From GridLAB-D Wiki
Jump to: navigation, search

Background

Typical distribution feeder model databases contain limited customer information, such as residential versus industrial or peak load data, but little to no information about the behavior of the loads. To perform analyses on the effects of strategies which modify the behavior of the end use loads, such as conservation voltage reduction or demand response, it is important to accurately capture the base line behavior. This involves using a parameterized model combined with known system characteristics, such as weather and HVAC penetration levels, to match historical data across multiple time horizons. This section will describe two methods used in previous studies to effectively describe advanced system load models from traditional power system information. Each of these described methods should not be considered the only methods, but rather an indication of the methodologies required to perform this type of calibration. The type of analysis to be performed, and the amount of information available to the modeler, will determine the population method to be used.

Previous Work

In previous work, a couple approaches have been used. The first, when general or prototypical feeders are created and second, when specific feeder information is available.

Generalized Feeders

In order to generate preliminary estimates of SGIG project benefits, the Pacific Northwest National Laboratory (PNNL) utilized the GridLAB-D simulation environment to conduct extensive simulations on representative technologies. A selection of representative technologies was simulated on a set of prototypical feeders to capture the national level benefits of the technologies [1]. This approach did not have SCADA data available, and was therefore only calibrated to the peak load study. This made the calibration portion of the study far simpler; however it is not able to represent any particular feeder in the United States.

The taxonomy of prototypical feeders was originally populated with a series of spot loads representing a standard peak load study. Each spot load was classified as residential, commercial, agricultural, or industrial. In this analysis, due to the broad nature of industrial and agricultural loads and the difficulty in accurately representing these loads, each of these loads was re-classified as commercial, leaving only residential and commercial loads. Each load was replaced with building models appropriate to the region of the United States where the prototypical feeder was located. The parameters of each home were determined by the climate region the feeder was located in. The number of buildings on the model was then adjusted until the peak annual load matched the available peak load flow study.

More details about this process can be found in [2]. The Matlab(R) scripts used to implement this on the prototypical feeders is available in the GridLAB-D source tree.

[1] K. Schneider, Y. Chen, D. Chassin, R. Pratt, D. Engel, and S. Thompson, “Modern Grid Initiative Distribution Taxonomy Final Report”, PNNL Report 2008.

[2] Fuller, J.C., Prakash Kumar, N., Bonebrake, C.A., "Evaluation of Representative Smart Grid Investment Grant Project Technologies: Demand Response", PNNL Report 2012.

Specific Feeders

In this analysis, the original powerflow loads were defined as static spot loads, where blocks of individual commercial and residential loads were summed to a single peak spot load on the primary system (no secondary voltage loads were defined). To more accurately classify the loads, Google Earth© images of the feeders were located and the physical dimensions of the feeder overlaid. The loads provided by the original model were then manually classified by the type of building found at that location, and were broken into nine different load types via visual inspection. These were classified as Residential 1-6, Commercial 1-2, and Industrial. Each load classification describes the properties of the load in that area, and the details that describe each type of load will be further described.

By defining each building as older or newer, and larger or smaller, approximate physical properties for those homes could be assumed. These were then used to define multiple building models at each load location, depending upon the type of building that was found through observation in Google Earth©. Defining these properties gives insight into the benefits of voltage reduction not only at a single given load level, but as a function of seasonal and daily variations in load. While a particular building model at that location does not accurately represent a specific building in reality, the aggregate of the distribution of the buildings should approximate the response of all of the real buildings. This is an important distinction. The goal isn’t to model every load within the system perfectly, but to capture the aggregate behavior of all the loads on the feeder through smaller individual load representations. Within each building, appliance loads were also modeled, as will be seen in the following sections.

Once each of the points of interconnection were classified, it was necessary to fully represent the load. Because of the complexity of end use load behavior, load models can be divided into two distinct classes: those without thermal cycles and those with thermal cycles. Loads without thermal cycles consume energy in a time-invariant manner, with the exception of voltage variations. Specifically, there is no control feedback loop. For example, a light bulb will consume energy when turned on in a fixed or voltage-dependent manner. In contrast, a load with a thermal cycle, such as a hot water heater, will have a varying duty cycle dependent on the supply voltage. If the supply voltage is lowered, the hot water heater will draw less instantaneous power, but it must remain on for a longer period in order to heat the same mass of water.

At each spot load location from the original feeder model, the load was replaced with a combination of building and ZIP load models. By varying the relative number of building models to the number of ZIP models, then varying their relative magnitudes, a reasonable approximation could be found that matched SCADA data for the entire year. By fitting data on approximately 6 to 12 days per year per feeder, with appropriate seasonal variation, it was found that the overall difference between simulated and actual demand (and reactive power) could be limited to approximately 5% of the total demand at all times throughout the year, except during times of topological changes (for example, if a large amount of load was shifted from one circuit to another). However, to develop an accurate annual load profile for the feeders each of the individual end use loads were calibrated. Relative loading across a feeder was approximately equal to that specified for the original database feeder models, but with added time-dependent outdoor temperature, solar insolation, voltage, etc. To create a model that accurately matches provided SCADA data over the provided time interval, a number of adjustments were needed. This process is further described in OMF Calibration.

Current OMF Work

See also

OMF_Scripting_Documentation

OMF Conversion Process

OMF Calibration

OMF Weather Extractor