Wind turbine electricity production is negatively correlated with system load; that is, whenever the system load is high, wind production tends to be low. The result is that any analysis or modeling of system performance and capacity credit needs to pay careful attention to wind-load correlations.
The Bonneville Power Administration (BPA) conducted statistical analysis of wind production vs ambient temperature in the Pacific Northwest. BPA wind turbines are located in eastern Oregon and eastern Washington. The distribution function showed strong negative correlations. Specifically, whenever the temperature was extreme, the probability that wind production was less than 10% of nameplate was 80% (see the paper). Remarkably it did not matter whether the temperature was extremely high or low, the result was the same. BPA was less successful with direct wind-load correlation in part because weekends interrupt load-temperature correlations.
Peak system load is a primary driver of system size and capital investment. The adjacent figure presents system load during daily peak hour for PJM-2015 (black dots). [Hourly load data is searched each day to determine the daily peak hour.]
The figure also presents the daily residual load peak (red X). Residual load is the hourly system load minus the corresponding hour wind production. [Published wind data is scaled so that its annual average is 50% of system load. For each hour, scaled wind is subtracted from system load, to produce residual load. Then hourly residual load data is searched each day to determine the daily residual load peak.] Residual load is the power that is required by backup generators after wind is subtracted from the primary system load.
From the figure, wind is strong and variable from January through April. For July and August, there is very little wind during peak load hours. Indeed there are some days when residual load is almost the same as system load. The addition of wind does very little to reduce system peak power requirements and nearly full power system backup would be required to maintain system reliability.
This analysis is based on published PJM data. The main assumption is that wind added to bring wind production up to 50% of total load has production similar to existing wind [this would be strictly true is wind addition maintained the same footprint, the same location of existing wind farms]. Wind-load correlations an important part of wind system performance. Many academic wind system models assume implicitly or explicitly that wind is independent of load.