基于自适应支持向量回归机的集输系统压力监测异常值识别

Anomaly Detection in Pressure Monitoring Data of Natural Gas Gathering System Based on Adaptive Support Vector Regression

为保障气田安全开发,针对气田集输系统压力监测数据异常值识别问题,提出了一种基于自适应支持向量回归机(ASVR,Adaptive Support Vector Regression)的方法。该方法将集输系统各关键节点压力值从上游到下游组成序列,取一组正常实测信号作为训练样本,以预测值和实测值间的均方差最小化为目标函数,通过遗传算法获取最佳惩罚因子、不敏感损失函数参数和核函数参数;利用序列最小优化算法(SMO,Sequential Minimal Optimization)对各工况实测信号进行回归拟合,通过非边界支持向量拟合误差判别监测数据是否为异常值,并用回归值对异常值进行修正。对现场信号处理表明,该方法可准确地模拟集输系统各关键节点压力间的函数关系,并能准确识别压力监测数据中的异常值,为安全控制系统提供正确的信号,对气田安全、高效开发具有实用价值。

Aiming at anomaly detection in pressure monitoring data of natural gas gathering system,a novel method based on ASVR（Adaptive Support Vector Regression） was proposed.Normal pressures of key points in the natural gas gathering system were used as training sample for SVR,and the penalty factor,insensitive loss function factor and kernel function factor were found by genetic algorithm（GA）,which used the minimization of mean square error between prediction and measurement values as the objective function.Then regression pressure values under each working condition were calculated by SMO（Sequential Minimal Optimization） algorithm,and then anomaly data was detected and amended by the fitting error of non-bound support vectors.Engineering data processing confirmed the excellent performance of the proposed method,and the anomaly data in pressure monitoring could be detected correctly,which is a matter of practical significance for the safe and effective development of gas field.