Overview
General Electric Company (GE) is an American multinational conglomerate operating mostly in the power, renewable energy, aviation, and healthcare industries. In 2020, GE ranks among the Fortune 500 as the 33rd largest firm in the United States by gross revenue.
To develop innovative technologies for GE’s business interests, GE Global Research (GEGR) was established, and it has become one of the world’s largest and most diverse industrial research labs. Improving business through technology, GEGR has developed expertise in simulation optimization and operations research.
Problem
When GE announced its commitment to electric vehicles (EV), it led to the need for advanced work in various related areas. Much of the technology was still developing and only just becoming widely available commercially, so many issues needed to be solved.
GE deployed a large fleet of electric vehicles for personal use, produced WattStation electric vehicle charging stations, and was conducting many other activities related to EVs, but a key question the company wanted to answer was: How would these markets evolve?
Relatively widespread use of EVs and the need to support them with services and charging network, prompted GEGR to conduct research into the related business demand and new technologies. Furthermore, any changes to electric power distribution patterns and methods were also of interest to GEGR.
In short, by conducting an EV transportation and charging network analysis, GEGR team sought to find answers to the following questions:
- What does the EV adoption curve look like?
- What will be the impact on the electricity distribution network?
- How will the charging infrastructure evolve?
- What new business models will emerge?
GEGR decided to test if advanced simulation modeling could help answer these questions. To this end, the research team was tasked with building demonstration prototypes:
- To evaluate modeling techniques as a tool to help gain insight, forecast, and make decisions in emerging business areas.
- To identify potential methods and approaches, as well as their integration possibilities that would help support the research.
Solution
For this EV transportation and charging network analysis project, GE Global Research chose AnyLogic multimethod simulation software because it supports the agent-based modeling method. Agent-based modeling allowed the engineers to describe the EV market as a system of individual agents that make their own choices. For example, consumers could individually decide whether to buy an EV or a conventional internal combustion engine (ICE) car.
With agent-based simulation, the team could also model adaptive driver behavior, such as when a driver takes an extended journey (as opposed to a regular home-work-home route) and needs to decide when and where to charge. Additionally, potential EV buyers might have very different priorities due to variables such as income, commute distance, personal preferences, and so on.
For the project, the team developed two prototype models: a granular EV adoption model and a charging network simulation. The agent-based models made use of AnyLogic’s Java platform to embed various rule-based functions and variables, and the software’s convenient graphical visualization capabilities.
The granular EV adoption model
As there was a lack of historical data on the potential-customer decision-making process, the GEGR team decided to simulate the process and used the Example-Based Evidential Reasoning (EBER) approach. This allowed the forecasting of how multiple factors could influence people’s decisions. The approach was developed by GE and has been used in several projects related to risk management and competitive pricing.
For the Granular EV Adoption model, the preference factors were selected by the team themselves. They included:
- Vehicle utility (reputation, range, battery life, etc.)
- Consumer attributes (commute distance, income, home charger availability, etc.)
- Finance (payback time, annual costs, etc.)
- Location (climate, government incentives, infrastructure, etc.)
The model’s inputs included outputs from other models (for example, financial simulations that calculated the payback and operational costs for an EV) and data from various open-source databases.
In the EV transportation network model, all factors and preferences, such as those related to an individual's finances and location, word-of-mouth, and vehicle availability, were organized in a way that corresponded to one potential buyer. Additionally, it tracked the transition of an agent from being a potential buyer to a user of either an EV or ICE vehicle. The output was the relative preference of a potential consumer for buying an EV over a conventional ICE vehicle.
The adoption rate was defined by tracking the number of EV and ICE vehicle drivers to give a picture of the total population. For a given geographical area, New York State, the model summarized adoption rates over time. The results could be displayed on a map by zip code region, or on charts.
The charging network simulation
The simulation was created to test the impact of various charging network designs on both the satisfaction rate of EV drivers and the utilization of charging points. The team also wanted to find answers to the following questions:
- What is the ROI for implementing a certain charging network design?
- Where are potential locations for charging points?
- How many charging points are needed?
Using the data on EV adoption rate from the previous simulation model, the team built a prototype simulation of EV usage in eleven New York State zip code regions with the help of GIS mapping. Using agent-based modeling, they created a custom library of objects which they could further use in other projects. For each object, the team could set properties and behavior so that, for example, households could be owned or rented and have different numbers of vehicles and drivers.
When run, the simulation showed the behavior of drivers, their movements, and decisions, based on logic rules which had been set in the model. It was possible to observe how the status of each object changed over time on a map or via an object’s statechart. By adding or removing charging points, they could alter the environment and monitor how it influenced potential EV owner satisfaction metrics.
Results
The GE Electric Vehicle transportation and charging network analysis project showed that simulation is a powerful tool for forecasting and planning in newly emerging business areas. The models developed by the research team were useful for:
- EV charging station manufacturers to understand demand.
- Store owners to simulate how installing EV chargers could influence their business revenue.
- Charging network operators to decide where to place chargers to maximize their utilization.
- City planners to decide on charger network design that would maximize returns on investments and the EV adoption rate.
AnyLogic software allowed the GE research team to build sophisticated models with numerous entities, including agents with decision-making abilities. The models made use of AnyLogic’s multimethod modeling possibilities and built-in graphical visualization capabilities.