PhD Proposal Defense:
Electric Vehicle Smart Charging Adoption,
Grid Peak-Shaving Quantification, and
The surveydown Survey Platform

Pingfan Hu
George Washington University

EV sales in US reaching ~10% of sales

Source: Argonne National Lab, www.anl.gov/ev-facts/model-sales

Introduction

• Unmanaged BEV charging is becoming a problem to the grid.
• Managed charging is cheaper and smoothes out the grid load.
• Smart charging: Supplier-Managed Charging (SMC) and Vehicle-to-Grid (V2G).

Unmanaged BEV charging is becoming a problem to the grid due to the spike. That’s why we need managed charging to smoothes out the grid load as well as keep the price low.

Smart charging is a type of managed charging, which includes Supplier-Managed Charging, shortened as SMC, and Vehicle-to-Grid, shortened as V2G.

SMC - Supplier Managed Charging

• SMC smooths out EV charging spike (called “Peak-shaving”).
• Electricity demand is controlled below capacity threshold.
• It saves money and reduces pollution.

For SMC, Supplier-Managed Charging, it smooths out the overnight EV charging demand.

As you can see, with unmanaged charging, the BEV load easily passes over the capacity threshold.

SMC - Supplier Managed Charging

• SMC smooths out EV charging spike (called “Peak-shaving”).
• Electricity demand is controlled below capacity threshold.
• It saves money and reduces pollution.

But with SMC, the electricity demand is controlled below the threshold. It reduces the peak demand and saves money.

V2G - Vehicle-to-Grid

V2G is short for Vehicle-to-Grid. It is another type of smart charging that allows BEV owners to sell electricity back to the grid. With V2G, BEVs become potential sources of electricity.

Smart charging effectiveness depends on
enrollment and peak-shaving.

Three Inter-connected Studies

Survey Research & Modeling

Electric Vehicle Smart Charging Adoption

Simulation Research

Grid Peak-Shaving Quantification

Software Development

The surveydown Survey Platform

Study 1
Electric Vehicle Smart Charging Adoption

Under second round of reviews:

Hu, Pingfan, Tarroja, B., Dean, M., Forrest, K., Hittinger, E., Jenn, A. & Helveston, J.P. (2024) “Measuring Electric Vehicle Owners’ Willingness to Participate in BEV Smart Charging Programs” Environmental Research Letters.

Prior studies have few EV owners in sample

1. A survey-based study by Wong et al. (2023) examined incentives effect on EV owners’ acceptance, but EV ownership in sample was only 19%.
2. A survey-based study by Philip and Whitehead (2024) found range anxiety matters, but EV ownership in sample was only 1.28%.
3. A survey-based study by Huang et al. (2021) indicates the importance of fast charging, but the sample size was only 157.

We need high EV ownership & large sample size.

There’s been a few studies that looked on people’s willingness to participate to smart charging programs, but they all have small sample size and low EV ownership. Therefore, we aimed to conduct a large-scale survey with high EV ownership to study BEV owners’ preferences.

Research Questions

1. Sensitivity: How do changes in smart charging program features influence BEV owners’ willingness to opt in?

2. Enrollment Rate: Under what combinations of features will BEV owners be more willing to opt in to smart charging programs?

Conjoint survey to collect BEV owners’ willingness.

Mixed logit model for utility simulations.

Here we have 2 research questions. Firstly, we want to know the sensitivity of each smart charging program feature; secondly, we want to know under what conditions will BEV owners be more willing to opt in to the programs.

In order to answer these questions, we designed a conjoint survey to collect BEV owners’ willingness to participate. Then we used mixed logit models to simulate the utility of each program feature.

Survey Design

Conjoint Questions

1. Monetary Incentives
2. Charging Limitations
3. Flexibility

Demographic Questions

1. BEV Ownership
2. Personal Info
3. Household Info

Conjoint Question Explained

A Sample Conjoint Question

1. Provide respondents with different sets of attributes.
2. Observe choices across random sets.
3. Estimate utility of each attribute.

But first, I want to explain what conjoint question is. If I care about apples, I would want to know how people value type vs price vs freshness. In this apple example, you can only choose one set of the attributes, so if you pick one option, you pick all the attributes in it. In short, conjoint is about forcing people to make trade-offs.

SMC Programs

Attributes

No.	Attributes	Range
1	Enrollment Cash	$50 to $300
2	Monthly Cash	$2 to $20
3	Monthly Override	0 to 5
4	Min Battery	20% to 40%
5	Guaranteed Battery	60% to 80%

Sample Program

Attributes	Values
Enrollment Cash	$300
Monthly Cash	$20
Monthly Override	5

(Range determined by stated vehicle they own)

Let’s move on to the conjoint question for SMC programs. Again SMC is short for Supplier-Managed Charging. These are the 5 attributes of SMC. You can see there are cash, override, and battery thresholds, and they all have reasonable ranges. For example, for enrollment cash we have $50 to $300.

Here is a sample SMC program. You can see the one-time enrollment cash, recurring monthly cash, and how many times per month the charging can be shifted back to normal.

The battery thresholds here are well-designed to provide an immediate visual aid of the minimum battery and guaranteed battery. We also used milage instead of percentage to make it more intuitive. The mileage will be shown according to the real values of each participant.

V2G Programs

Attributes

No.	Attributes	Range
1	Enrollment Cash	$50 to $300
2	Occurrence Cash	$2 to $20
3	Monthly Occurrence	1 to 4
4	Lower Bound	20% to 40%
5	Guaranteed Battery	60% to 80%

Sample Program

Attributes	Values
Enrollment Cash	$300
Occurrence Cash	$20
Monthly Occurrence	1

(Range determined by stated vehicle they own)

This is the conjoint question design for V2G programs. V2G is short for Vehicle-to-Grid. The attributes are similar to SMC, but with some differences. For example, we have monthly occurrence instead of monthly override.

The battery thresholds are in the same layout, but since V2G drains the battery, the lower bound is how much the battery is drained to, and guaranteed battery is how much it’s charged back to.

Sample SMC Question

There are 6 SMC conjoint questions. In each question, participants are provided with 2 valid options and a “Not Interested” option. It’s important to have this “Not Interested” option, because it is a validation of willingness. When participants choose from these 2 valid options, they are surely to be interested in what they choose. Otherwise, they will choose “Not Interested”.

Sample V2G Question

Likewise, there are 6 V2G conjoint questions.

Survey Fielding - 1,356 in Total

Meta Ads: Voluntary participants

• 803 responses
• March to July in 2024

Dynata Recruitment: Paid survey

• 553 responses
• September to November in 2024

We collected in total of 1,356 complete, filtered responses from Meta and Dynata throughout 2024, and we made sure that all participants are real BEV owners.

Survey Question - Car Ownership

Survey Results - Top 10 BEV

This is a showcase of the top 10 BEV models from our survey results. You can see that Tesla is the most popular brand, followed by Ford, Chevrolet, and some other popular brands. These results are also representative of the BEV market in the United States.

Survey Results - Demographics

Other than the BEV models, there are also some important demographics information collected in the survey. You can see most families have 2 cars; most of owners charge at home for more than 8 hours; and most of them care of climate very much. And more importantly, we have 100% BEV ownership.

Survey Results - Willingness to Participate

Mixed Logit Models

\[ \small \begin{align*} u_j = v_j + \epsilon_j = \beta' x + \epsilon_j \qquad P_j = \frac{e^{v_j}}{\sum_{k=1}^{J} e^{v_k}} \end{align*} \]

Utility esimated using maximum likelihood estimation (MLE).

SMC Estimates

V2G Estimates

Without compensation, users will not participate.

The willingness to participate is quantified with the mixed logit model, shortened as MNL. The utility of each option is a linear combination of the attributes, which is the \(v_j\) here, plus an error term, shown as \(\epsilon_j\). The probability of choosing an option is the exponential of its utility divided by the sum of exponentials of all utilities.

Here are the coefficient estimate plots of both programs. As you can see, the “No Choice” has highgest coefficient, which simply means that without compensation, users will not participate in the programs.

However, coefficients alone are not enough to interpret the willingness to participate. A more intuitive approach is the sensitivity, which means by how much the enrollment rate will change when a feature value changes.

Enrollment Sensitivity

Baseline Simulation

Choice between “None” and this program:

Attributes	Values
Enrollment Cash	$0
Monthly Cash	$0 - $20
Monthly Override	0

Sensitivity Plot

This is one example of the sensitivity plots. As the enrollment cash increases, the user enrollment increases as well. The solid part is from the results, and the dotted part is the extrapolation with the model.

Enrollment Sensitivity

1. Steeper slope indicates higher sensitivity.
2. Diminishing returns exist.

And here we have a full demonstration of all sensitivity plots. Steeper slope indicates higher sensitivity. Also, as the feature values become higher, we see diminishing returns. But the solid parts are basically linear.

Equivalencies of 5% Enrollment Increase

SMC

Attribute	Equivalence Value	Unit
Enrollment Cash	77.7	$
Monthly Cash	4.0	$
Override Days	2.5	Days
Minimum Threshold	65.5	%
Guaranteed Threshold	6.3	%

V2G

Attribute	Equivalence Value	Unit
Enrollment Cash	55.7	$
Occurrence Cash	2.9	$
Monthly Occurrence	1.9	Times
Lower Bound	11.7	%
Guaranteed Threshold	9.1	%

1. Smaller value indicates higher efficiency.
2. Monetary incentives are valued more in V2G than SMC.
3. Guaranteed thresholds are in high efficiency, indicating range anxiety.
4. Attribute equivalencies can be used to inform incentive design.

Study 2

Quantifying the supply of peak-shaving from controlled EV charging

Recall the SMC Peak-shaving

• SMC smooths out EV charging spike (called “Peak-shaving”).
• Electricity demand is controlled below capacity threshold.
• It saves money and reduces pollution.

Research Questions

1. Peak-shaving: How much does peak-shaving from controlled EV charging cost utilities?

2. Variations: How do peak-shaving results change based on regions, seasons, EV-to-house ratios, and enrollment rates.

Data collection from NHTS, NREL, and U.S. Census.

Peak-shaving simulation using model from study 1.

Plan of Action

Step 1: Data Collection

• NHTS: Charging behavior data
• NREL: Household electricity consumption data
• U.S. Census: Count of single family households
• Study 1: MXL model of SMC enrollment utilities

Step 2: Simulation

• Simulate grid demand profiles with different EV charging loads
• Simulate peak load reduction under different levels of SMC enrollment

Step 3: Sensitivity Analysis

• Compare the peak-shaving results of different scenarios
• Summarize and provide policy implications

NHTS - National Household Travel Survey

Travel patterns of 26,000 vehicle owners.

Simulated EV Charging Profile

Energy consumption and concurrent charging % of 10,000 EVs.

NREL - National Renewable Energy Lab

Data of household electricity consumption in each GEA region.

18 GEA Regions

CAISO as an Example

7.69M single-family homes, collected from U.S. Census.

Peak-shaving Simulation

Original Consumption Profile

Assumptions

• 200-mile range BEV.
• Level 2 charger.
• Peak: 10AM-12:00AM.
• Valley: 12:00AM-7AM.
• Battery: 20%-80%.
• Auto-override.

Shifted Consumption with Peak-shaving

90% SMC enrollment, EV-to-house ratio 1:1.

Assumptions

• 200-mile range BEV.
• Level 2 charger.
• Peak: 10AM-12:00AM.
• Valley: 12:00AM-7AM.
• Battery: 20%-80%.
• Auto-override.

MXL Model of Monthly Incentives

EV Peak Increase vs Monthly Incentives

More incentives lead to higher SMC enrollment, thus more peak shaving.

Sensitivity Analysis

Different Scenarios

• All 18 GEA regions in all 4 seasons.
• Varying EV-to-house ratios (1:4, 1:2, 1:1).
• Varying SMC enrollment rates (from 0% to 100%).
• Varying peak and valley windows.

Policy Implications

1. Utilities must plan a budget that scales with enrollment levels.
2. Governments could get engaged (e.g. provide subsidies to offset the cost).

Study 3
The surveydown Survey Platform

Published in PLoS ONE:

Hu, Pingfan, Bunea, Bogdan, & Helveston, J. P. (2025). “surveydown: An open-source, markdown-based platform for programmable and reproducible surveys” PLOS ONE, 20(8), e0331002. https://doi.org/10.1371/journal.pone.0331002

Typical Web Survey

Google Form - GUI Interface

Limitations

❌ Reproduciblity

❌ Version control

❌ Limited features

❌ Open source

Qualtrics - Commercial Platform

Expensive!

Why not make surveys from code?

Limitations
Features

✅ Reproducibility

✅ Version control

✅ Lots of features

✅ Open source

✅ Free!

Introducing surveydown!

R package that renders Quarto files into surveys

Quarto is a publishing system

Original qmd file

Markdown + R code chunks

---
format: html
title: "HTML Page with R Code"
---

# Hello, World!

This is a simple **HTML** page with *R* code.

```{r}
library(ggplot2)

df <- data.frame(x = rnorm(100))
ggplot(df, aes(x = x)) +
  geom_histogram()
```

Rendered HTML

Original qmd file

Markdown + Python code chunks

---
format: pdf
title: "PDF File with Python Code"
---

# Hello, World!

This is a simple **PDF** file with *Python* code.

```{python}
import matplotlib.pyplot as plt
import numpy as np

data = np.random.normal(0, 1, 100)
plt.hist(data)
plt.show()
```

Rendered PDF

survey.qmd

---
format: html
echo: false
warning: false
---

```{r}
library(surveydown)
```

::: {#welcome .sd-page}

# Welcome to `surveydown`!

```{r}
sd_question(
  type  = "mc",
  id    = "has_fav_hero",
  label = "Do you have a favorite super hero?",
  option = c(
    "Yes" = "yes",
    "No"  = "no"
  )
)

sd_next()
```

:::

Rendered Survey

survey.qmd

---
format: html
echo: false
warning: false
---

```{r}
library(surveydown)
```

::: {#welcome .sd-page}

# Welcome to `surveydown`!

```{r}
sd_question(
  type  = "mc",
  id    = "has_fav_hero",
  label = "Do you have a favorite super hero?",
  option = c(
    "Yes" = "yes",
    "No"  = "no"
  )
)

sd_next()
```

:::

YAML header for a “clean” output

survey.qmd

---
format: html
echo: false
warning: false
---

```{r}
library(surveydown)
```

::: {#welcome .sd-page}

# Welcome to `surveydown`!

```{r}
sd_question(
  type  = "mc",
  id    = "has_fav_hero",
  label = "Do you have a favorite super hero?",
  option = c(
    "Yes" = "yes",
    "No"  = "no"
  )
)

sd_next()
```

:::

Load the surveydown Package

survey.qmd

---
format: html
echo: false
warning: false
---

```{r}
library(surveydown)
```

::: {#welcome .sd-page}

# Welcome to `surveydown`!

```{r}
sd_question(
  type  = "mc",
  id    = "has_fav_hero",
  label = "Do you have a favorite super hero?",
  option = c(
    "Yes" = "yes",
    "No"  = "no"
  )
)

sd_next()
```

:::

Use Quarto fences (:::)

to define survey pages

survey.qmd

---
format: html
echo: false
warning: false
---

```{r}
library(surveydown)
```

::: {#welcome .sd-page}

# Welcome to `surveydown`!

```{r}
sd_question(
  type  = "mc",
  id    = "has_fav_hero",
  label = "Do you have a favorite super hero?",
  option = c(
    "Yes" = "yes",
    "No"  = "no"
  )
)

sd_next()
```

:::

Page content

• Markdown for texts, images, etc.
• sd_question() for survey questions
• sd_next() for page navigation

survey.qmd

---
format: html
echo: false
warning: false
---

```{r}
library(surveydown)
```

::: {#welcome .sd-page}

# Welcome to `surveydown`!

```{r}
sd_question(
  type  = "mc",
  id    = "has_fav_hero",
  label = "Do you have a favorite super hero?",
  option = c(
    "Yes" = "yes",
    "No"  = "no"
  )
)

sd_next()
```

:::

Rendered Survey

survey.qmd

---
format: html
echo: false
warning: false
---

```{r}
library(surveydown)
```

::: {#welcome .sd-page}

# Welcome to `surveydown`!

```{r}
sd_question(
  type  = "mc",
  id    = "has_fav_hero",
  label = "Do you have a favorite super hero?",
  option = c(
    "Yes" = "yes",
    "No"  = "no"
  )
)

sd_next()
```

:::

Rendered Survey

Quarto renders to static html pages.
Shiny turns them interactive.

A complete surveydown survey

survey.qmd

A Quarto doc defining the survey content (pages, texts, images, questions, etc).

app.R

An R script defining the survey Shiny app.

PostgreSQL for response data storage.

supabase.com

Technologies of surveydown

surveydown is feature-packed!

surveydown is highly flexible and customizable to user needs

Randomization: Conjoint question

Shiny compatibility: leaflet map

surveydown vs other platforms

Star Count on GitHub

surveydown.org

Research Contributions

• Study 1: First large N survey study of EV owner preference sensitivities to smart charging. Introduces attribute equivalencies for incentive design.
• Study 2: Quantifies grid supply curves for peak-shaving, considering regions, seasons, EV-to-house ratios, and enrollment rates.
• Study 3: surveydown, a free open-source survey platform for programmable, reproducible survey designs.

I start with a survey research on smart charging, followed by a nation-wide grid simulation, and a survey software.

I hope these studies could benefit sustainable transportation, and give persistent contributions for future research.

Timeline of Study 2

• Nov 2025: Sensitivity analysis on various scenarios.
• Dec 2025: Complete & submit manuscript for peer review.
• Mar 2026: Defend dissertation.

Thanks for listening!

Appendix

SMC Scenario Analysis

1. Flexibility is highly valued.
2. Recurring incentives are more important than one-time.
3. Payment alone is not enough.

So, after analysis of all these individual features, we can combine them into programs. We call them “scenarios”. I want to emphasize that the purpose of doing these evaluations is to wisely allocate the resources of the grid. Therefore, we need to maximize the enrollment of each program with least efforts.

Here for SMC we have 3 scenarios. They are flexibility, recurring cash, and one-time cash designated.

As we can see, for SMC, flexibility is highly valued, and recurring incentives are more important than one-time. So due to the range anxiety, payment alone is not enough. There should be a combination of flexibility and monetary incentives.

V2G Scenario Analysis

1. Still, recurring incentives are more important than one-time.
2. But flexibility is not as important compared with SMC.
3. Owners are willing to leverage BEV as a source of income.

For V2G programs, recurring incentives are also more important than one-time. But flexibility is compromised since owners are willing to use their BEVs as a source of income. So the V2G programs are more incentive powered than SMC programs.

Smart Charging Enrollment Simulator

Here we’ve made a smart charging enrollment simulator. We can play around with it and see the anticipated enrollment of each program, with different feature values.

1. More enrollment increase in SMC: Guaranteed threshold from 60% to 100%.
2. More enrollment increase in V2G: Recurring cash from $2 to $20.

SMC Logit Model

\[ \begin{align*} u_j = \beta_1 x_j^{\text{enroll_cash}} + \beta_2 x_j^{\text{monthly_cash}} + \beta_3 \delta_j^{\text{override_allowed}} + \beta_4 x_j^{\text{num_overrides}} \notag \\ + \beta_5 x_j^{\text{min_threshold}} + \beta_6 x_j^{\text{guaranteed_threshold}} + \beta_7 \delta_j^{\text{no_choice}} + \epsilon_j \end{align*} \]

Attribute	Coef.	Est.	SE	Level	Unit
Enrollment Cash	β₁	0.0037	0.0002	50, 100, 200, 300	USD
Monthly Cash	β₂	0.0728	0.0031	2, 5, 10, 15, 20	USD
Override Days	β₃	0.1191	0.0140	0, 1, 3, 5	Days
Override Flag	β₄	0.4357	0.0654	Yes, No	-
Minimum Threshold	β₅	0.0044	0.0023	20, 30, 40	%
Guaranteed Threshold	β₆	0.0490	0.0028	60, 70, 80	%
No Choice	β₇	2.8984	0.2215	-	-

V2G Logit Model

\[ \begin{align*} u_j = \beta_1 x_j^{\text{enroll_cash}} + \beta_2 x_j^{\text{occur_cash}} + \beta_3 x_j^{\text{num_occurrences}} + \beta_4 x_j^{\text{lower_threshold}} \notag \\ + \beta_5 x_j^{\text{guaranteed_threshold}} + \beta_6 \delta_j^{\text{no_choice}} + \epsilon_j \end{align*} \]

Attribute	Coef.	Est.	SE	Level	Unit
Enrollment Cash	β₁	0.0051	0.0003	50, 100, 200, 300	USD
Occurrence Cash	β₂	0.0972	0.0045	2, 5, 10, 15, 20	USD
Monthly Occurrence	β₃	0.1595	0.0262	1, 2, 3, 4	Times
Lower Threshold	β₄	0.0263	0.0036	20, 30, 40	%
Guaranteed Threshold	β₅	0.0368	0.0037	60, 70, 80	%
No Choice	β₆	2.4283	0.1964	-	-

surveydown Feature Highlights

Question types

Conditional logic

12 Built-in Question Types

text

textarea

numeric

mc

mc_multiple

mc_buttons

mc_multiple_buttons

select

slider

slider_numeric

date

daterange

Question Type: text

sd_question(
  type  = "text",
  id    = "fav_hero_name",
  label = "Who is your favorite super hero?"
)

–>

Question Type: mc_buttons

sd_question(
  type  = "mc_buttons",
  id    = "dream_power",
  label = "If you could have ONE superpower?",
  option = c(
    "🕸️ Web-slinging"   = "webslinging",
    "🛡️ Super Strength" = "strength",
    "✈️ Flight"         = "flight",
    "🧠 Telepathy"      = "telepathy",
    "⚡️ Super Speed"    = "speed"
  ),
  direction = "vertical"
)

–>

Conditional logic

Conditional showing

Conditional skipping

Conditional stopping

Conditional showing - sd_show_if()

Conditional showing - sd_show_if()

survey.qmd

# Conditional Question
sd_question(
  type  = "mc",
  id    = "has_fav_hero",
  label = "Do you have a favorite hero?",
  option = c("Yes" = "yes", "No" = "no")
)

# Target Question
sd_question(
  type  = "text",
  id    = "fav_hero",
  label = "Who is your favorite super hero?"
)

app.R

# Inside server
sd_show_if(
  input$has_fav_hero == "yes" ~ "fav_hero"
)

Condition ~ Target

“If {condition}, then show {target}”

Generate surveys with LLMs!

The companion sdstudio package

The companion sdstudio package

The companion sdstudio package

Reference List

Huang, Bing, Aart Gerard Meijssen, Jan Anne Annema, and Zofia Lukszo. 2021. “Are Electric Vehicle Drivers Willing to Participate in Vehicle-to-Grid Contracts? A Context-Dependent Stated Choice Experiment.” Energy Policy 156 (September): 112410. https://doi.org/10.1016/j.enpol.2021.112410.

Philip, Thara, and Jake Whitehead. 2024. Consumer Preferences Towards Electric Vehicle Smart Charging Program Attributes: A Stated Preference Study. Rochester, NY. https://doi.org/10.2139/ssrn.4812923.

Wong, Stephen D., Susan A. Shaheen, Elliot Martin, and Robert Uyeki. 2023. “Do Incentives Make a Difference? Understanding Smart Charging Program Adoption for Electric Vehicles.” Transportation Research Part C: Emerging Technologies 151 (June): 104123. https://doi.org/10.1016/j.trc.2023.104123.