Avoidance of the high societal costs of electricity blackouts requires that the amount of power generation in a balancing authority area match exactly, on a near-instantaneous basis, the system load, net of losses and interchange with other balancing authority areas. In a truly smart grid, the actual flow of electricity along a given transmission line would equal the expected flow. But electricity flows on an alternating current transmission network follow the paths of least resistance and thus there can be large differences between scheduled and actual electricity transmission flows between any two electricity control areas. One example of this are the flows between PJM, the regional transmission organization that coordinates the movement of wholesale electricity in all or parts of 13 states and the District of Columbia, and the New York Independent System Operator (NYISO). Over the period 1 June 2007 through 31 December 2008 the root-mean-squared-errors of the actual electricity flows between PJM and NYISO were approximately 78 percent of the scheduled flows. Indicative of the challenge faced by system operators, the inadvertent flows on this interchange are so large that PJM can sometimes inadvertently export electricity to New York even when it is a scheduled net importer. This is not an isolated case. At PJM's interface with the Michigan Electric Coordinated System, the root-mean-squared-errors of the actual electricity flows are almost 500 percent. To put this number in perspective, the correlation between actual flows and scheduled flows at this interface over the period 1 June 2007 through 31 December 2008 was about -0.02. Ideally, the correlation should equal 1.00. This is truly a very challenging state of affairs. It is no wonder that various PJM State of the Market Reports have called for improved analysis of electricity flows across the Eastern Interconnection.

In this paper, we present evidence that electricity flows between control areas can be forecasted using multivariate econometric methods. The modeling framework makes use of temperature data, proxies for the day-ahead expected level of transmission conductivity, measures of “network effects,” and a measure of geomagnetically induced currents (GICs). The latter variable is included based on evidence that GICs can affect the paths of least resistance. The results of the analysis indicate the level of operational uncertainty faced by transmission system operators can be significantly reduced via forecasting.