How do you find out what is already known?

Search, snowball and map twelve papers

Research studio R2 of Energy Systems, Step by Step: build logged search strings against a dated OpenAlex snapshot, snowball from a review, code a twelve-paper concept matrix, search its empty cell directly, read the closest paper in a first and a second pass and write a gap statement checked against the course rubric.
  • Researcher · L4
  • Research studio
  • 30 min guided · about 4 h practice
  • After R1 and N1

The task

  1. Produce: a search log, a concept matrix of 12 papers with an empty cell searched directly, a second-pass extraction sheet on the closest paper, and a gap statement with its evidence.
  2. From: the question below, two open databases (OpenAlex and one more, such as Semantic Scholar or Crossref) and one review to snowball from.
  3. Done when: the rubric's bar for positioning: a concept matrix of at least 12 sources and a stated gap with evidence.

About 30 minutes guided here, then about 4 hours on your own.

The worked question: How do capacity-expansion models stand in for a whole year with a few representative periods, and what happens to storage and to extreme periods when they do?

Five search strings, OpenAlex titles and abstracts, 3 November 2026

Change String Works
Plain words representative days capacity expansion 344
Quote one phrase, drop the model words (two changes) "representative days" 649
Synonyms joined by OR "representative days" OR "representative periods" OR "time series aggregation" OR "typical days" 2,191
AND the model block ("representative days" OR "representative periods" OR "time series aggregation" OR "typical days") AND ("capacity expansion" OR "generation expansion" OR "energy system model" OR "investment planning") 113
AND storage ("representative days" OR "representative periods" OR "time series aggregation" OR "typical days") AND ("capacity expansion" OR "generation expansion" OR "energy system model" OR "investment planning") AND storage 49
The empty cell, searched directly ("representative days" OR "representative periods" OR "time series aggregation" OR "typical days") AND extreme AND storage 23

Each count links the same search on OpenAlex's search page; counts there may differ as OpenAlex updates its records.

Snowballing from Hoffmann et al. (2020): 196 references (backward) and 229 citing works (forward) on 3 November 2026. Of the 12 matrix papers, the AND-the-model string finds 9, backward snowballing 9 and forward snowballing 1; together, all 12.

The concept matrix, coded from the abstracts (first pass)

Tick a cell only when the abstract says so. “not stated” means the abstract does not say: it is not the same as no, and a second pass can change it. yes: the abstract reports it; partly: the abstract hints at it; not stated: the abstract does not say (not the same as no).

Paper Links storage between periods Keeps extreme periods Checks against a full-year run
Poncelet et al. (2016)
Picks representative days with an optimisation and compares it with clustering and simple rules, using new measures of how representative a set is.		 not stated not stated not stated
Kotzur et al. (2017)
Compares aggregation methods across several energy-system designs; the system matters more than the method, and typical periods as usually modelled cannot carry seasonal storage.		 not stated not stated partly
Pfenninger (2017)
Compares downsampling, clustering and simple rules in a Great Britain power model with 25 years of wind and solar data; the best method depends on the data and the model.		 not stated not stated partly
Kotzur et al. (2018)
Links typical periods through a storage level that carries energy from one period to the next, so seasonal storage can be modelled, and compares with optimising the full year.		 yes not stated yes
Liu, Sioshansi and Conejo (2017)
Picks representative days with two hierarchical clustering methods that keep correlations in time and between places, tested on the Texas power system against the full data.		 not stated not stated yes
Tejada-Arango et al. (2018)
Adds links between representative days so that long-term storage fits an investment model, and benchmarks it against an hourly unit-commitment model.		 yes not stated yes
Mallapragada et al. (2018)
Compares a model with up to 12 representative days against a time-slice model for Texas, then runs both plans through an hourly operations model.		 not stated not stated yes
Teichgraeber and Brandt (2019)
Compares five clustering methods on two operating problems, battery and gas-turbine scheduling; a better clustering score does not always give a better result.		 not stated not stated partly
Sun et al. (2019)
Groups days by the investment decisions each day would lead to, rather than by the input data alone, for large power systems with many locations.		 not stated not stated not stated
Yeganefar et al. (2020)
Adds extreme days of high and low net load to the representative days, picked with self-organising maps, and tests them on two power systems.		 not stated yes partly
Teichgraeber, Küpper and Brandt (2021)
Adds extreme periods one at a time, using the planning model's own results, until the design meets demand in every hour of the full data.		 not stated yes yes
Reichenberg and Hedenus (2022)
Measures how representative periods distort cost and capacity results against a full chronological year, in a European model with 90% wind and solar.		 not stated not stated yes

Abstracts reporting both a storage link and extreme periods: 0 papers (the empty cell). Storage link only: 2. Extreme periods only: 2. Neither stated: 8.

Searching that cell directly (the empty cell, searched directly) gave 23 works on 3 November 2026; at least 3 of them do both, by their abstracts: Hilbers, Brayshaw and Gandy (2023); Moradi-Sepahvand and Tindemans (2023); Moradi-Sepahvand and Tindemans (2024). An empty cell is a lead, not a gap.

Reading the closest paper: Hilbers, Brayshaw and Gandy (2023)

  1. First pass (5–10 min): Title, abstract, introduction, section headings, conclusions; glance at the references you already know. Out: The five Cs: category, context, correctness, contributions, clarity.
  2. Second pass (up to an hour): Figures, tables and the argument; skip the proofs. Mark the references to read next. Out: The extraction sheet, filled.
  3. Third pass, optional (4–5 hours for a beginner, outside this studio's 4 hours): Rebuild the work under the same assumptions, and challenge each one. Out: What you would do differently, and what it would change.

Its headings (open preprint, October 2022), for the first pass: 1. Introduction (capacity expansion planning; time series aggregation; inter-period links and storage) · 2. This paper's contribution · 3. Methods (overview; framework; remarks) · 4. Simulation studies (overview; setup; validation; example; discussion) · 5. Conclusions (conclusions; extensions).

The first pass, worked: the five Cs
Category
What kind of paper is it? A methods paper: a new aggregation framework, tested in simulations on a six-region test system.
Context
Which papers does it build on? Two from your matrix: Kotzur et al. (2018) for linked storage, and Teichgraeber, Küpper and Brandt (2021) for extreme periods taken from model results.
Correctness
Which assumptions will you check in pass two? A test system based on the IEEE six-bus system; demand and wind only, no solar; operation with perfect foresight.
Contributions
What is new? Periods chosen after a first model run that keep chronology for storage and keep the extreme periods that drive capacity; models, data and code released.
Clarity
Is it well written? Your judgement: note one thing you would ask the authors.
The second pass, worked: the extraction sheet
Question
Can choosing periods after a first model run cut the error of capacity plans built from long weather series, in models with storage?
Scale and boundary
A six-region test grid: baseload, peaking and wind generation, transmission and storage, built from scratch.
Data vintage
Hourly demand and wind for European countries, 1980–2017, from a published reanalysis-based dataset.
Method
Capacity-expansion optimisation on representative days, then an operations run over the full series.
Counterfactual
Three conventional schemes chosen from the input data alone and, on 3-year samples, the plan from the full series.
Sensitivity and validation
30 and 120 representative days; 40 resampled series; unserved energy checked over the full series.
Not modelled
Solar; operation without perfect foresight (named by the authors as an extension).
Code and data
Public, with a Zenodo DOI.

Worked from the open preprint (October 2022); the published version is at the DOI above.

Write the gap statement

Draft your gap statement: what you searched (where, when, strings), what the matrix shows, what the direct search found, what a second pass on the closest paper showed, and the question that remains.

Your own literature map: about 4 hours

  1. Write two concept blocks for your own question; run five logged strings in OpenAlex and one other open database, such as Semantic Scholar or Crossref (string, date, count). (30 min)
  2. Screen titles and abstracts down to about 30 candidates; note why you dropped the rest. (30 min)
  3. Find one review; snowball backward and forward from it; add what the searches missed. (30 min)
  4. Do a first pass on 12 papers (about 7 minutes each) and code them in a concept matrix with a written coding rule. (90 min)
  5. Search your empty cell directly and log the string, date and count. (15 min)
  6. Read the closest paper in a second pass, fill the extraction sheet, then write the gap statement. (45 min)

Optional, beyond the 4 hours: a third pass on the closest paper (4–5 hours for a beginner).

With JavaScript on, you build the strings yourself, predict each count, code four papers and test the empty cell step by step; your drafts stay in this browser.

Counts and citations: a snapshot of OpenAlex (open data, CC0) taken on 3 November 2026. Counts may change as OpenAlex updates its records: record the date and keep the snapshot.
Coding: the summaries and the matrix are the course's own first pass from each abstract; no abstract is quoted, and “not stated” means the abstract does not say.
Worked answers: the question and the answers on this page are illustrative.

Two answers to the same task

Weak answer · Developing

Representative periods are widely used, but few studies [1] consider storage and extreme events together [2]. This study is the first [3] to fill this gap [4].

  1. Falls short: No count, no database, no date: a reader cannot check it.
  2. Falls short: Undefined: linked between periods? Extremes chosen how?
  3. Falls short: A novelty claim the evidence cannot carry; a direct search finds at least three papers that do both.
  4. Falls short: Asserted, not shown: no matrix, no search of the cell.

Strong answer · Proficient

Search: OpenAlex titles and abstracts, 3 November 2026, five logged strings [1]; one review snowballed both ways [2]. Matrix: 12 papers coded from their abstracts by a written rule [3]; no abstract reports both a storage link and extreme periods [4]. Direct check: a search of that cell returned 23 works; by their abstracts, at least three do both [5] (Hilbers et al. 2023; Moradi-Sepahvand and Tindemans 2023, 2024). Closest work: a second pass on Hilbers et al. finds a six-region test grid with demand and wind only, no solar [6]. Remaining question: does choosing periods after a first model run still keep extremes and storage right once solar is added? The two other papers get a second pass before we claim it [7].

  1. Works: Database, date and strings: anyone can repeat the search and see what has changed since.
  2. Works: Search and snowballing find different papers; together they found all twelve.
  3. Works: The rubric's bar for positioning: at least 12 sources, coded with a stated rule.
  4. Works: Said as the coding allows: what the abstracts report, not what the papers lack.
  5. Works: The empty cell is tested, not assumed, and the gap as first stated does not survive.
  6. Works: The closest work is read, not just named: the extraction sheet supplies the distinction.
  7. Works: A narrowed gap with its evidence, held as provisional: Proficient. Strong needs the other two read and the gap to survive a reviewer's search.
Weak answer Strong answer
Evidence None given Search log, review, matrix of 12
The empty cell Assumed to be a gap Searched directly: 23 works, at least 3 do both
After the check "The first" No novelty claim; a narrowed question, tied to the closest work's second pass
Search
Synonyms inside a block with OR, blocks joined with AND, phrases in quotes; log the database, string, date and count.
Snowball
Backward through a review's references, forward through the works that cite it: each finds papers a search misses.
Matrix
Papers in rows, ideas in columns, coded by a written rule; an empty cell is a lead.
Test the cell
Search the empty cell directly before calling it a gap, then read the closest work in two passes.
Abstracts only
A first-pass matrix records what abstracts report; “not stated” is not “no”.
Go deeperWhy the course never calls a question “novel”

Few papers on a topic is not evidence that a question is new, important or feasible. It can mean:

  • the field uses other words;
  • the field is small;
  • the question was tried and dropped.

The rubric asks for a gap stated with evidence; its Strong level asks that the gap survive a reviewer's search.

The question studio (R10) shows how much has been written on a topic only as context, for the same reason.

Go deeperThe same workflow for other question types

The twelve papers here share one method family, optimization models. That is this example's choice, not the studio's: optimization is one family among several (R3).

Search, snowballing and the matrix work for every question type (R1). Only the matrix columns change, to fit the evidence each type uses. The questions below are illustrative.

Type Question Matrix columns
Descriptive How often did a region's wind output stay under a tenth of its capacity for a whole day? Data source and years · time resolution · how a low day is defined
Explanatory Why did members of some rural co-ops adopt rooftop solar faster than others? Data (survey, records, interviews) · method (statistics, case study) · setting
Evaluative Did a time-of-use tariff lower household peak demand, compared with a flat rate? Comparison group · design (trial, difference-in-differences, matching) · metered or surveyed
Predictive What will a city's summer peak demand be in ten years, and with what uncertainty? Forecast method · how uncertainty is shown · validation period
Qualitative (awareness level) How do households describe their winter heating choices? Who was asked, and how many · interviews or documents · how answers were coded
Go deeperWhat this studio leaves out
  • Systematic reviews with a registered protocol, two independent screeners and a flow diagram of every record: the next step when a review is itself the study.
  • Reference managers: any one works if it stores the DOI, your tags and the date you added each paper.
  • Grey literature (agency reports, grid-operator studies): search the agencies' own sites and log them the same way.

Check your own work against the rubric

Competence Developing Proficient (the bar) Strong
Positioning
Look for: Your matrix has at least 12 papers and a written coding rule; the empty cell was searched directly; the closest prior work is read in a second pass.		 A few sources, no map. Concept matrix of at least 12 sources; a stated gap with evidence. The gap survives a search by a reviewer; closest prior work engaged directly.
Go deeper · ResearcherFrom this map to your capstone proposal

The capstone research project starts from a typed question (R1) and needs this studio's matrix and gap statement in its proposal.

The closest prior work you found here becomes your reference point when you build and check your analysis (R4, then R5): a benchmark for a model, a comparison study for empirical work.

Sources and what’s next

Tools and data named on this page

Tool or source What it is for Link
OpenAlex An open index of research works, with a search page and an API; its data are CC0. This page's counts are a snapshot; its links open the same searches on the search page. openalex.org · snapshot of 3 November 2026
OpenAlex licence The terms of the index data (CC0). help.openalex.org · checked 2 November 2026
Semantic Scholar, Crossref Open databases for the second search in your log. Google Scholar is free to search but not open and has no API: log what you screened there. semanticscholar.org · search.crossref.org

Next

R3 · Which system, which boundary, which method?

Sources and further reading

Spotted an error, or stuck? Write to hlli@ksu.edu. Corrections are made in place and dated here.

Text licensed CC BY 4.0, code MIT: teachers and students may reuse and adapt it, with credit to Honglin Li.