基于支持向量回归机及混沌理论的气田监测预警技术

Gas Field Pre-alarm Technique Based on Support Vector Regression and Chaos Theory

针对气田开发过程中安全监测预警问题,结合支持向量回归机(Support Vector Regression,SVR)和相空间重构,提出了气田监测预警方法。对训练样本进行相空间重构,提取用于SVR离线训练的自变量和因变量。通过遗传算法获取最佳SVR参数,采用序列最小优化算法(Sequential Minimal Optimization,SMO)算出Lagrange乘子,构造在线预测模型,对监测信号进行预测,判别未来10个时间步长内监测值是否超出安全范围,在时间相对较充足的情况下采取有效措施,避免事故的发生。现场应用表明,所述方法在混沌信号预测方面具有较高精度,可为气田监测提供准确的预警信息。

Aiming at the prealarm problem of-safety inspection during the development of gas fields, a novel method combined Support Vector Regression （SVR） and Phase Space Reconstruction （PSR） is proposed. Firstly, Independent and dependent variables are obtained through the PSR of training samples. Secondly, SVR parameters are obtained through Genetic Algorithm（C-A）. Then, Lagrange multipliers are calculated by Sequential Minimal Optimization（SMO） algorithm, and then an online prediction model can be constructed. Finally, the future ten monitoring data are predicted for prealarming to avoid accidents in case of sufficient time. Simulation and field data processing show that the method has higher precision for chaotic signal prediction, and the accurate pre-alarming information can be provided for gas filed monitoring.