Abstract: High-dimensional, non-stationary, and strong noise characteristics of drilling parameters under deep and complex geological conditions result in low accuracy and poor timeliness in anomaly detection. To address this issue, this paper proposes a method for identifying abnormal drilling parameter data based on extreme value analysis. By establishing a generalized Pareto distribution model for extreme value analysis of drilling parameters and combining it with a multi-parameter correlation discrimination strategy, a complete set of abnormal data identification methods is constructed. The experimental results show that the accuracy of this method in drilling parameter anomaly detection reaches 94.7%, which is 12.3%, 9.0%, 5.4% and 2.2% higher than 3σ method (82.4%), box plot method (85.7%), isolation forest (89.3%) and LSTM-AE (92.5%) respectively. It also exhibits strong robustness under various abnormal working conditions such as stuck drilling, lost circulation, and well kicks. The research results not only enrich the theory of drilling parameter data analysis, but also provide technical support for intelligent drilling in oil and gas fields, which has important practical significance for improving drilling safety and efficiency.