Machine learning can help energy suppliers better identify faulty or compromised components in power grids. A research project led by the Massachusetts Institute of Technology describes a technique that allows modeling complex interconnected systems consisting of many variables whose values change over time. By matching connections in these so-called multiple time series, the Bayesian network can learn to identify anomalies in the data.
The state of the power grid can be made up of a variety of data points, including the magnitude, frequency and angle of the voltage in the entire network, as well as the current. Anomaly detection depends on identifying abnormal data points that can be caused by things like cable breakage or insulation damage.
“In the case of the electric grid, people tried to collect data using statistics, and then determine the detection rules with knowledge of the subject area. For example, if the voltage rises by a certain percentage, the network operator should be warned. 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 data using advanced machine learning methods,” the experts explained.
The new method uses unsupervised learning to identify abnormal results, instead of using manually created rules. When the researchers tested their model on two private datasets recording measurements of two interconnects in the United States, they revealed the superiority of the model over other machine learning methods based on neural networks.
The general method of detecting abnormal data changes can even be used to signal an alarm in the event of a power system hack.
“It can be used to detect the devaluation of the power grid failure for the purposes of cyber attacks. Since our method is essentially aimed at modeling the power grid in a normal state, it can detect anomalies regardless of the cause,” the experts noted.
According to the researchers, the model cannot indicate the exact cause of the anomalies, but it can determine which part of the power system is out of order. The model can be used to monitor the state of the power grid and can report a network failure within one minute.