Research on the Quantitative Evaluation and Methods of Drilling Overflow Risk Based on While-Drilling Information

Abstract

In order to solve the problems of insufficient generalisation ability and poor interpretability of evaluation results in the complex geological environment of the current data-driven overflow risk assessment model, this paper proposes a quantitative overflow risk assessment method based on the combination of CNN-LSTM data-driven model and fuzzy reasoning system. The data features were extracted, the overflow risk was predicted, the fuzzy comprehensive evaluation method was further adopted, the membership function was generated based on the normal distribution fitting historical data, the risk probability threshold was dynamically calibrated, the risk was refined into low, medium and high levels, and the expert experience was integrated through fuzzy reasoning rules to improve the transparency and flexibility of overflow risk assessment. The experimental results show that the accuracy of the proposed method is 99.95%, and the false alarm rate is only 0.5%, which significantly improves the adaptability and reliability of risk assessment compared with the traditional static threshold method, and enhances the interpretability of the model through feature visualisation and dynamic rules, which provides intelligent decision support with both high precision and strong robustness for drilling safety.