Skip to main content

Making the Support Vector Machine-Based Relays Secure

Researchers: Marija Ilic, Ozan Tonguz

Research Area: Next Generation Secure and Available Networks

Abstract

Making the Support Vector Machine (SVM)- Based Relays Secure

Today’s electric power systems are very large and complex interconnected networks. The key role of electric power systems in modern society imposes stringent requirements for an acceptable level of reliable and secure operation, even under unusual circumstances and adverse conditions such as blackouts.   While such conditions may occur with very low probability they may have high impact; hence, the system must be sufficiently adaptive to minimize the adverse effects on equipment safety and the continuity of electricity service. To keep the continuity and the stability of the power system, various procedures are put in place to adjust the system when hard-to-predict large equipment failures occur and to disconnect the service only where it is essential to do so to provide service to the rest of the customers. These procedures are currently a combination of system operator’s actions, automated control, and automated system protection.

This project concerns, in particular, the problem of protection for transmission lines. The primary role of a relay dedicated to a piece of equipment is to disconnect this equipment in order to prevent it from getting damaged when the electrical conditions (e.g., the current going through that element) are unacceptable. The protection has by and large, except for some special protection schemes (SPS), fixed thresholds of variables to which it responds when disconnecting equipment. While different traditional relays respond to different physical variables, they all share a common characteristic that their thresholds are set under fixed conditions, and are not adjusted as the system operating conditions vary. They are, instead, deployed and left alone. Consequently, currently implemented relays are not sophisticated enough to protect equipment against serious failures such as blackouts. In some situations they are not adaptive enough to discriminate between the fault and non-fault (normal) conditions, or to react correctly to faults. This malfunctioning of relays has been inherently associated with all major system blackouts, and has been known to have contributed to the accelerated cascading of the initial equipment failure (outage). It is widely recognized by the industry that more intelligent relays are needed which would be capable of reducing the damage to the power systems hardware and, at the same time, localizing the effects of outages without causing cascading equipment failures.

In our group at Carnegie Mellon University (CMU) we are pursuing work toward a new generation of protective relaying. In this project, we improve the protection relays via applying support vector machine classification. Support Vector Machine Classification Based (SVMCB) smarter relays determine their decision boundary based on more than one feature. SVM classification helps to improve the performance of the smarter relays when dealing with complex fault conditions in large scale systems. In our experiments, to improve the accuracy of smarter relays on complex conditions, more than one feature is taken into consideration at one time. Besides the magnitude of current, which was selected to be the representative feature on conventional relays, phase of current, magnitude of voltage, phase of voltage, real power and reactive power are all candidates in SVM based smarter relays.

Experiments, which are done on IEEE 118 bus systems,  conducted at CM show that a high accuracy can be achieved when SVMCB smarter relays are making the decision to discriminate the zone 1 faults, zone 3 faults and normal conditions. Besides these, simulation results indicate that the real and reactive power are two major features in most of the cases. These two features are neglected in conventional relays, which is one of the reasons why conventional relay may malfunction in some critical conditions. The application of these two features in SVM based-relays helps increase the accuracy in critical decision making. Moreover, simulation results also indicate that the SVMCB smarter relays can achieve high accuracy even in the “N-1 condition”, which is a more critical condition during systems control process.  Besides decision-making accuracy, scalability has also been demonstrated by simulation.