As part of a project led by the Massachusetts Institute of Technology, researchers have described a technique for modeling complex interconnected systems from many variables whose values ​​change over time. The "Bayesian network" matches connections across these multiple time series and learns to spot anomalies in the data.

The new method uses unsupervised learning instead of hand-crafted rules to detect anomalies.

Such a system would enable energy providers to better identify faulty or compromised components in the power grid. Thus, the state of the electrical network can be composed of many data points, including the magnitude, frequency and angle of the voltage in the entire network, current. The system will detect anomalous data points that could be caused by a broken cable or broken insulation.

“In the case of the power grid, people tried to collect data using statistics and then define discovery rules with domain knowledge. For example, if the voltage rises by a certain percentage, then the network operator must be alerted. Such systems, even enhanced by statistical data analysis, require a lot of work and experience. We can automate this process as well as extract patterns from the data using advanced machine learning techniques,” the authors of the study say.

The authors of the development tested their model on two private datasets recording measurements of two interconnections in the United States, and revealed its superiority over other machine learning methods based on neural networks.

The general data anomaly detection method can also be used to generate an alarm in the event of a power grid breach and to detect the devaluation of a power failure for cyber attack purposes. “Because our method essentially aims to simulate an electrical network in a normal state, it can detect anomalies regardless of the cause,” the experts say.

The model cannot pinpoint the exact cause of the anomalies, they say, but determines which part of the power system is failing. The model can be used to monitor the state of the power grid and report a network failure within one minute.

bbabo.Net