1 High Temporal Resolution Load Variability Compared to PV Variability Matthew Lave, Jimmy Quiroz, Matthew J. Reno, Robert J. Broderick Sandia National Laboratories, Livermore, CA and Albuquerque, NM, 94550 and 87185, USA Abstract — While solar variability has often been quantified and its impact to distribution grids simulated, load variability, especially high-frequency (e.g., 1-second) load variability, has been given less attention. The assumption has often been made that high-frequency load variability is much smaller than PV variability, but with little evidence. Here, we compare load and PV variability using 1-second measurements of each. The impact on voltage regulator tap change operations of using low- resolution (e.g., 15- or 30-minute) interpolated load profiles instead of 1-second is quantified. Our results generally support the assumption that distribution feeder aggregate PV variability is much greater than aggregate load variability. I. INTRODUCTION Many studies (e.g., [1-3]) have demonstrated the impact of PV variability to voltage fluctuations on distribution feeders and hence voltage regulator tap change operations. The spatial smoothing due to geographic diversity of PV modules has been well documented (e.g., [4-6]) and modeled (e.g., [7-9]). Additionally, the importance of using high-frequency PV samples for distribution grid studies was shown in Lave, et al. [2], where errors in simulated tap change operations of 20% or more resulted from using low-frequency (5-minute or 15- minute) PV power samples instead of sub-minute. However, comparatively little analysis of load variability exists. Historically, load measurements have been low- frequency (e.g., 15-minute resolution), sparse (e.g., only measured at the distribution substation), and low quality (e.g., low accuracy and reliability due to lack of maintenance), but recently load data availability (e.g. AMI data) and resolution have been improving. So far the assumption in most PV integration studies has been that high-resolution load data is not necessary because the PV variability is much larger than load variability, but this has not been directly verified. In this work, we present a direct comparison of the variability of both PV and load on a distribution feeder using measured data from Ota City, Japan. Additionally, we investigate the impact of using low resolution load, PV, or both on distribution simulation accuracy. II. DATA 1-second load and PV power output data from nearly 500 homes in Ota City, Japan with PV was used for this analysis. The maximum load of all houses was 1.0MW and the installed capacity of PV was 1.9MW. The data is described in detail in [10]. Figure 1 shows the load and PV profiles for the aggregate of all houses, binned into averages by month of year and hour of day. The Ota City load had both a morning and an evening peak, due in part to household heating demands. These peaks are largest in the winter; in the summer, loads are low. Houses in Ota City do not typically have air conditioning. This load profile is different from many United States load profiles. In the Southwestern United States, for example, loads tend to have a single daily peak, with maximums in summer afternoons. The PV profile generally follows seasonal solar cycles, but suffers from reduced power output in July due to many cloudy days. In this way, it may be similar to locations in the United States that experience summer cloud cover/fog, such as coastal San Diego or San Francisco. III. PV VS. LOAD VARIABILITY Figure 1: Average load and PV power output values for each month of year/hour of day combination for the aggregate of 483 houses in Ota City, Japan.