IEEE Electrification Magazine - March 2016 - 28

Figure 5. A sample of the graphical user interface for DEx.py.

G
G
Bus 2

Bus 1

20 MW

6
4

Bus 3
85 MW

7
Bus 4

Bus 5

8
40 MW

9
20 MW
Bus 6
20 MW

Figure 6. The Roy Billinton test system.

have at least one subfeeder (level 2). Each primary feeder and
subfeeder consists of at least one load point, at level 3. Each
load point may consist of additional subload points (level 4).
The upper bounds for the number of primary feeders, subfeeders per primary feeder, load points per feeder, and additional
subload points per load point per feeder for each transmission
node are set by the user to create a new distribution topology.
The link between the primary feeders and subfeeders can
be configured as an overhead line or an underground cable.
The load points and subload points in the distribution

I E E E E l e c t r i f i cati o n M agaz ine / March 2016

topology represent the connection points for the low-voltage
(LV) network where the residential loads are connected, as
shown in Figure 8. The LV network consists of step-down distribution transformers, triplex nodes, and triplex lines for each
of the three phases. The residential loads are connected at the
end points of each phase of the LV network at distribution
voltage level. At the time of writing, only residential homes are
modeled in DEx.py; however, we are working to add commercial- and industrial-type loads as well. Once the distribution
system network is generated, the load profiles need to be generated for each of the newly attached homes. This leads to the
second step of DEx.py: end-user load population.
To split the total load at the PCC among individual house
loads, we designed a combined top-down, bottom-up
approach. First, the total bus load is split into the aggregate
load of each individual house connected on a distribution
network, using a stochastic approach (top down). One-dimensional random walk theory is used to generate the load profile for the residential loads that are distributed across a distribution feeder network. Here, the load profile for each residential load is scaled from a time-varying system load at the
transmission node and used as the input for the one-dimensional random walk process. Load curves for a small subset of
the residential loads that populate a GridLAB-D feeder are
presented in Figure 9. Each house has a different, yet realistic,
load curve; the ensemble sum of the residential load curves
represents the load curve at the PCC.
To obtain the usage pattern of individual smart grid
technologies at each household (bottom up), we use a
method based on queueing theory. We model each household as an infinite capacity computing server with asset
usage analogous to task arrivals. In the Mt/G/∞ queue,
applications arrive nonhomogeneously with Markovian
distribution (i.e., time-varying Poisson process), generally
distributed execution times, and infinite capacity. By

http://www.DEx.py http://www.DEx.py http://www.DEx.py

Table of Contents for the Digital Edition of IEEE Electrification Magazine - March 2016