Physics-shaped room model learns hidden door leak from heating logs

The author's earlier attempt, where a language model fitted room physics from few readings, gave an unreliable heating plan. This follow-up builds a physics-shaped model with three weights that treat heat loss as proportional to the indoor-outdoor temperature gap, a fixed heater contribution, and a latent field capturing an unlogged leak. The model reuses the earlier training setup on 4,800 rows, with only small changes to the prediction error and gradient calculations. The second half searches for heater schedules against the model, progressing from exhaustive search to random guesses and then to a learned search method. A planner that replans at every step is also introduced and tested on an evening run against a thermostat. The code is in tools/heater_plan.ts.
This is an AI-generated summary. ShortSingh links to the original source for the complete article.
Discussion (0)
Log in to join the discussion and vote.
Log in