In this second part, I’ll describe how to write the review document. In what follows, “collection” refers to your high-quality collection.
Second-level Review
The next step is a second, closer look at the papers in your collection. How close? You mostly want to understand what was done, and only a rough sense of how it was done. You are not attempting to understand the fine implementation details of the “hows.”
Capture this information for all of your papers by answering the seven key questions below. Your answers should be fairly brief—the review for each paper can be contained within a single page. There is definitely no need to reproduce any of the equations, tables, etc.
Start a new page for each paper. At the top of the page, display the title. At the end of the title, use the \cite command to include the bibliographic reference number (check that the paper then appears in the References section). Then answer these questions:
Key Questions
Author/Institution. This is meant to give you a feel for who is doing active research in the space.
Primary Problem. Give a succinct summary of the problem that’s being solved.
Approach. What was the general approach they used? Again, there is no need to state equations or exact algorithms; a verbal description is sufficient.
Author’s Assessment and Limitation of Other Works. In the paper’s introduction and literature review they will give an overview of the problem domain, prior research, existing gaps—sometimes specific flaws—in previous approaches. This is highly useful information.
Key Contributions. Give a succinct summary of what the authors’ claimed as the key contributions of the paper.
Author’s Statement of Value/Limitations. At the end of a paper, there will typically be a statement (admission) of limitations of the paper and suggested areas for future research. Again, this is highly valuable information.
My [Your] Assessment of Value/Limitations. What is your candid assessment of the value of the paper and its limitations?
You can see a representative example is below, or skip to the next part.
Example
This following example was taken (with modification) from Jarret Wendell’s literature review.
Stochastic Energy Management in Distribution Grids [1]
Author/Institution: G. Wang, G. Gainnakis (University of Minnesota-Minneapolis); V. Kekatos (Virginia Tech)
Primary Problem: Determine the lowest-cost energy management strategy in a power distribution system comprising grid, loads, solar panels and energy storage devices connected via smart inverters, where the smart inverters can accept real and reactive power commands from a central controller.
Approach: The authors argue that some elements in the system are stochastic in nature and that certain flexibilities can be exploited to admit a non-deterministic optimization approach. The stochastic variables are assumed to be stationary and ergodic. Grid constraints are assumed to allow deviation for short time periods, while longer-term averages are kept within constraints (i.e., a temporal separation that admits relaxation of short-term constraints). The stochastic problem is solved with a Stochastic Dual Sub-gradient Solver. The approach is evaluated numerically on a 56-bus distribution feeder system, and compared to a deterministic control strategy.
Authors Assessment and Limitation of Other Works: The authors claim that other (deterministic) solutions don’t fully exploit the system’s flexibility during short time frames.
Key Contributions: The key contribution is a proposed Ergodic Energy Management (EEM) algorithm that exploits the relaxation of grid constraints to optimize the power distribution by allowing factors (voltage) to exceed specifications for short periods of time.
Authors Value/Limitations: The authors plot the deterministic energy management algorithm against the ergodic energy management algorithm and show significant cost savings over a 4-5 hour period. No author comments are given on limitations of their work [note: this is rare].
My Assessment of Value/Limitations: They evaluated the approach against another method that was also developed by (at least one of) the authors. In my view, the performance should have been evaluated against known data or a validated deterministic algorithm. The assumption of ergodic stochastic variables may not be valid in practice, meaning that the relaxation of constraints may also be invalid.
References
[1] G. Wang, V. Kekatos, and G. B. Giannakis, “Stochastic energy management in distribution grids,” 2016 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Shanghai, China, pp. 3476-3480, March 2016.
Go to the next part: