Abstract: In order to ensure the safe and efficient production of sour gas wells, a model was built for predicting sour gas wells integrity risk level on the basis of PCA and BP neural network method. Before building the model, a bow-tie model was built for identifying integrity failure risk factors. The factors were quantified by fuzzy evaluation method. Comprehensive indexes were extracted by PCA. An application of the prediction model was made to 5 sour gas wells. The results show that the input data of BP neural network can be reduced from 28 to 4 by PCA, and the prediction accuracy of risk grade is higher than that of BP neural network without PCA, and that the model built on the basis of combination of PCA and BP neural network can be used to identify the risk factors of integrity failure in sour gas well development and improve the risk grade prediction technology of sour gas wells.

Key words: sour gas well, wellbore integrity, principal component analysis(PCA), risk assessment, BP neural network, fuzzy evaluation