(after
Esri, 1997)
Kevan Q. Newton, David E.
Schirmer
Abstract
As Electric Vehicles (EVs)
make their way onto the roads of California, Southern California
Edison (SCE) is taking steps to ensure that it can meet the future
electric demand these vehicles will impose upon the existing infrastructure.
SCE's first concern with regard to this new demand is localized
load impact--that is, where are the substations, circuits, and
other equipment within the SCE service territory where EV load
is expected to tax existing capacity? Predicting the infrastructure
that will be impacted is all but impossible without accurate spatial
definition of the geographic extent of the equipment served by
each substation - its 'sphere of influence' (SOI). Once defined,
a spatial context will exist upon which predictive EV purchasing
behavior data may be overlaid, and potential load problem areas
can be identified and remedied.
This paper will detail the
methodology employed in determining spheres of influence for Southern
California Edison substations, for the purposes of load forecasting
within the wider context of an electric vehicle program.
Introduction/Background
Electric Vehicles promise
to become an increasingly common feature on the California landscape.
Providing electric service to more than four million customers
in California, SCE is taking steps to understand what impact these
vehicles will have on the efficient distribution of electricity
within the area it serves. Southern California Edison has calculated
that EV electrical demand does not pose a significant challenge
to its generation system (SCE, 1996). Rather, SCE is concentrating
on how this new demand will impact the distribution system.
Paramount to this effort is knowing where the highly localized
demand created by EVs will take place, within the context of the
existing distribution infrastructure. This has been accomplished,
in part, through the daunting task of defining the specific area
served by each distribution substation-- its sphere of influence.
Armed with this information, electric distribution planners may
now view spatially the area served by a substation, and translate
purchase forecasting data into real-world impacts to the distribution
system.
The first and most important
step is determining when and where this increased load from Electric
Vehicles will impact the distribution system. To do this, SCE
relies on the complex California Alternative-Fuel Vehicle Demand
Forecasting Model developed by the Institute of Transportation
Studies (ITS) at the University of California at Irvine and Davis.
Utilizing a variety of socio-economic data derived from the census,
transportation studies, and surveys, the model is capable of yielding
crucial information in terms of where likely EV buyers reside
and work, typical driving behavior, and destination information
(see Morales, 1995, Schirmer et al., 1996). It has been calculated
that a household of four will double its electrical demand by
charging an EV at home (SCE, 1996). This has important implications
for SCE because mature neighborhoods within SCE's territory, whose
demand was not expected to increase much in the future, now have
the potential to increase electricity demand substantially. Furthermore,
following Tobler's Law of spatial auto-correlation, it is not
unreasonable to assume that the impact EVs posses is highly localized,
compounding SCE's reason for concern.
Spheres of Influence
While the purpose of this
paper is not an introduction to electrical distribution engineering,
it is perhaps necessary to backtrack a bit and lay out a couple
of key points. First, substations feed circuits which, by way
of equipment (e.g. transformers), are attached to meters. Second,
the collection of circuits, equipment, and meters which feed off
of a substation define its Sphere of influence. Given these
points, determining the actual existing demand on a substation,
or its load, is already measured and thus is not earth-shattering
information. What has never before been accomplished at SCE is
the assembling of all the sources of this demand and the space
they occupy.
SCE, like most large organizations,
suffers from data fragmentation. SCE maintains large volumes
of data about meters, circuits, equipment, transmission lines,
substations, etc., much of which resides in different databases,
maintained by separate departments, serving different purposes.
Extensive effort has been made in an attempt to bring these systems
together, but making the fundamental link between meter and substation,
and vice versa, has proven to be an all but impossible task given
the existing database structure. In other words, determining
the connectivity between a given substation and the electric meters
it serves has not been possible. However, by applying a GIS approach
to the existing datasets, SCE's GIAS group has pioneered the delineation
of each substation's Sphere of influence.
Defining each substation's
sphere of influence has implications for predicting where, within
SCE's distribution infrastructure, future load will occur in
terms of both 'traditional' load, and the less traditional load
associated with EVs. When planning for a new residential or commercial
development (traditional load), distribution planners can now
know instantly which substation will be impacted. Regarding EV
load, the predictive results of the ITS Microsimulation Model
can be aggregated to the sphere of influence level, allowing distribution
engineers a window into the load future of each substation.
Methodology
A key dataset utilized for
this analysis consists of traditional AM/FM-type CAD drawings
of individual circuits. These .dwg files first needed to be converted
into a format usable by ArcInfo. To do this each circuits' drawing
file was brought into ArcView using the CADReader extension.
The drawing file was then queried for the desired layer, and subsequently
converted to a shapefile. Each circuit shapefile was then aggregated
to the appropriate SCE district level to yield a single large
shapefile for each district. A typical district is comprised
of approximately 300 circuits. (The aggregation of the circuits
to one of 35 SCE districts is appropriate, as distribution circuits
rarely cross district boundaries. This issue will be discussed
further in the 'Limitations' section.) The shapefile is then
converted to an ArcInfo coverage where it is projected into the
appropriate coordinate system. One key attribute of the arcs
within this coverage was the numeric ID of the substation to which
each circuit was attached.
The problem, however, of
creating two-dimensional, extensive polygons (spheres of influence)
from collections of one-dimensional circuit lines still remained.
In other words, how do we assign the space around these arcs
the attributes of the most appropriate nearby arc? A variety
of methods for creating these extensive polygons were attempted,
but only three will be described here. Before describing the
chosen method, other procedures and the reasons they were rejected
will be addressed.
One common thread among
all of the methods described here is that each required the input
feature class be a point. After extensive trial and error, a
determination was made that in order to provide a reasonably high
degree of accuracy, while at the same time, avoiding the creation
of unmanageably large datasets, a resolution of three meters was
suitable. The ARC command DENSIFYARC was used such that circuit
line vertices were at most 3 meters apart. All vertices were
then converted to points using the ARC command ARCPOINT, with
the numeric ID of the substation serving as the spot item. But
the problem still remained how to convert these points to areas.
The first rejected methodology
was to create a TIN (using the ARC command CREATETIN) from the
now-'densified' circuit points, and then to convert that TIN to
a polygon coverage (using the ARC command LATTICEPOLY). This
coverage would then be dissolved to create one sphere of influence.
The second rejected method
was to KRIG the circuit points such that the resulting surface
would be dissected into each substation's SOI, again to be converted
to a polygon coverage.
While both of these methods
were somewhat successful, the problem with each is the same.
Each incorrectly (for our purposes) assumes that the data are
continuous. That is, values are assigned to the area between
circuit points that belong to different substations by interpolating
the values of the substation ID. Thus, the area midway between
points with substation ids of 5200 and 5450 would be assigned
a value of 5325. Obviously this is of little value, as the substation
data are discrete.
In light of this very important limitation, the most appropriate method was to create Thiessen (also known as Voroni) polygons from each of the circuit points. Thiessen polygons are created such that each region contains only one point and that each region has the unique property that any location within that region is closer to the region's point than to the point of any other region (Esri, 1997) The idea is that given a two-dimensional array of sampling points, the 'best' information about an unvisited point can be gleaned from the data point nearest to it (Burrough, 1986). The following illustrates the process of creating Thiessen polygons:
(after
Esri, 1997)
(A proximal tolerance of
.01 meters was used in the Thiessen command). Once we had a Thiessen
polygon for each circuit point, those polygons with the same substation
value were dissolved together, and the collection of spheres of
influence for that district was complete (from now on referred
to as the "district SOI coverage"). But Spheres of
Influence are only valuable if they are one continuous coverage
for SCE's entire service territory - not district by district.
The next challenge was how to bring together all of the SOI's
for the entire service territory.
As you can see from the
illustration above, in the bottom-right diagram, the TIN bisectors
extend out until they reach either another bisector OR the edge
of the mapextent of the input coverage. Because the SOI's could
only be created one district at a time, each based on its own
input mapextent, when it came time to append all of the districts
together, there were areas along the border of adjacent districts
which overlapped considerably. This meant that this area of overlap
would be assigned to TWO different SOI's - an impossibility in
the world of electricity distribution.
Resolving these areas shared
by two SOI's was a lengthy process. When the circuit lines for
a district area were drawn up against the district SOI coverage,
it was clear that the district SOI coverage extended far beyond
the area covered by that district's circuit lines. Since the
area not served by circuit lines shouldn't be included in the
district SOI coverage, these outlying areas had to be removed.
This was done by creating a 300 meter grid around each of the
district's circuit lines on the assumption that no customer meter
would be further than 300 meters away from the circuit line that
fed it. Each district's grid was then turned into a polygon coverage
and used to clip out only the desired part of the district SOI
coverage. Adjacent clipped district SOI coverages were then intersected,
creating an area shared by both districts - our new area of concern.
The circuit lines in this area of intersection were clipped out
and points created just as before. These points were then Thiessened,
and the resulting polygons again dissolved by their substation
number. This created a new set of Spheres of Influence for just
the area of intersection. This new set and the clipped district
SOI coverages were then appended, and again dissolved by their
substation number erasing any artificial boundaries inserted during
the clipping process, and Spheres of Influence existed for SCE's
entire service territory. While there are a few drawbacks to
this method (see Limitations section below, and Burrough, 1986),
in the context of our data limitations, it was the best fit.
It is of interest to note
that in the beginning a much simpler approach for creating the
district SOI coverages was attempted. Instead of creating Thiessen
polygons for each and every circuit point, we created polygons
using only the substation points themselves. The hope was that
given the resolution of the ITS model is the census tract, it
would be enough to find the census tracts that fell within each
of those polygons. After testing, this was rejected because the
edge of a substation's Sphere of influence can rarely, if ever,
be approximated as the midpoint between itself and the next nearest
neighbor.
By 'Thiessening' the circuit
points, accuracy was greatly improved, as the distances between
neighboring circuits is vastly smaller than the distances between
substations. In other words, this method accounts for the following
scenario: Circuits from substation A spread out to the north
and west only, such that substation A falls in the very southeast
corner of its own Sphere. Southeast of substation A is an area
whose circuits 'belong' to the neighboring substation, B. 'Thiessening'
only the substation points would result in the incorrect assignment
of this area to substation A. This would not be the case by utilizing
the final methodology described above.
Limitations
This approach may not always
be a practical option where scale becomes an issue. Southern
California Edison serves more than 4.3 million customers, via
4000 circuits that run throughout the 50,000 square mile service
territory. Once the circuits were aggregated and converted to
points, the size of the data was much too great to consider as
a whole. By way of example, the resulting point coverage of the
SCE service territory contained over 40 million points requiring
4.5 gigabytes of disk space. The processing overhead required
to create Thiessen polygons proved to be too great.
Disaggregating the data
meant that at some point it would have to be appended back together.
This too presented problems. If for example, we were to break
up and Thiessen the circuit data by county, those circuits near
county borders would look for the next nearest circuit, but in
some cases would find the edge of the county instead, with it
then forming the edge of the SOI. More than likely, SOI do not
follow county boundaries. To combat this, the circuit data was
broken up by SCE district with the understanding that circuit
lines tend not to serve more than one district.
While this is generally
the case, circuits lines have been observed crossing district
boundaries. In these rare cases, one side of the Thiessen polygon
would continue on to the limits of the current extent as the Thiessen
process "searched" for the next nearest circuit point.
There are multiple methods for accurately approximating where
to cut off these endless polygons. After weighing each possibility,
the circuit lines were buffered, and the portion of this buffer
that extended past the district boundary was appended to the existing
"open-ended" SOI, providing an approximation of where
the SOI actually ends. This too posed problems as there are a
few cases where the buffers of two separate circuits overlap,
presenting the impossible scenario where one area is assigned
to two SOI. These areas were further analyzed by again "Thiessening"
the shared circuit points, dissolving together those Thiessen
polygons with the same substation id, and then appending the result
to the main SOI. Fortunately, these subsequent processes had
to be performed on only a very small minority of the circuits.
Conclusions
Limitations notwithstanding, the creation of substation SOI have been of great benefit to SCE. The delineation of the areas served by each substation are of great help in the electric distribution planning arena, and make the translation of EV purchasing forecast data into real world electric demand possible. As the number of 'plugged-in' EVs increases, so too will the rewards of this effort. As a result, any concerns SCE might have had about predicting the impact of future electric demand on the distribution infrastructure have all but been erased.
References
Southern California Edison
(1996) Electric Transpiration. Southern California Edison, Electric
transportation Department.
Morales, Ernest (1995) Geographic
Information System (GIS) Applications for Electric Vehicle Demand
modeling. Paper Presented at the 7th National Demand-Side Management
Conference. June 28, 1995, Dallas, Texas.
Schirmer, David E. (1996)
On the Integration of GIS Within an Electric Vehicle Program for
Predictive Analysis. Paper Presented at the 1996 Esri Users Conference.
May, 1996, Palm Springs, California.
Esri (1995) ARCDOC - "THIESSEN".
Environmental Systems Research Inc. Redlands, California.
Burrough, P.A. (1986) Principles
of Geographical Information Systems for Land Resources Assessment.
Oxford University Press. New York.
Authors
Kevan Q. Newton
GIS Analyst
Southern California Edison
2131 Walnut Grove Ave.
Rosemead, CA 91770
(818) 302-7545
email: newtonkq@sce.com
David E. Schirmer
Site Coordinator
Southern California Edison
2131 Walnut Grove Ave.
Rosemead, CA 91770