DSL Masters / NRF Sprint
Research discussion feedback note
This note is a thinking guide from your research discussion. It is not a proposal draft, not a decision, and not something to paste into an AI tool. Use it to think before you rewrite.
How To Use This
Read it once, close the page, and answer the questions in rough bullets. The goal is to make your own reasoning clearer before the next version of your NRF application.
What you are trying to do
You are exploring smarter control of a home solar, battery, and grid system under scheduled outages. The useful research version is not simply to build another controller. It is to test whether a learning or optimisation method adds value over clear operating rules when future outages, solar variation, household demand, tariffs, and battery reserve are all part of the decision.
What is strong
The strongest foothold is the local problem. Scheduled outages create a different control setting from the reliable-grid assumptions used in many energy-management examples. That gives you a reason to build a simulation and benchmark control strategies instead of only describing a home-energy product.
Main issue
The idea still needs a sharper research centre. If the project is about reinforcement learning, you must explain why learning is needed and what it is compared against. If the better route is optimisation or model-predictive control, choose that honestly. The contribution must be the evidence from the comparison, not the fact that a controller exists.
Three thinking questions
- What exact decision is the controller making at each step, and what time horizon does it need to look ahead?
- Which baselines will you compare against so the method is tested fairly: simple rules, scheduled reserve rules, optimisation, or model-predictive control?
- What result would show that the added complexity is worth it, and what result would show that simple rules are enough?
What to do next
Bring rough bullets that define the simulated system, the data route, the control action, the baseline set, the metrics, and the contribution or claim you would be able to defend if the method works.