus make the forecast. So this article unified the artificial neutral network and the time series forecast carried on the establishment, the training and the simulation forecast model research based on theory BP network time series and so on.This article simply introduced the basic principle of artificial neural network, the basic structure of BP network, the learning algorithm, the network model and the neural network toolbox of MATLAB calculate software. We also analyzed and compared with several kinds of forecasts models which at present uses in forecast. The time series forecast model based on BP network is the key introduction among them.This article established the accidents death rates of all the industrial enterprises form 1974 to 2003, also trained and simulated time series model which is based on BP network. The simulation result of neural network was more accurate than the result of ARMA through compared with each other. The results showed that the forecasting model which based on BP network had the ideal forecast ability. The next,through the research in the casualty of the kinds of coal mine accidents from 1981 to 2002, it proofed the established forecast model which based on BP network had the validity in the in putting and multi- factors.