Abstract: The safety valve is the last barrier to ensure the safety of petrochemical equipment. The more accurate the prediction of the risk of safety valve failure, the more conducive it is to the safe long-term operation of petrochemical equipment and the reasonable arrangement of safety valve maintenance plans. To efficiently and accurately predict the failure risk of safety valves, a risk assessment method based on XGBoost algorithm for safety valve failure is proposed. The method is a data-driven prediction method based on the XGBoost algorithm modelling, and the key feature parameters affecting the failure risk of safety valves are preferred to predict safety valve failure risk. The experimental data indicate that this method has good performance in predicting safety valve failure risk with an accuracy of 94.0% on the safety valve test set, which is better than the accuracy of traditional machine learning methods. In addition, this method can also provide reference basis for the determination of test and calibration cycle of safety valves.

Key words: safety valve, XGBoost, feature screening, risk prediction