Abstract: To address the issue of low accuracy in existing environmental monitoring and early warning methods for oilfield bases, an online early warning method for abnormal data based on LS-SVM is proposed. The least squares support vector machine (LS-SVM) is adopted to construct a regression model following the principle of structural risk minimization. The complexity of the model is characterized by kernel function parameters. Predictions are then performed on environmental monitoring data, and the data are normalized to unify their scales. Empirical parameters are set to describe the prediction error. When the error satisfies the standard normal distribution condition, an anomaly is determined. The error sequence is then compared with a threshold, and an early warning is triggered once the error exceeds the threshold range. Experimental results show that the proposed method can effectively identify abnormal fluctuations in monitoring data with an early-warning accuracy of 100%. It can fully detect abnormal time series and markedly improve the timeliness and accuracy of environmental anomaly early warning for oilfield bases.

Key words: LS-SVM, abnormal data, environmental monitoring, early warning