Supernova Award Category

Data to Decisions

The Problem

Creating a faster-than-real-time Energy Management System (EMS) is a dream and a huge challenge for the power industry. In order to be considered faster-than-real-time, the EMS must be capable of completing EMS execution within a Supervisory control and data acquisition (SCADA) sample cycle, typically 5 seconds.

Faster-than-real-time EMS capability is key to maintaining awareness over power system events, helping prevent power system blackouts from occuring. For over a decade, power system engineers have been unsuccessful developing faster EMS applications for large-scale power systems. This project is designed to ultimately help operators in control centers to make power system operations more secure and cost-effective. So far, no commercial EMS has been able to perform at such faster-than-real-time levels.
 

The Solution

Technical challenges to curb EMS application computation efficiency include: A) The power system becoming larger and larger to meet growing demand, B) High renewable energy penetration, FACTS devices and HVDC transmission lines making power system models more complex, C) Rapid fluctuations in power demand and supply combined with fast-response devices leads to more frequent and intensive recalculations.

Parallel computing is a breakthrough solution for speeding up EMS applications. Power system engineers have investigated different parallel computing approaches based on relational database structure to improve the EMS application computing efficiency, but until now have been unable to achieve faster-than-real-time EMS.

Using TigerGraph’s graph database and computing platform, GEIRINA has achieved a faster-than-real-time EMS prototype, verified by provincial power system cases.

The results

Conventionally, power systems are modeled using relational databases in a collection of interlinked tables.  As different components of power systems are stored in separate tables, they need to be linked together using shared key values to model the connectivity and topology of the power system. Connecting or linking across separate tables (database joins) takes about 25% for power flow calculation and 35% for power grid state estimation out of the total processing time according to the results of the case study.
 
The standard approach of solving large-scale linear equations require bulky, time-intensive matrix operations. Modeling a power system as a graph (instead of a matrix) represents connections and topology more naturally. No data preparation is needed, cutting 25-35% of the generally required time for power flow calculation and state estimation.  Bus ordering and admittance graph forming are performed in nodal parallel (with all nodes performed at the same time to ensure the ordering sequence and admittance graph are in parallel). Symbolic and numerical factorization and forward/backward substitution are performed in hierarchical parallel, with nodes partitioned into levels. Nodes at the same level are calculated in parallel, starting with the lowest level. Core calculations are all conducted on graph, and solved values are stored as attributes of vertices and edges of the graph - rather than unknown variables in the vector or matrix.

Metrics

In the conventional approach, power system problems are solved as unknown variables and assigned to the X vector. To visualize the solved value, a mapping process is used to link the unknown variable to displays.

Using TigerGraph, solved values are stored as attributes of vertices and edges on a graph - forgoing the need for a mapping process. According to the test case, output visualization takes about 70% of total time for power flow calculation, and 28% for state estimation when using the conventional approach. Using TigerGraph, that portion of the time is eliminated.

This table summarizes test results of the faster-than-real-time EMS comparing with D5000, the widely used commercial EMS covering most control centers in China & many other countries.

Test on a Real Provincial 2650 Bus System

Test Systems           *SE*         *PF*          *CA*

Commercial EMS      4488ms    3817ms    18000ms

TigerGraph EMS       172ms      79ms        772ms

The total execution time of the three major EMS applications - State Estimation, Power Flow, and Contingency Analysis - is a little above 1 second combined, which is much less than the SCADA sample cycle standard of 5 seconds. With TigerGraph, we have achieved the first viable faster-than-real-time commercial EMS solution.

The Technology

TigerGraph Graph Platform & technology

Disruptive Factor

Faster than real-time Energy Management System (EMS) has been the holy grail for power management especially as any power outages directly affect the productivity and economic output / GDP (Gross Domestic Product) of the nation.  Such as a system can identify mismatches between power demand and supply, lower power consumption for non-critical parts of the grid and divert power to higher priority areas for industrial output and national security and be able to accomplish all of this in under one second.

Shining Moment

We teamed up Dr. Eugene Litvinov, senior director at the ISO-NE,  Xiaochuan Luo, technical manager at the ISO-NE, Tongxin Zheng, technical director at the ISO-NE and several respected IEEE fellows/professors to propose the creation of a faster-than-real-time computing task force to the Technologies and Innovation Subcommittee at the IEEE PES GM18 conference in Portland last week, which was approved. Additionally, seven of our papers on this topic were published and presented at the conference.