The introduction of roof top solar generation has given rise to a grid supplied load profile that is conditional on solar conditions. This can reduce the accuracy of the load forecasts that are used to optimize the grid’s resources. The data science methods employed by EEDS can offset this loss in predictability. In the case of New York City, the method works because of the magnitude of the autocorrelations in grid supplied load.
Figure 1. The autocorrelations in the hourly grid-supplied load in NYC, Jan 1, 2015 to Dec 31, 2018
The correlation in the grid-supplied load in market period t and the load level in market period t-1 equals 0.9820. This correlation subsequently declines and reaches a nadir at lag 12 with a value of 0.2577.
The correlation subsequently rises until it attains a value of 0.8959 at lag 24, the lag associated with the operating period 24 hours previously.
This pattern in autocorrelations continues in subsequent days. As a result, the electricity load in hour t is highly correlated with the load levels in hours t-48, t-72, t-96, t-120, t-144, t-168, etc.
This has obvious implications for being able to predict the level of load in hour t given that the past levels of grid supplied load are known,