Abstract In this paper artificial neural network (ANN) and adaptive neuro-fuzzy inference system (ANFIS) were applied to predict the separation percentage of gas and gas condensate in a wellhead separator in Naar oil field (Boushehr province, IRAN). The operating parameters including valve opening percentage, gas flow, design pressure, and design temperature are considered as the inputs of the models. The accuracy of the proposed models were evaluated using statistical parameters such as correlation coefficient (R2), average percent relative error (APRE), average absolute percent relative error (AAPRE), and root mean square error (RMSE). Based on the achieved data, R2 values were 0.9691 and 0.9807 for ANN and ANFIS models, respectively, while the values of RMSE were 6.117 and 4.57 for the applied models, which denote the higher accuracy of ANFIS model. Moreover, risk analyzing and consequence assessment of probable explosion of separator using PHAST (Process Hazard Analysis Software) software showed that inspection of separators is very important. Considering the calculated results, it can be concluded that ANFIS was better than ANN in prediction of gas and gas condensate separation percentages, since its output showed higher affinity to the real data. Generally, the findings obtained from the current work suggest that it is possible to predict the separation efficiency of a wellhead separator using intelligent systems.

中文翻译：

---

凝析气藏井口分离器效率预测与风险评估

摘要 本文应用人工神经网络（ANN）和自适应神经模糊推理系统（ANFIS）预测Naar油田（伊朗Boushehr省）井口分离器中天然气和凝析油的分离率。包括阀门开度、气体流量、设计压力和设计温度在内的操作参数被视为模型的输入。使用相关系数 (R2)、平均百分比相对误差 (APRE)、平均绝对百分比相对误差 (AAPRE) 和均方根误差 (RMSE) 等统计参数评估所提出模型的准确性。根据获得的数据，ANN和ANFIS模型的R2值分别为0.9691和0.9807，而应用模型的RMSE值分别为6.117和4.57，这表示 ANFIS 模型的精度更高。此外，使用PHAST（过程危害分析软件）软件对分离器可能爆炸的风险分析和后果评估表明，分离器的检查非常重要。考虑到计算结果，可以得出结论，ANFIS 在预测气体和气体冷凝物分离百分比方面优于 ANN，因为其输出显示出与实际数据更高的亲和力。一般而言，从当前工作中获得的发现表明，可以使用智能系统预测井口分离器的分离效率。考虑到计算结果，可以得出结论，ANFIS 在预测气体和气体冷凝物分离百分比方面优于 ANN，因为其输出显示出与实际数据更高的亲和力。一般而言，从当前工作中获得的发现表明，可以使用智能系统预测井口分离器的分离效率。考虑到计算结果，可以得出结论，ANFIS 在预测气体和气体冷凝物分离百分比方面优于 ANN，因为其输出显示出与实际数据更高的亲和力。一般而言，从当前工作中获得的发现表明，可以使用智能系统预测井口分离器的分离效率。