Application Of Artificial Neural Network Technique On Safety Assessment

With the rapid development of scientific technique and modern industry, the safety forecast work becomes an important link of both modern safety production and safety management. The selection of assessment method is the key of safety assessment work. It directly influences the deep and accuracy of assessment. The requirement for assessment medium and method is getting higher and higher. It is increasing the quality of safety forecast nowadays.Artificial neural network (ANN) technique has a very strong nonlinear ability and the self-adaptability. Especially, it is suitable for resolving the problems such as complex uncertain inference of the causalities, determination, forecast, classification and so on. The object of safety forecast is a complex system which contains many factors of both certainty and uncertainty. The ANN technique can just be applied in the safe-forecast. The implicative experience and knowledge of mankind, and the intuition thought of the viewpoint of the importance of each factor etc. can be achieved through learning of the existing safety system and its evaluated result. Once forecast by means of the technique, these experiences, knowledge and thought, and rational determination for the complex problem can be reappeared through the network.It was tried that neural network is drawn into the work of the quantitative safety forecast method. The particularly applied research of the casualties accident forecast and the dangerous degree classification of MOND Method were done.After introducing the basic theory of the ANN, the three-layer BP neural network was established by using neural network toolbox in the MATLAB software. Using the industrial casualty statistical datum from 1985 to 2002 of England, as training sample, the casualties were predicted for 2003 by adopting different number of neuron and different training function. By contrasting to practical casualties, the predicted results of each ANN and the predicted results of both the traditional regression and grey system method. It shows that the result of accident predicted model of ANN is more accurate than the others.At the same time, the research of ANN model of the fire disaster in MOND Method and explosion dangerous degree classification was done. The ANN model of the dangerous degree classification also trained by random selected enough samples in the range of known parameters. The training constructing model and real measurement classification for DOW/ICI main index D were done. The result of classification shows that the ANN classification is accurate, reliable and rapid.The application of ANN technique in safety evaluation has both the theoretical research significance and the project of practical value. The preliminary exploration in this area was done. And, making further improvement on works will be done in the future research.