David Schirmer, Ernest T. Morales, Habibur Boruah

On The Integration of GIS within an Electric Vehicle Program for Predictive Analysis

Abstract

As regulatory agencies seek to improve air quality through legislation requiring zero-emission vehicles, Southern California Edison (SCE) is aggressively preparing for the eventuality of increased numbers of electric vehicles (EV's). Introduction and widespread use if Ev's throughout SCE's service territory could potentially produce significant impacts on the electrical distribution infrastructure. SCE is currently developing methods and technologies that will enable Edison to anticipate and proactively design the distribution infrastructure to support the expected introduction of EV's within the service territory. Chief among the tools being developed is a geographic information system (GIS) that allows for the incorporation of a comprehensive EV purchase behavior model, electric load forecasting methodology, and an EV charging site optimization model. Data yielded from these models may then be further integrated with electric utility infrastructure data, and other cartographic databases to create a powerful locational decision support system that will enable infrastructure planners to set sound and defensible policy with respect to generation planning, distribution engineering and planning, as well as small area or service planning (Morales, 1995).

Introduction

In 1990, The California Air Resources Board (CARB) adopted rules requiring that by 1998 two percent of all new vehicles available for sale in California by the largest automakers must produce no tailpipe emissions. This figure will rise to ten percent by the year 2003. Currently, the only vehicles that satisfy these requirements are electric. And, while these rules have recently been relaxed somewhat -- taking on a more voluntary nature, it is clear that SCE's service territory will see dramatic increases in the numbers of electric vehicles.

In preparing for the eventuality of electric vehicles, many utilities have focused their attention on the generation, or 'on the supply side' of the equation. That is, does the existing system have enough generating capacity to support the expected numbers of EV's? Indeed, SCE estimates that the existing 'fuel' supply is more than adequate, able to charge approximately 600,000 EV's nightly in Southern California without the need for additional generating capacity, more than enough to satisfy the foreseeable demand (SCE, 1996).

EV Load

This, however, does not capture the full picture. While the generation and transmission infrastructure could support this new demand, wide adoption of EV's within the service territory could produce significant impacts on the localized distribution systems. By way of example, the use of even a single charger could potentially double a household's electrical demand. Using Tobler's Law of autospatial correlation, it is not unreasonable to assume that likely EV owners will be spatially clustered, thus putting a disproportionate burden on the localized substations and circuits.

In order to prepare the distribution infrastructure to serve this new load it is necessary to make predictions about where and when load will appear on the system. Edison's distribution infrastructure, like most other utility systems, has differing carrying capacities depending on which part of the system is being considered. Some areas of the system are more robust, and thus can handle additional load without system upgrading (e.g., commercial feeder systems are typically have more capacity than most residential circuits). Predicting the location and characteristics of the EV load and matching that information against discrete distribution elements is critical to the process of managing the load in a cost effective manner (Morales, 1995). However, electric vehicles present utilities with unique problems with respect to demand management that have not previously required consideration. Some considerations include:

* EV's represent an entirely new type of non-linear load that cannot be characterized by examining any previous comparable load on the system.

* EV's are a mobile rather than stationary source of demand.

* Because EV's are charged for relatively long time periods, EV's could place significant coincident demand on the system.

* EV-created load is unpredictable. It can appear on the system without advance notice and in multiple locations. Without proper planning and load management, substation and circuit rebuild costs could be substantial.

* Because of its random nature, knowledge of load behavior (EV purchasing and charging behavior in particular) is required in order to determine the location and consequences of load impacts.

* The impact of multiple EV chargers on the system may contribute to significant power quality problems not previously experienced by utilities on a large scale. Some studies have estimated that harmonic distortion levels could conceivably reach 90 percent at some locations given a large number of chargers connected to almost completely charged batteries during the night or early morning, This level of harmonic distortion could damage customer as well as utility distribution system components and seriously disrupt service (Rice, 1994).

Solutions

The Electric Transportation Division at SCE has devoted considerable resources to the development of datasets, applications, and models with the primary purpose of gaining insight into the patterns of potential EV owners, and predicting the impact of this behavior. Edison has developed a comprehensive EV purchase behavior model, including a load forecasting methodology, and an EV charging-site optimization model. Further, loacational data on EV ownership and usage patterns derived from a vehicle forecasting system may be spatially linked to utility infrastructure data, population growth statistics, land use data, demographics, and economic indicators. These datasets, models, and methodologies have been integrated within a single sophisticated GIS application, creating a powerful decision support system that will enable infrastructure planners to provide solutions to meet the challenges that increased numbers of EV's pose.

EV Forecasting

As discussed, in order to prepare the infrastructure to serve the anticipated electrical load, it is necessary to make predictions about where and when the load will be spatially located. To do this, SCE is utilizing the state-of-the-art California Alternative-Fuel Vehicle Demand Forecasting Model developed by the Institute of Transportation Studies (ITS) at the University of California at Irvine and Davis. The principal objective of the micro simulation model is to forecast market penetration of alternative fuel vehicles, with respect to numbers of vehicles, miles traveled, temporal recharge demand, concentrations of likely buyers, etc.. 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.

Forecasting EV penetration relies on predicting consumer behavior in terms of purchasing and usage. However, EV's differ from conventional vehicles with respect to monetary considerations, e.g., vehicle price, fuel costs, incentives, and repair costs, and other non-monetary considerations, such as availability, range, and vehicle performance. These considerations are accommodated within the model as described by the various inputs. These inputs include:

* Demographic variables (e.g., income, education, attitudes, etc.);

* Vehicle characteristics (e.g. body type, charger characteristics, recharge time, range, price, operative costs, speed, payload, performance);

* Fuel price/availability (i.e. gasoline, compressed natural gas, electricity, public charging, quick charging, workplace charging);

* Incentives (e.g. sales tax, registration fees, subsides, free parking, solo access to HOV lanes, and reduced household electricity rates).

Outputs from the model include:

* EV market penetration by type and total number of vehicles. This number may then be converted to kilowatt demand since the charger technology is known;

* Geographic location of likely owners by Edison service districts;

* Occurrence of EV load by time of day;

* Average vehicle miles traveled.

The EV forecasting model consists of a number of linked sub-modules determining household structure, income and employment status. These are then used to predict vehicle transactions, and determine annual vehicle mileage and refueling requirements. The model is continually being calibrated using information collected from personal vehicle surveys, vehicle trials programs, and the census of vehicle registrations (Morales, 1995).

The role of GIS

The primary role of the GIS application is to link the EV forecasting model with data on existing SCE facilities and distribution system infrastructure. However, the real power and potential of the application may be realized through its analytical capabilities in terms of electrical load forecasting, and public charging site optimization. This is in addition to the more general, but no less important, display, query, and output capabilities.

General description and functionality
The application is AML-based designed to run within the Arcplot, module of ArcInfo version 7.04. The entire application is driven by a single AML containing some 3000 lines of code. Using the Esri standard, the AML consists of approximately 150 separate routines that serve as a 'task library' being called by several dozen individual menus. The application was developed on a SUN SPARC 20 running Solaris 2.5. The workstation is equipped with 256 megabytes of memory with 13 gigabytes of disk space. Digital Thomas Brothers Maps data that include, among others, the street network, hydrography, cultural features, political boundaries serve as the principal land base. These data are 'tiled up' and stored using ArcInfo's Librarian module. Onto this base layer may be drawn various Edison-specific cartographic datasets including the service territory boundary, substations, transmission lines, electric meters, etc. Additionally, the application utilizes various other cartographic datasets including census blocks, census tracts, zipcodes, zip+4, etc.. These spatial data may also be linked with a variety of large corporate aspatial data tables held in a SYBASE environment.

It is envisioned that the principal users of the system will be distribution engineers, and planners who have little or no UNIX or GIS background. For this reason, simplicity, and ease of use were of paramount consideration when designing the application. To this end, the program utilizes a WIMP (windows, mouse, icons, and pop-ups) architecture that requires little or no special knowledge of computers. To avoid overwhelming the user, a single main menu remains on screen for the duration of the session. It is from this main menu that all functionality may be accessed. Broadly speaking, the application may be thought of as having three separate modules, each with a discrete area of functionality. These three modules include, an areal extent tool, a display management tool, and assorted general tools.

The areal extent tool simply allows the user to enter and browse the data at whatever resolution of interest. This is to say, that the user may, with a few mouse clicks, set the mapextent to a city, to a circuit's sphere of influence, to an individual meter, or to the area served by a particular substation. This may be done both logically (i.e., via a pre-programmed reselect), or graphically (i.e., by pointing to the object of interest and clicking). Also included in this module is a variety of pan and zoom tools, thereby providing the user with nearly unlimited assess to the data.

The display management module enables the user to control what data layers are drawn to the graphics window. Currently, the application allows for some fifty separate cartographic data sets to be drawn. This number, however, continually increases as the user requirements expand and change. Additionally, all of the draw symbols, whether point, line, or polygon, may be modified should the user deem the defaults to be inappropriate for the task at hand. One important feature of the display management module is that all of the active cartographic data layers remain resident when the user changes the mapextent, thus eliminating the need for the user to rebuild the display every time the area of interest changes.

It is here within the display management module where the results of the micro simulation model may be viewed. The user is prompted, via checkboxes and choice fields, to view the desired outputs from the model. For example, the user may wish to view the predicted penetration of mid-sized EV's for a given area for the year 2005. With the built-in graphing and 'animation' tools, the user may view how this market penetration is expected to change over a user-defined period time. Additionally, the application allows for the user to alter the parameters of the model as a means of running 'what if' scenarios. This is particularly important in light of rapid EV technology innovation. For example, as the range of the vehicles improves, and as the number of public charging stations increases, the users is able to capture what these changes could potentially mean in terms of EV penetration for a given area

The tools module represents a suite of wide-ranging ArcInfo functions automated and bundled into easy- to-use menu choices. The functionality ranges from the ability to convert ASCII address data into ArcInfo coverages, to allowing users to identify, query, display, and generate reports on all spatial and aspatial data sets included within the application.

One important feature contained within the tools module is the ability to produce high quality vector hard copy map output. As discussed, all active cartographic data layers remain resident in memory via variable management. This allows for output files to be generated with a single mouse click. . In addition to drawing the layers in the correct order (i.e., polygon features first, followed by lines and points), an appropriate legend is also scaled and drawn. The data that are drawn to the graphics screen are placed into a standard layout that draws border boxes, the scale bar, north arrow, logos, creation date, etc. The user may then input map-specific information including the map's title, the creator's name, the specific project the map was created for, etc. Other features include the ability to adjust the output size of the map (e.g., from A-size to E-size), the output format (e.g., a graphics file, a postscript file, an hpgl2 file, etc.), and the destination plotting device (e.g., an HP 750, an HP 1200, or a SPARC printer).

Load Forecasting
As discussed, the real power of the application will be in its ability to forecast the impact EV's will have on the distribution infrastructure. While the steps involved in this process are fairly straightforward, the data requirements, programming and processing have proven to be quite formidable. A brief synopsis of these steps is as follows.

On-going work being undertaken by the GIAS group at SCE is resulting in the geocodeing of all of SCE's 4.3 million electric meters. Through a series of ArcInfo relates and remote connections to various dynamic SYBASE tables, the geocoded meters are tied to transformers that are tied to circuits, that are ultimately tied to substations. The micro simulation model illustrates where potential EV owners are located, and gives the predicted numbers of vehicles by census tract. Given that the electrical demand of the charging technology is know, this can be directly translated into kilowatt demand upon the circuit, substation, or indeed, whatever level of aggregation that best meets the needs of the task at hand.

By way of example, if the model predicts that there will be ten EV's in a given census tract and that each EV will require an addition 10 kilowatts of power, we can query the meters for the circuits that terminate in that tract. If it is found that two circuits feed the census tract, we may then (by using the less-than-prefect assumption that there is a uniform distribution of vehicles) spread the load accordingly, resulting in an additional load of 50 kilowatts per circuit. Further queries determine that both circuits feed the same substation resulting in a total additional load of 100 kilowatts.

Public Charging Site Optimization
As utilization of the application increases users are continually providing input on how the applications might be improved through additional functionality outside the original scope of work. One such addition suggested by the users that is currently under development is the ability to optimize the siting of remote electric vehicle charging stations. The impetus for this work is the desire by Edison to minimize or eliminate the construction of unnecessary or under-utilized (stranded) electrical infrastructure, while at the same time, optimize the utilization of the EV charging infrastructure that is deployed.

One of the outputs from the micro simulation model is data describing the driving behavior of potential EV owners in terms of average miles traveled, time of travel, etc. To further verify and potentially calibrate the model, a means of chronicling the actual driving behavior of an EV owner has been suggested. To this end, SCE is undertaking a project that will deploy global positioning system units in a sample set of the vehicles participating various testing programs. These units would map (record latitude and longitude data) vehicle location throughout the duration of the trip. Additionally the drivers would input attribute data regarding the nature of the trip (e.g., for work, shopping, etc.). The collected data would then be down loaded into the application where the actual data would be compared with predicted data.

Further calibration of the micro simulation model could then be made, if necessary.

Armed with 'real world' empirical data, and with the outputs from the model, SCE should have a fairly clear notion of not only where the potential EV owners will be located, but also the nature of the trips undertaken. From here, utilizing the location/allocation built into ArcInfo, EV charging demand surfaces could be constructed within the application illustrating those areas where public investment for utility infrastructure may be warranted, and conversely, where it makes little sense.

Conclusion

The process of predicting patterns of EV purchasing and usage, and determining the impacts of EV penetration on the electrical infrastructure are necessarily complex. SCE's GIS application briefly described above established a framework that can support that process, by spatially organizing and consolidating the many layers of spatial and aspatial data essential to the development of intelligent predictions, and proactive solutions to meet the challenges that the introduction of electric vehicles poses. As the micro simulation model is refined, additional functionality and applications will be designed and implemented. We believe that the system is robust enough to meet the existing demand, while at the same time flexible enough to accommodate future needs.

References

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.

Rice, Richard (1994) Modeling Electric Vehicle Utility System Impact Utilizing a Geographic Information System. Southern California Edison, Electric Transportation Department.

Southern California Edison (1996) Electric Transpiration. Southern California Edsion, Electric Transportation Department.

David Schirmer
Geographic Information and Analysis Systems (GIAS)
2131 Walnut Grove Avenue, Room 228
Rosemead, California 91770

Ernest T. Morales
Electric Transportation Division
Southern California Edison
2244 Walnut Grove Avenue Suite 419
Rosemead, California 91770

Habibur Boruah
Electric Transportation Division
Southern California Edison
2244 Walnut Grove Avenue Suite 419
Rosemead, California 91770