Electric submersible pump anomaly monitoring based on unsupervised multivariate streaming data analysis

Electric Submersible Pump is an essential artificial lift equipment in the petroleum industry, and its anomaly detection is of great significance for maintaining system performance and ensuring normal production. In this paper, we propose a novel unsupervised multivariate streaming data analysis approach that directly uses the raw structured data as input. It eliminates the need for complex modeling and parameter optimization. Then, metrics that can reflect the state of the system are built through feature mapping and fusion. Based on the fused metric, the smoothness of the metric is monitored in real-time by the control chart detection method, which does not require pre-training of known label samples or any domain knowledge. Finally, the feature selection algorithm ANOVA is introduced into the algorithm to locate the most relevant features to anomalies. We evaluate the algorithm using the ESP signals from three regions in Penglai, Bohai Oilfield, China. The experimental results show that the algorithm proposed in this paper can capture system performance anomalies in time and realize the location of suspicious features. Moreover, the inexpensive computational costs and the second-level inspection speed make it ideal for on-site deployment.