← AcademiaOne PhD Hub · 返回首页
AcademiaOneEDUCATION
PhD Research Proposal Guide · HK · Science & Engineering 香港理工科研究计划书指南
AcademiaOneEDUCATION
Your PhD application, from idea to offer你的博士申请,从构想到录取
DOCTORAL APPLICATIONS · HONG KONG · SCIENCE & ENGINEERING   香港理工科博士申请

PhD Research
Proposal Guide · Science & Engineering

香港理工科博士研究计划书写作指南

A section-by-section model of a self-proposed science or engineering PhD research proposal for Hong Kong, with the supervisor-first email that comes before it. Each part shows the structure, a fully worked example from power-systems engineering, and the difference between a weak and a stronger version, with the current requirements of HK departments taken from their official pages.

逐部分拆解一份香港理工科自主课题博士研究计划书,并附写在计划书之前的联系导师邮件。每一部分都包含结构说明、一个完整的电力系统工程范例,以及“较弱”与“更好”写法的对比,并附香港各院系官网上的最新要求。

Start here 开始之前

What Hong Kong universities ask for香港各校要求

Hong Kong universities ask for a research proposal or a research plan in almost every field, but the form differs: HKU faculties set their own format, PolyU uses Form GSB/1A, HKUST takes a research plan typed into the online form, and CityU calls the document a research plan in several schools. The Hong Kong PhD Fellowship Scheme (HKPFS) and the main round at HKU and CUHK all close on 1 December 2026, the earliest deadline of the year, so plan backwards from that date.

香港各校几乎所有学科都要求提交研究计划书(有的称为“研究计划”),但形式各不相同:港大各学院自定格式,理大使用 GSB/1A 表格,港科大要求在网申表格中直接填写研究计划,城大多个学院称之为“研究计划”。香港博士研究生奖学金计划(HKPFS)与港大、港中文的主轮申请都在 2026 年 12 月 1 日截止,是一年中最早的截止日期,请以此倒推时间表。

Hong Kong science and engineering applications are mostly self-proposed and go in on a university form. HKU uses a faculty proposal form for Engineering and Science; CUHK Engineering asks for a proposal in English, and Biomedical Engineering for a research-interest statement of about 500 words; HKUST takes a research plan in the online form with no extra proposal; PolyU asks for Form GSB/1A plus a CV; and CityU for a research plan, with supervisor consent advised.

香港理工科申请多为自主选题,并通过学校表格提交。港大工程与理学使用学院计划书表格;港中文工程要求英文计划书,生物医学工程只要约 500 词的研究兴趣陈述;港科大在网申表格中填写研究计划,无需另交计划书;理大要求 GSB/1A 表格加简历;城大要求研究计划,并建议先取得导师同意。

Science & Engineering: what each programme asks for理工科:各项目要求

UNIVERSITY 大学WHAT THE PROGRAMME ASKS FOR 项目要求UNIVERSITY-WIDE RULE 全校规定OFFICIAL PAGE 官方页面
HKU
香港大学
QS 2027: 11
Faculty proposal form (Engineering, Science)
References: 2
Proposal or statement of research interest in faculty format
gradsch.hku.hk
gradsch.hku.hk
CUHK
香港中文大学
QS 2027: 18
Proposal in English (Engineering); BME: ~500-word research interest
Engineering pages differ (2 or 3)
References: 2-3
Not stated centrallyeng.cuhk.edu.hk
HKUST
香港科技大学
QS 2027: 33
Research plan in the online form; no extra proposal
References: 2+
Research topic and plan in form; extra proposal for Humanities and Social Science
fytgs.hkust.edu.hk
fytgs.hkust.edu.hk
PolyU
香港理工大学
QS 2027: 50
Form GSB/1A + CV (personal statement in most departments)
References: 2
Compulsory on Form GSB1a
polyu.edu.hk
polyu.edu.hk
CityU
香港城市大学
QS 2027: 52
Research plan; supervisor consent advised
Referees must hold PhDs (EE)
References: 2
Required for Liberal Arts & Social Sciences, Creative Media, Energy & Environment, Law
cityu.edu.hk
cityu.edu.hk

Requirements are quoted from the official page linked in each row, in English as published. The third column is the university-wide proposal rule from the graduate school page. 表中要求按各行所附官方页面的英文原文摘录;第三栏为研究生院页面上的全校计划书规定。

Golden rule 核心原则

The programme page always beats this guide. Use this guide to decide what to write; use the programme page to decide how long it is and what form it takes.

项目官网的要求永远优先于本指南。用本指南决定“写什么”,用项目页面决定“写多长、用什么格式”。

HKPFS comes first HKPFS 最先截止

The HKPFS has two steps. Get an HKPFS reference number on the Research Grants Council (RGC) system by 12:00 noon HKT on 1 December 2026, then apply to up to two programmes at one or two universities, quoting it, by 11:59 pm the same day. Universities pre-screen and nominate candidates, and HKPFS applicants are considered for each university’s own awards. Later rounds exist: CUHK clearing closes on 31 March 2027 and HKU Medicine on 30 April.

HKPFS 分两步:先在研究资助局(RGC)系统取得 HKPFS 编号,截止时间为 2026 年 12 月 1 日香港时间中午 12 点;再在当天晚上 11:59 前,凭该编号向一至两所大学的最多两个项目提交申请。各校先行筛选并提名候选人,HKPFS 申请人也会自动参与各校自设奖学金的评选。之后还有补录轮:港中文补录轮于 2027 年 3 月 31 日截止,港大医学院为 4 月 30 日。

Referees 推荐人

HKU, PolyU and CityU ask for two academic referees, HKUST for at least two (the form takes up to five), and CUHK for three for research doctorates. At PolyU your proposed supervisor cannot be one of your referees.

港大、理大和城大要求两位学术推荐人,港科大至少两位(表格最多可填五位),港中文研究型博士要求三位。理大规定拟定导师不能担任推荐人。

Supervisor and referees 导师与推荐人

CityU asks you to seek a supervisor’s consent before you apply, and in Electrical Engineering your referees must hold PhDs. At HKUST the research plan goes into the online form, so prepare a condensed version of your proposal.

城大要求申请前先取得导师同意,电机工程的推荐人须具有博士学位。港科大的研究计划直接填写在网申表格中,请准备一份压缩版计划书。

Table built from 5 official programme pages (each checked on the programme page itself; data checked 2026-11-18). Rules change each cycle; the programme page always wins. 表格依据 5 个官方项目页面整理(均在项目页面本身核对;数据核对日期 2026-11-18)。规则每年可能调整,以项目页面为准。

Start here 开始之前

Two routes into a STEM PhD理工科博士的两条申请路径

In science and engineering the funding usually comes with the project. A supervisor wins a grant, advertises a funded PhD on a defined problem, and recruits the applicant who fits it best. Elsewhere you design the project yourself and apply for a scholarship to fund it. Decide which route each target uses before you write anything.

理工科的资助通常跟着项目走:导师获得科研经费,发布一个课题明确、带资助的博士项目,并招收最匹配的申请人。在另一些地方,课题由你自己设计,再去申请奖学金。动笔之前,先弄清每个目标院校走哪条路径。

ROUTE A · ADVERTISED PROJECT A 路径:已发布项目ROUTE B · SELF-PROPOSED PROJECT B 路径:自主课题
Who sets the topic
谁定课题
The supervisor. The aims, methods and often the equipment are fixed in the advert.
导师。目标、方法乃至设备通常已在广告中确定。
You, refined with a supervisor who agrees to host it.
你自己,并与同意接收你的导师共同完善。
Funding
资助
Attached to the project (studentship, grant or salaried post).
随项目提供(奖学金、科研经费或带薪职位)。
A separate scholarship competition (university or government scholarships, CSC).
另行竞争奖学金(校级或政府奖学金、国家留学基金委 CSC)。
Main documents
主要文书
Project fit statement, personal statement or motivation letter (1-2 pages), CV, referees. Some ask for a short research statement.
项目匹配陈述、个人陈述或动机信(1-2 页)、简历、推荐人。部分还要求简短的研究陈述。
Research proposal (usually 2,000-3,000 words or 2-5 pages), CV, personal statement, referees.
研究计划书(通常 2,000-3,000 词或 2-5 页)、简历、个人陈述、推荐人。
What the panel judges
评审看什么
Whether your skills and experience match the project, and whether you can start quickly.
你的技能与经历是否与项目匹配,能否尽快上手。
Whether the problem is worth a PhD, the method can answer it, and you can carry it out.
问题是否值得做博士、方法能否回答问题、你能否完成。
Where you meet it
常见于
Less common in Hong Kong: CUHK Biomedical Engineering asks only for a research-interest statement of about 500 words, and HKUST takes a research plan in the online form with no extra proposal.
在香港较少见:港中文生物医学工程只要约 500 词的研究兴趣陈述,港科大在网申表格中填写研究计划,无需另交计划书。
HKU (faculty proposal form, Engineering and Science), CUHK Engineering (proposal in English), PolyU (Form GSB/1A plus CV), CityU (research plan, supervisor consent advised), funded through HKPFS or university studentships.
港大(工程与理学学院表格)、港中文工程(英文计划书)、理大(GSB/1A 表格加简历)、城大(研究计划,建议先取得导师同意),由 HKPFS 或大学研究生奖学金资助。
First 先做Both routes start with an email

In many departments a supervisor’s support comes before the application. The next two pages cover that email.

在许多院系,导师支持先于正式申请。接下来两页讲这封邮件。

Route ARoute A: a fit statement

For an advertised project, the result depends on the fit statement and on how you present experience, publications and lab skills. The AcademiaOne personal statement and CV guides cover these.

申请已发布项目,成败取决于项目匹配陈述,以及你怎么写经历、成果和实验技能。AcademiaOne 个人陈述与简历指南专门讲这些。

Route BRoute B: this guide

Sections 0 to 6 of this guide build the proposal for a self-proposed application.

本指南第 0 至 6 部分逐步写出自主课题申请所需的计划书。

Before you write 动笔之前

The Supervisor-First Email联系导师邮件

Typical length 典型篇幅 150-200 words · CV and one-page summary attached

In Hong Kong, CityU asks you to seek a supervisor’s consent before you apply, and universities pre-screen and nominate HKPFS candidates, so a supervisor who already knows your plan is an advantage everywhere. In under a minute, the first email has to show that your background matches the lab’s current work. It should also ask a question the professor can answer quickly.

在香港,城大要求申请前先取得导师同意;各校还会先筛选并提名 HKPFS 候选人,因此一位已经了解你计划的导师在任何学校都是优势。第一封邮件要在一分钟内让教授看到你的背景与实验室当前研究相契合,并提出一个对方能快速回答的问题。

Blueprint结构

1Subject line

Topic and status, so it can be found later.

写明方向与身份,便于日后检索。

2Who you are

Degree, institution, intake year.

学位、学校、入学年份。

3Their work

One specific recent paper and why it matters to you.

对方近期一篇具体论文及其与你的关联。

4Your evidence

One result from your own work, with a number.

你自己研究中的一个结果,带数据。

5The ask

A yes/no question and attachments.

一个可用“是/否”回答的问题,加附件。

Worked example · Supervisor-first email 范例 · 联系导师邮件

1Subject: Prospective PhD student (2027): graph learning for fault location in inverter-rich distribution networks

Dear Professor [Surname],
2I am completing an MSc in Electrical Power Engineering at [University] and would like to join your group as a PhD student in 2027. 3I read your recent paper on protection in networks with high inverter-based generation (Authors, Year). Its finding that limited fault current weakens overcurrent protection is the starting point of my own work.

4In my dissertation I trained a graph neural network to locate faults on the IEEE 33-bus test feeder from measurements at six nodes; it placed 94% of simulated faults within one bus of the true location. I would now like to test whether adding power-flow constraints keeps this accuracy on larger feeders with high solar penetration, using hardware-in-the-loop simulation.

5Do you expect to take a PhD student in this area for 2027, and would a self-proposed project of this kind suit the [university scholarship / studentship] round? I attach my CV and a one-page research summary.
Kind regards,
[Full name] · [MSc programme, University] · [link to thesis or code repository]

Why it works 写法解析
1Searchable subject 可检索的标题

Year, topic and status in one line. Professors file and search these emails.

一行写明年份、方向与身份。教授会归档并检索这类邮件。

2Status first 先说身份

The professor knows in one sentence who is writing and when you would start.

一句话让教授知道你是谁、何时入学。

3One real paper 一篇真实论文

A specific finding in place of “I am very interested in your research”.

写出具体发现,而不是“我对您的研究非常感兴趣”。

4A number 用数据说话

A measured result from your own work usually does the most to persuade.

你自己研究中可量化的结果,通常最能打动人。

5Easy question 容易回答的问题

A yes/no question about places and funding gets a reply more often than a request for a call.

询问名额与资助的“是/否”问题,比请求通话更容易得到回复。

Before you write 动笔之前

Emails that get a reply能得到回复的邮件

✕ Weak 较弱“Dear Professor, I am very interested in your research and I hope to join your excellent team. I am hard-working and passionate about science.”

This could go to any professor in any field. It reads like a mass email and is often deleted unread.

可以发给任何领域的任何教授,看起来像群发邮件,常常未读即删。

✓ Stronger 更好“Your finding that limited fault current weakens overcurrent protection (Authors, Year) is the starting point of my MSc dissertation.”

Shows you read one paper closely and links it to your own work in one sentence.

表明你认真读过一篇论文,并用一句话把它与你自己的研究联系起来。

✕ Weak 较弱“I have rich research experience and strong programming skills.”

A claim with no evidence. The professor cannot check it.

只有断言没有证据,教授无从核实。

✓ Stronger 更好“My graph model placed 94% of simulated faults within one bus; the code is on [repository link].”

A result with a number and a link the professor can open.

有数据的结果,加上教授可以打开的链接。

✕ Weak 较弱A 600-word email with the full proposal pasted into the body.

Too long to read between meetings, and the main point gets lost.

太长,教授在会议间隙读不完,重点被淹没。

✓ Stronger 更好A 180-word email with the CV and a one-page summary attached.

Short enough to read on a phone, with the detail one click away.

短到可以在手机上读完,细节就在附件里。

Send it well 发送须知

Use a university or professional address and write to one professor at a time. Attach clearly named PDFs. If there is no reply after about two weeks, send one short follow-up, then move on. At CityU the supervisor’s consent comes before the application, so this email is the first step.

请用学校或正式邮箱,一次只写给一位教授,附件为命名清晰的 PDF。约两周无回复可发一封简短跟进邮件,之后转向其他导师。在城大,导师同意先于正式申请,所以这封邮件就是第一步。

Common email mistakes邮件常见错误

Wrong name or title 称呼错误

Misspelt names, “Dear Sir”, or Professor for a lecturer. Check the staff page.

名字拼错、写“Dear Sir”,或把讲师称为教授。请核对教职员页面。

Mass sending 群发

The same email to ten people in one department; they talk to each other.

同一院系发给十个人,老师之间会互相交流。

Asking what to research 问“我该研究什么”

Asking the professor to suggest a topic. Bring your own direction, even a rough one.

请教授替你想课题。哪怕方向还粗糙,也要自己带着方向来。

Biology 生物 Name the lab’s model system and one technique you have used on it: “I used CRISPR-Cas9 to knock out a gene in HEK293 cells and confirmed it by Western blot.”
Chemistry 化学 Name the material class and your synthesis or characterisation skills: “I synthesised and characterised layered oxide cathodes using XRD and SEM.”
Computer sci. 计算机 Link to code or a paper, and name one open problem from their recent work you would like to try.
Civil eng. 土木 Name the structure type and the test or model: “I ran four-point bending tests on reinforced concrete beams and built the matching FE model in ABAQUS.”
Start here 开始之前

How long, and how the words are shared篇幅与字数分配

For a self-proposed project (Route B), write one master proposal of about 2,500 words, then cut it down to fit each template. STEM panels spend most of their time on the methodology and work plan, so it gets the largest share. References are usually extra. In Hong Kong the limit is often set in pages or by a form (HKU Arts 4-5 pages double-spaced, HKUST Social Science 10-12 pages, HKU Education up to 20 pages), so check the spacing and whether references count before you cut.

B 路径请先写一份约 2,500 词的完整母版,再按各校模板裁剪。理工科评审把大部分时间花在研究方法与工作计划上,因此这部分字数最多。参考文献通常不计入字数。在香港,篇幅上限常以页数或表格来规定(港大文学院 4-5 页双倍行距,港科大社会科学部 10-12 页,港大教育学院不超过 20 页);裁剪之前请确认行距要求以及参考文献是否计入。

200500200950300300
0 Title & abstract1 Background & gap2 Aims & questions3 Method & work plan4 Contribution & fit5 Timeline & risks
Route A documents A 路径文书

For an advertised project you write a fit statement or personal statement (usually 1-2 pages) and a CV of about 2 pages instead of a proposal. The AcademiaOne personal statement and CV guides cover them; the email on the previous pages serves both routes.

申请已发布项目时,你写的是匹配陈述或个人陈述(通常 1-2 页)以及约 2 页的简历,而不是计划书。AcademiaOne 个人陈述与简历指南专门讲这些;前几页的联系导师邮件两条路径通用。

The five questions every STEM panel asks理工科评审必问的五个问题

#PANEL QUESTION 评审问题WHERE YOU ANSWER IT 在哪里回答
1Is the problem real, current and worth three to four years? 问题是否真实、前沿,值得三到四年?Background · Abstract · Fit statement 背景 · 摘要 · 匹配陈述
2Is the gap technical and specific, beyond “more research is needed”? 研究空白是否具体、有技术含量,而不只是一句“还需更多研究”?Background & gap 背景与空白
3Will the method answer the question, with baselines, controls and validation? 方法能否回答问题,是否有基准、对照与验证?Methodology & work plan 方法与工作计划
4Has this person already used these tools, techniques or instruments? 申请人是否已经用过这些工具、技术或仪器?CV · Experience · Methodology 简历 · 经历 · 方法
5Does it fit this lab’s equipment, funding and current projects? 是否契合该实验室的设备、经费与在研项目?Contribution, fit & resources 贡献、匹配与资源
Before you write 动笔之前

The golden thread逻辑主线

A science or engineering application holds together when each part follows from the one before: a technical problem leads to a gap in current methods, the gap defines the question, and the method and validation answer it. If you cannot write these four sentences, narrow the topic before drafting anything else. The same four sentences become the core of your email and your fit statement.

好的理工科申请,每一部分都接着上一部分往下走:一个技术问题暴露出现有方法的不足,这个不足决定了研究问题,方法和验证再来回答它。如果这四句话写不出来,先把课题缩小,再写别的。联系导师的邮件和项目匹配陈述,也是围绕这四句话写的。

PROBLEM技术问题

As rooftop solar and other inverter-based generation grow, faults on distribution networks produce smaller and less distinctive fault currents, so overcurrent protection and impedance-based fault location can miss or misplace them.

GAP研究空白

Data-driven fault locators perform well in simulation, but most assume dense measurements, a fixed network topology and little inverter-based generation.

METHOD研究方法

Train a physics-informed graph neural network on simulated faults in IEEE test feeders with varying solar penetration, then validate it with hardware-in-the-loop tests.

CONTRIBUTION研究贡献

A fault detection and location method that keeps working as solar grows, an open benchmark dataset, and guidance on where to place the few sensors a utility can afford.

The worked example used in this guide本指南使用的范例

Example applicant 范例申请人

Background 背景
BEng Electrical Engineering and Automation (Mainland China); MSc Electrical Power Engineering (UK), with a dissertation on graph neural networks for fault location on the IEEE 33-bus test system; one year as a research assistant on a phasor-measurement project; second author of one conference paper.

Applying to 申请
PhD in Electrical Engineering in Hong Kong: a self-proposed project (Route B), with advertised studentships (Route A) in the same cycle.

Working title 暂定题目
Physics-Informed Graph Learning for Fault Detection and Location in Distribution Networks with High Inverter-Based Generation

How to use the example 如何使用范例

The same power-systems project appears in every section, from the first email to the timeline, so you can watch one application take shape. Don’t copy the topic. Study the moves in the numbered notes beside each extract, then make the same moves with your own field, data and supervisor. Citations shown as (Authors, Year) are placeholders; in your own documents every claim needs a real source you have read, and every result you report must be your own. The examples are in English only, because your application will be written in English.

同一个电力系统项目贯穿每个部分,从第一封邮件到时间安排,方便你看到一份申请如何成形。不要照搬课题,要学习写法:先看每段旁边的编号解析,再用你自己的方向、数据和导师做同样的事。(Authors, Year) 为占位符,你自己的文书里每个论断都必须有你读过的真实文献,你报告的每个结果都必须是你自己的。范例只有英文,因为你的申请材料也是用英文写的。

Section 0 第 0 部分
0

Title & Abstract题目与摘要

Typical length 典型篇幅 title 10-20 words · abstract 150-250 words

STEM titles name the method, the task and the condition or system that makes the problem hard. A title that only names a field is too broad. The abstract states the problem, the gap, the question, the method and validation, and the expected output in plain technical language.

理工科题目要写出方法、任务,以及让问题变难的条件或系统,避免只写一个领域名称。摘要用平实的技术语言说清问题、空白、研究问题、方法与验证、预期产出。

Anatomy of a title题目结构

Physics-Informed Graph Learning Method 方法
for Fault Detection and Location Task 任务
in Distribution Networks with High Inverter-Based Generation System and condition 系统与条件

Weak vs stronger titles较弱与更好的题目

✕ Weak 较弱“Research on Artificial Intelligence in Smart Grids”

This names a field, with no project in it. “Research on” is a direct translation of 研究 and adds nothing.

这只是领域,看不出项目。“Research on”是“……研究”的直译,没有信息量。

✓ Stronger 更好“Physics-Informed Graph Learning for Fault Location in Distribution Networks with High Inverter-Based Generation”

Method, task and the hard condition, in one line.

一行写出方法、任务和难点条件。

✕ Weak 较弱“A Novel High-Accuracy Deep Learning Method for Fault Detection”

“Novel” and “high-accuracy” are claims the panel cannot judge from a title.

“新颖”“高精度”是评审无法从题目判断的自我评价。

✓ Stronger 更好“Fault Detection from Sparse Measurements under Changing Network Topology”

Names the constraints that make the problem worth a PhD.

点明让问题值得做博士的约束条件。

Biology 生物 Role of [Gene X] in Chemoresistance of Triple-Negative Breast Cancer Cells: A CRISPR Knockout and Rescue Study
Chemistry 化学 Surface-Coated Nickel-Rich Cathodes for Long-Cycle Lithium-Ion Batteries: Synthesis and In-Situ Characterisation
Computer sci. 计算机 Calibrating Large Language Model Confidence on Rare Cases with Retrieval-Based Uncertainty Estimates
Civil eng. 土木 Vision-Based Crack Detection and Width Measurement for Reinforced Concrete Bridge Inspection
Section 0 continued 第 0 部分(续)

The abstract摘要

Blueprint结构

1Move 1 · Problem

The technical problem and why it is growing.

技术问题及其为何日益突出。

2Move 2 · Gap

What current methods cannot do.

现有方法做不到什么。

3Move 3 · Question

The question or aim, stated outright.

直接写出研究问题或目标。

4Move 4 · Method

Model or experiment, and how it is validated.

模型或实验,以及验证方式。

5Move 5 · Output

What will exist at the end.

结束时会产出什么。

Worked example · Abstract 范例 · 摘要

1Distribution networks are connecting more rooftop solar and other inverter-based generation, which limits fault current and changes how faults appear in measurements (Authors, Year). Overcurrent protection and impedance-based fault location, designed for networks fed by large synchronous generators, can then miss faults or misplace them.

2Data-driven methods locate faults accurately in simulation, but most assume dense measurements, a fixed topology and low solar output (Authors, Year). 3This project asks whether a graph neural network that encodes the network’s topology and power-flow equations can detect and locate faults from sparse measurements under high and changing inverter-based generation.

4The model will be trained on simulated faults in IEEE test feeders across a range of solar penetration levels and network configurations, compared with impedance-based and purely data-driven baselines, and validated in hardware-in-the-loop tests on a real-time simulator. 5The project will produce a fault locator that keeps working as solar grows, an open benchmark dataset, and guidance on where to place a limited number of sensors.

Why it works 写法解析
1Problem with a trend 问题与趋势

States the change in the system that creates the problem.

写出系统中引发问题的变化。

2Gap in methods 方法上的空白

Names three assumptions current methods make; the project removes them.

点明现有方法的三个假设,本项目将去除它们。

3One question 一个问题

The question repeats the three conditions from the gap.

研究问题呼应空白中的三个条件。

4Baselines and validation 基准与验证

Says what the method is compared with and how it is tested beyond simulation.

写明与什么方法比较,以及如何在仿真之外检验。

5Concrete outputs 具体产出

A method, a dataset and design guidance. A reader can use each of them.

方法、数据集和设计建议,每一项读者都能使用。

Write it last 最后再写

Write the abstract after the rest of the proposal is finished, then check that every sentence matches a section. On forms with a short summary box, sometimes limited to 100 words, keep moves 2, 3 and 4.

摘要最好在整份计划书完成后再写,并核对每句话都能对应到某一部分。如表格只有简短摘要栏(有时限 100 词),保留第 2、3、4 步即可。

Section 1 第 1 部分
1

Background, State of the Art & Gap背景、研究现状与空白

Typical length 典型篇幅 400-600 words · about 20% 约占 20%

In STEM the literature review works as a map of methods. Group existing work by approach, say what each approach achieves and under which assumptions, and end on the assumption that fails in your setting. That assumption is your gap.

理工科的文献综述是一张“方法地图”:按研究路线对现有工作分组,说明每类方法取得了什么、依赖哪些假设,最后落在你的场景中不成立的那个假设上,这就是你的研究空白。

Blueprint结构

1Para 1 · The system

The technical setting and the change that creates the problem.

技术背景,以及引发问题的变化。

2Para 2 · Approach families

Two or three groups of methods, with what each achieves.

两三类方法及其成效。

3Para 3 · Shared assumption

What they all assume, and why it fails here.

它们共同的假设及其为何在此失效。

4Para 4 · The gap

One sentence: what has not been done.

一句话:什么还没做。

Worked example · Background and gap 范例 · 背景与空白

1Inverter-based resources such as solar photovoltaic systems control their output electronically and limit their fault current to little more than rated current (Authors, Year). As their share in distribution networks grows, fault currents become smaller and their direction and waveform change, which weakens protection schemes designed for networks fed by synchronous generators (Authors, Year).

2Fault location methods fall into three groups. Impedance-based methods estimate distance from voltage and current at the substation and are cheap to deploy, but lose accuracy when generation along the feeder supplies part of the fault current (Authors, Year). Travelling-wave methods are fast and precise but need high-sampling-rate devices that are costly to install on every feeder (Authors, Year). Data-driven methods, including recent graph neural networks, learn the relation between measurements and fault location and report high accuracy on test feeders (Authors, Year).

3Most data-driven studies, however, train and test on one fixed topology, assume measurements at many nodes, and include little or no inverter-based generation (Authors, Year). Real distribution networks are reconfigured during operation and have few sensors. 4No method has yet been shown to locate faults from sparse measurements when both topology and inverter-based generation change.

Why it works 写法解析
1Mechanism explained 讲清机理

Explains why inverter-based generation causes the problem, with a citation.

用引用解释逆变器型电源为何导致问题。

2Map of approaches 方法地图

Three families, each with its strength and its limit.

三类方法,各有优势与局限。

3The failing assumption 失效的假设

The gap comes from assumptions in existing work. A shortage of papers is not a gap.

空白来自现有研究的假设。“论文不够多”算不上空白。

4Gap in one sentence 一句话的空白

Precise enough that the panel can check whether it is true.

精确到评审可以核实其真伪。

Section 1 continued 第 1 部分(续)

Writing a gap a panel believes写出评审认可的研究空白

✕ Weak 较弱“With the rapid development of the economy and society, electricity has become more and more important in people’s daily life.”

A generic opening that any proposal could use. It tells the panel nothing about the technical problem.

任何计划书都能用的泛泛开头,没有告诉评审任何技术问题。

✓ Stronger 更好“Inverter-based resources limit their fault current to little more than rated current (Authors, Year).”

Starts from a technical fact with a citation, so the problem is clear in one line.

从有引用的技术事实开始,一句话就讲清问题。

✕ Weak 较弱“There are few studies on this topic, so it is worth studying.”

A small number of studies is not a gap. The panel cannot tell what is missing or why it matters.

研究少不等于有空白;评审看不出缺什么、为何重要。

✓ Stronger 更好“Most data-driven studies assume a fixed topology and measurements at many nodes; real feeders are reconfigured and have few sensors.”

Names the assumption and the real condition that breaks it.

点明假设,以及打破这一假设的现实条件。

Common background mistakes背景部分常见错误

Textbook background 教科书式背景

Explaining what a neural network or PCR is. The panel knows; spend the words on the problem.

解释什么是神经网络或 PCR。评审都懂,把篇幅用在问题上。

Paper-by-paper list 逐篇罗列

One paragraph per paper, with no grouping or comparison.

每篇论文一段,没有归类与比较。

Out-of-date work 文献陈旧

In fast fields such as machine learning, leaving out the last two or three years suggests you are not keeping up.

在机器学习这类变化快的领域,漏掉近两三年的工作,会让人觉得你没在跟进。

Biology 生物 Background: what is known about the pathway. Approaches: genetic, pharmacological, clinical association. Gap: no causal test in this cell type.
Chemistry 化学 Background: why the material degrades. Approaches: doping, coating, electrolyte additives. Gap: no coating survives high-voltage cycling beyond a stated limit.
Computer sci. 计算机 Background: where models fail. Approaches: post-hoc calibration, ensembles, retrieval. Gap: none tested on long-tail cases at scale.
Civil eng. 土木 Background: inspection cost and risk. Approaches: manual, sensor-based, vision-based. Gap: crack-width accuracy not validated in field lighting.
Section 2 第 2 部分
2

Aims, Objectives & Research Questions研究目标与研究问题

Typical length 典型篇幅 150-250 words · about 8% 约占 8%

State one aim and three or four objectives. In STEM each objective should end in something that can be measured or delivered: a model that reaches a stated accuracy, a material with a stated property, a dataset or a validated prototype. Research questions or hypotheses follow from the objectives.

写一个总目标和三到四个具体目标。理工科的每个具体目标都应以可测量或可交付的结果结尾:达到某一精度的模型、具备某一性能的材料、一个数据集或一个经过验证的原型。研究问题或假设由具体目标导出。

Blueprint结构

1Aim

One sentence: what the project achieves overall.

一句话:项目总体要实现什么。

2Objectives O1-O3

Each with a measurable end point.

每个都有可测量的终点。

3Questions or hypotheses

One per objective, answerable with data.

每个目标对应一个,可用数据回答。

Worked example · Aims and objectives 范例 · 目标

1Aim. To develop and validate a fault detection and location method for distribution networks that remains accurate from sparse measurements under high and changing inverter-based generation.

2O1. Build an open benchmark of simulated faults on two IEEE test feeders, covering fault types, fault resistances, solar penetration from 0 to 100% of peak load, and at least five network configurations.
O2. Develop a physics-informed graph neural network and compare it with impedance-based and purely data-driven baselines on detection rate and location error.
O3. Identify the smallest set of measurement points that keeps location error within one bus, and validate the method in hardware-in-the-loop tests.

3RQ1. Does adding power-flow constraints reduce location error relative to a purely data-driven graph model when solar penetration is high?
RQ2. How does accuracy change when the network topology at test time differs from training?
RQ3. How many measurement points are needed, and where, to keep location error within one bus?

Why it works 写法解析
1One aim 一个总目标

Repeats the gap’s conditions: sparse measurements, changing generation.

重复空白中的条件:稀疏量测、变化的电源。

2Measurable objectives 可测量的目标

Each objective names its own deliverable and test.

每个目标都写明交付物和检验方式。

3Matched questions 对应的问题

RQ1 to RQ3 map one to one onto O2 and O3, so the panel can follow the plan.

RQ1 到 RQ3 与 O2、O3 一一对应,评审能看清计划。

✕ Weak 较弱“Objective: to study fault detection using AI and improve accuracy.”

No system, no comparison, no end point. When is the objective met?

没有系统、没有比较、没有终点。何时算完成?

✓ Stronger 更好“O2: compare the model with two baselines on detection rate and location error on two IEEE feeders.”

Method, comparison, metric and test system are all there, so the finish line is clear.

方法、比较、指标与测试系统俱全,终点明确。

Biology 生物 H1: knocking out [Gene X] increases drug sensitivity in two cell lines; rescue restores resistance.
Chemistry 化学 O1: coated cathode keeps at least a stated share of its capacity after a stated number of cycles at a stated voltage.
Section 3 第 3 部分
3

Methodology & Work Plan研究方法与工作计划

Typical length 典型篇幅 800-1,100 words · about 38% 约占 38%

A STEM panel decides here whether the project can be done, and whether you are the person to do it. Organise the method by work package, one per objective. For each, give the data, simulation or experimental set-up, the model or procedure, the baselines or controls, the metrics, and how you will validate the result beyond your own set-up.

理工科评审主要看这一部分来判断项目做不做得成,以及你做不做得成。按工作包组织方法,每个目标对应一个工作包。每个工作包写明数据、仿真或实验设置,模型或流程,基准或对照,评价指标,以及如何在自己的设置之外验证结果。

Blueprint结构

1 · Set-up

Data, test systems, materials or samples.

数据、测试系统、材料或样本。

2 · Method

Model, procedure or synthesis route.

模型、流程或合成路线。

3 · Comparison

Baselines, controls, replicates.

基准、对照、重复。

4 · Metrics

How success is measured.

如何衡量成功。

5 · Validation

Tests beyond your own set-up.

超出自身设置的检验。

Worked example · Methodology by work package 范例 · 按工作包写方法

1WP1 · Benchmark (O1). I will model the IEEE 33-bus and IEEE 123-node test feeders in OpenDSS, adding solar systems at a range of locations. Faults of all common types will be simulated across a range of fault resistances, solar outputs and at least five switching configurations, giving an estimated 50,000 labelled cases. Electromagnetic-transient checks in PSCAD will be run on a subset to confirm that inverter fault behaviour is represented correctly.

2WP2 · Model (O2). The network will be represented as a graph, with buses as nodes and lines as edges. A graph neural network will predict the faulted line from measurements at a subset of nodes, with a loss term that penalises predictions inconsistent with the power-flow equations (Authors, Year). 3It will be compared with an impedance-based method and a purely data-driven graph model trained on the same data. 4Metrics are detection rate, false-alarm rate, and location error in buses and in kilometres, reported separately for each solar level and for topologies not seen in training.

5WP3 · Sensors and validation (O3). A greedy search will find the smallest set of measurement points that keeps location error within one bus. The final model will then be tested in hardware-in-the-loop on the group’s real-time simulator, with realistic measurement noise and delays. If a utility partner can provide anonymised fault records, they will be used as a final test set.

Why it works 写法解析
1Named tools and systems 具体工具与系统

Standard test feeders and named software make the plan checkable and reproducible.

标准测试系统与具体软件让计划可核查、可复现。

2Method in plain steps 方法分步清楚

Graph, inputs, output and the physics term, each in one sentence.

图结构、输入、输出与物理约束项,各用一句话说明。

3Fair baselines 公平的基准

Two baselines trained on the same data show whether the gain holds up.

两个基准用同样的数据训练,才能看出提升站不站得住。

4Metrics by condition 分条件报告指标

Reporting by solar level and unseen topology tests the gap directly.

按光伏水平与未见拓扑分别报告,直接检验研究空白。

5Beyond simulation 走出仿真

Hardware-in-the-loop and optional field data answer “does it work outside your laptop?”.

硬件在环与可选的现场数据,回答“离开你的电脑还管用吗”。

Section 3 continued 第 3 部分(续)

Making the method convincing让方法更有说服力

✕ Weak 较弱“I will use deep learning to analyse the data and achieve high accuracy.”

No data, no model, no baseline, no metric. The panel cannot judge feasibility.

没有数据、模型、基准和指标,评审无法判断可行性。

✓ Stronger 更好“A graph neural network with a power-flow loss term, compared with two baselines on detection rate and location error, by solar level.”

A supervisor could start planning experiments from this one line.

导师凭这一句就能开始规划实验。

✕ Weak 较弱“The expected result is that my method outperforms all existing methods.”

Predicts the result and leaves no plan if it does not happen.

预设了结果,也没有结果不理想时的预案。

✓ Stronger 更好“If the physics term does not help, WP3 still delivers the sensor-placement results and the open benchmark.”

Shows the project still produces useful outputs if the main result is negative.

表明无论结果如何,项目都有有用的产出。

Feasibility checklist可行性清单

CHECK 检查项WHAT THE PANEL WANTS TO SEE 评审希望看到
Equipment and compute
设备与算力
Instruments, simulators or GPU time the lab already has. Check the lab page before naming them.
实验室已有的仪器、仿真器或 GPU 资源。写之前先核对实验室页面。
Data and samples
数据与样本
Public datasets, standard test systems, or samples you can obtain; a fallback if a partner’s data does not arrive.
公开数据集、标准测试系统或可获得的样本;合作方数据未到位时的备选方案。
Skills
技能
Methods you have already used, and which ones you will learn in Year 1.
你已用过的方法,以及第一年将学习的方法。
Safety, ethics and approvals
安全、伦理与审批
Lab safety training, animal or human-tissue licences in biology, export-control or ATAS checks in sensitive engineering fields.
实验室安全培训;生物方向的动物或人体组织许可;敏感工程领域的出口管制或 ATAS 审查。
Biology 生物 Name cell lines or model organisms, conditions, controls (non-targeting guide, rescue), biological replicates and the statistical test.
Chemistry 化学 Give the synthesis route, characterisation (XRD, SEM, XPS), electrochemical tests and how many batches you will make.
Computer sci. 计算机 Name datasets, baselines, metrics, compute budget and ablations; say what code and data will be released.
Civil eng. 土木 Give specimen design and number, test set-up and sensors, the FE model and how it is calibrated against tests.
Section 4 第 4 部分
4

Contribution, Fit & Resources贡献、匹配与资源

Typical length 典型篇幅 250-350 words · about 12% 约占 12%

Say what the project adds to the field and to practice, then why this lab is the place to do it. In STEM, fit depends on people and on resources such as equipment, test facilities, data access and industrial partners. A lab head reads this section to judge whether you would strengthen current projects.

说明项目对学科和实践的贡献,再说明为什么要在这个实验室做。理工科的匹配要看人,也要看资源,比如设备、测试平台、数据渠道和产业合作方。实验室负责人读这一部分,是想判断你能不能给在研项目帮上忙。

Blueprint结构

1Contribution to the field

The method, data or finding that is new.

新的方法、数据或发现。

2Contribution to practice

Who uses it, and how.

谁会使用、如何使用。

3People

Two staff and a specific link each.

两位老师,各有具体联系。

4Resources

Facilities the project needs and the lab has.

项目需要且实验室具备的设施。

Worked example · Contribution and fit 范例 · 贡献与匹配

1The project contributes a fault-location method designed for the conditions under which current data-driven methods have not been tested, and an open benchmark that other groups can use to compare methods on the same cases. 2For distribution network operators, it gives guidance on how many sensors are needed, and where, as solar connections grow.

3I am applying to [School] because Professor A’s work on protection in inverter-dominated networks (Authors, Year) defines the problem this project addresses, and Dr B’s research on graph-based state estimation (Authors, Year) provides the modelling approach I would extend. 4The school’s real-time digital simulation laboratory makes the hardware-in-the-loop validation in WP3 possible, and its links with network operators offer a route to field data.

Why it works 写法解析
1Two kinds of contribution 两类贡献

A method for researchers and a dataset others can reuse.

给研究者的方法,加上他人可复用的数据集。

2Practice in one line 实践意义一句话

Names the user and the decision the work informs.

写明使用者以及研究支持的决策。

3Problem and method 问题与方法

One staff member for the problem and one for the method.

一位老师对应问题,另一位对应方法。

4Named facility 具体设施

Links a facility to a specific work package.

将具体设施与某个工作包挂钩。

✕ Weak 较弱“I chose this university because it has a world-class engineering school and an excellent research environment.”

True of many universities, and says nothing about this lab.

适用于许多大学,与这个实验室毫无关系。

✓ Stronger 更好“The school’s real-time digital simulation laboratory makes the hardware-in-the-loop validation in WP3 possible.”

A specific resource tied to a specific part of the plan.

具体资源对应计划中的具体部分。

Section 5 第 5 部分
5

Timeline & Risks时间安排与风险

Typical length 典型篇幅 150-300 words or one table · about 12% 约占 12%

Show the work packages across the PhD, with an output at each stage, and name the main technical risks with a fallback for each. Many STEM supervisors expect a conference paper in the second year and journal papers before submission, so build them into the plan.

展示各工作包在博士期间的安排及各阶段产出,并写明主要技术风险与对应备选方案。许多理工科导师期望第二年有会议论文、提交论文前有期刊论文,请在计划中体现。

Worked example · Timeline (three-year plan) 范例 · 时间安排(三年计划)

MONTHS 月份WORK 工作OUTPUT 产出
1-6Literature review; WP1 benchmark on two IEEE feeders; PSCAD checksReview chapter; dataset v1
7-15WP2 physics-informed model and baselines; first-year reviewConference paper
16-24WP2 unseen-topology tests; WP3 sensor placementJournal paper 1; open benchmark released
25-32WP3 hardware-in-the-loop validation; field data if availableJournal paper 2
33-36Thesis writing and submissionThesis

Why it works 写法解析
1Outputs each stage 每阶段有产出

Each row ends in something a supervisor can see.

每一行都以导师能看到的产出结尾。

2Papers planned 论文已规划

Conference and journal papers appear where the work allows them.

会议与期刊论文安排在研究进度允许的位置。

Risks and fallbacks风险与备选方案

RISK 风险FALLBACK 备选方案
Simulated inverter fault behaviour differs from real devices
仿真中的逆变器故障特性与真实设备不同
Use the PSCAD subset to calibrate; report results with both models.
用 PSCAD 子集校准,并报告两种模型下的结果。
The physics term gives no gain
物理约束项没有带来提升
Report it as a finding; WP3 sensor-placement results still stand.
作为研究发现报告;第 3 工作包的传感器布置结果依然成立。
No field data from a utility
无法获得电网公司的现场数据
Hardware-in-the-loop tests become the final validation.
以硬件在环测试作为最终验证。
Four-year and cohort programmes 四年制与成组培训项目

Some PhDs, including cohort-based doctoral training programmes, run for four years with taught training at the start. Add the training year and move each work package back. Where there is a confirmation or upgrade review in the first year, put it in the plan.

部分博士项目(包括成组培训的博士项目)为四年,开头有课程培训。请加入培训年,并把各工作包顺延。如果第一年有确认或升级评审,请写进计划。

Section 6 第 6 部分
6

References参考文献

Usually excluded from the word count 通常不计字数 15-30 main sources

Cite the leading journals and conferences of your field, the papers behind each method you use, and recent work from the lab where it applies. Engineering and computer science usually use numbered IEEE style; chemistry often uses ACS style; biology often uses author-date or Vancouver style. Follow the template if the university gives one.

引用本领域顶级期刊与会议、你所用每种方法的出处,以及在确实相关时引用目标实验室的近期工作。工程与计算机通常使用编号式 IEEE 格式;化学常用 ACS 格式;生物常用作者-年份或温哥华格式。学校提供模板时以模板为准。

Blueprint结构

Field venues

Leading journals and, in computing, top conferences.

顶级期刊;计算机方向还包括顶级会议。

Method papers

The source of each model, technique or test standard.

每个模型、技术或测试标准的出处。

Test systems and data

Cite the test feeder, dataset or protocol you use.

引用所用测试系统、数据集或实验规程。

One style

One citation style, used consistently.

统一使用一种引用格式。

Worked example · Reference list in IEEE style (excerpt) 范例 · IEEE 格式参考文献节选

[1] A. Author and B. Author, “Fault current characteristics of inverter-based resources,” Journal Name, vol. 00, no. 0, pp. 000-000, Year.

[2] C. Author et al., “Graph neural networks for fault location in distribution systems,” in Proc. Conference Name, Year, pp. 000-000.

[3] D. Author, “Physics-informed machine learning for power systems,” Journal Name, vol. 00, pp. 000-000, Year.

Why it works 写法解析
1Real sources only 只用真实文献

In your own proposal every reference must be a paper you have read.

你自己的计划书里,每条参考文献都必须是你读过的论文。

2Numbered in order 按出现顺序编号

In IEEE style, references are numbered in the order they first appear in the text.

IEEE 格式按正文首次出现的顺序编号。

Blogs and tutorials 博客与教程

Citing websites or online tutorials instead of papers and standards.

引用网页或网络教程,而不是论文和标准。

Preprints only 只有预印本

In fast fields preprints are normal, but balance them with peer-reviewed work.

在快速发展的领域引用预印本很正常,但要与同行评审文献相平衡。

Mixed styles 格式混用

Numbered and author-date citations mixed in one document.

同一份文件里混用编号式与作者-年份式。

Putting it to work 实际应用

Adapting to each university按校调整

Keep two master documents: the 2,500-word proposal and a one-page fit statement built from it. Cut and adjust from these. Do not start again for each application.

保留两份母版:2,500 词的计划书,以及由它改写而成的一页项目匹配陈述。之后只做裁剪与调整,不要为每份申请从头写起。

TARGET 目标WHAT TO DO 做法
Short proposal (one page to 1,000 words)
短篇计划书(一页至 1,000 词)
CUHK Biomedical Engineering (research-interest statement of about 500 words) 港中文生物医学工程
Keep the gap, the aim with three objectives, one paragraph per work package and a three-line timeline. Name the supervisor you have contacted.
保留研究空白、目标与三个具体目标、每个工作包一段,以及三行时间安排。写明已联系的导师。
Proposal or research statement (2-5 pages)
计划书或研究陈述(2-5 页)
CUHK Engineering (proposal in English); CityU (research plan) 港中文工程、城大
Cut the master to the department limit; keep Sections 1-4 and a short timeline with risks. Name the supervisor you have contacted.
按院系上限裁剪母版;保留第 1-4 部分以及含风险的简短时间安排。写明已联系的导师。
Department template or form
院系模板或表格
HKU faculty proposal form (Engineering, Science); PolyU Form GSB/1A plus CV; HKUST research plan in the online form 港大学院表格、理大 GSB/1A、港科大网申研究计划
Paste the matching parts of the master under the template headings and keep within each limit; add the extra statements the form asks for.
把母版对应内容填入模板标题下并遵守各栏字数;补充表格要求的其他陈述。
Supervisor first
先联系导师
CityU (seek a supervisor’s consent first; referees in Electrical Engineering must hold PhDs) 城大
Send the email and a one-page summary first; finalise the proposal with the supervisor before you apply.
先发邮件和一页摘要;申请前与导师共同定稿计划书。

The email before the proposal计划书之前的那封邮件

In Hong Kong your first email is part of the application. CityU asks for a supervisor’s consent before you apply, and because universities pre-screen and nominate HKPFS candidates, a supervisor who already knows your plan helps everywhere. Write in September or October, well before 1 December. Give the supervisor enough detail to judge the fit: one specific link to their work, your question and method in one sentence each, one easy question, and your CV plus a short summary attached. Write the first draft yourself; your academic will check the tailored paragraph.

在香港,第一封邮件就是申请的一部分。城大要求申请前先取得导师同意;由于各校会先筛选并提名 HKPFS 候选人,一位已经了解你计划的导师在任何学校都有帮助。请在 9 月或 10 月发出,远早于 12 月 1 日。写得足够详细,让导师能判断是否匹配:联系对方一项具体研究,用一句话写研究问题、一句话写方法,只问一个容易回答的问题,并附简历和一份简短摘要。第一稿由你自己撰写,导师只负责核对针对导师的那一段。

Before you submit 提交之前

Read it like a panel用评审的眼光重读

Reread your documents as a lab head would, with ten minutes to spare and forty applications to read. Score each question 1 (weak), 2 (adequate) or 3 (strong), and fix the lowest score first.

像一位只有十分钟、还有四十份申请要读的实验室负责人那样重读。每项打 1(弱)、2(一般)、3(强)分,先修改得分最低的一项。

#CHECK 检查SELF-SCORE 自评
1Have I used the right route and document for this school? 我对这所学校用了正确的路径和文书吗?1 2 3
2Can I state the problem and the gap in two sentences from memory? 我能凭记忆用两句话说出问题与空白吗?1 2 3
3Does every claim about my experience have a method, a result or a link? 关于我经历的每个说法都有方法、结果或链接吗?1 2 3
4Does the method name the set-up, baselines or controls, metrics and validation? 方法是否写明设置、基准或对照、指标与验证?1 2 3
5Is the fit specific to named staff and to equipment this lab has? 匹配部分是否针对具体老师和该实验室已有的设备?1 2 3
6Does the timeline show outputs and a fallback for each main risk? 时间安排是否有阶段产出,并为主要风险准备备选?1 2 3
7Is every publication status and skill level stated accurately? 每篇论文的状态与每项技能的程度都如实写明了吗?1 2 3

Three tests三个测试

The lab-head test

Would the supervisor see you working in their lab next month? If they would have to train you in everything, say what you will learn first.

导师能想象你下个月就在实验室工作吗?如果样样都要培训,请说明你会先学什么。

The reviewer test

Imagine a journal reviewer reads your method. What is their first objection? Answer it in the text.

想象期刊审稿人读你的方法部分,第一个质疑是什么?在正文中回答它。

The two-minute test

Explain the problem, the gap, the method and the expected output in two minutes. Where you stumble, the writing is unclear.

用两分钟讲清问题、空白、方法与预期产出。卡壳的地方就是写得不清楚的地方。