Explore the project results
Project community · D6 preparation

OPTI-WIND-AI Community Forum

Discuss wind-farm modelling, turbine placement, numerical results and the next steps toward terrain-aware optimisation.

Read the project team’s starter discussions and share reproducible examples when posting opens. Public discussion stays separate from confidential proposals and customer work.

Community access

Checking account status…

Discussions

6 prepared starter topics · 6 published topics. Starter content is written by the KYAMOS team.

Welcome to the OPTI-WIND-AI project forum. This is a place for public questions about the wind-farm research software and its examples. Start with the project page for current results and limitations.

KYAMOS team ·

Computational fluid dynamics (CFD) calculates how air moves. A passing software check shows that a specific calculation behaves as expected; it does not establish accuracy at a real wind farm.

KYAMOS team ·

The development software can generate synthetic Gaussian hills, which are smooth, bell-shaped terrain examples with configurable size. Source experiments include slope-related forcing and roughness terms.

KYAMOS team ·

The 4, 8, 16 and 24 turbine demonstration uses a Jensen top-hat wake model: a simplified downstream region of reduced wind speed with a uniform cross-section. It uses flat terrain, a steady 8 m/s input, identical 5 MW reference turbines and…

KYAMOS team ·

The project explores reinforcement learning, where a program learns choices through trial and feedback, alongside regular, random, greedy and genetic layout methods. A two-worker IMPALA learning setup and bounded placement evaluator are dev…

KYAMOS team ·

Reporting Problems

Starter topic

For a reproducible problem report, tell us: what you did; what you expected; what actually happened; the software version; date and time; operating system and GPU; the smallest public example; and the exact error text.

KYAMOS team ·

All discussions

AI Placement Optimization

KYAMOS team · · Updated 20 Nov 2026

The project explores reinforcement learning, where a program learns choices through trial and feedback, alongside regular, random, greedy and genetic layout methods. A two-worker IMPALA learning setup and bounded placement evaluator are development components. A fair comparison gives every method the same inputs and evaluation budget, then tests on examples kept out of learning. Thirty-five OPTI test methods are recorded in the delivery evidence; passing software checks does not prove that a learned layout beats the alternatives. Use this topic for comparison design, reproducibility, operating limits and cases where a simple method is sufficient. A general AI performance advantage and terrain-aware real-site optimisation remain unestablished.

Replies (0)

Join this discussion

Checking account status…

Forum rules & privacy

  1. Be respectful. Keep discussion relevant to the project.
  2. Support factual claims with sources and state the inputs and limits of an example.
  3. Share only material you have permission to publish.
  4. Do not post passwords, personal contact details, customer data or confidential files. Use the private contact channel for proposals.
  5. Moderators may edit, hide or lock spam, abuse, unsafe disclosures and misleading claims. Contact KYAMOS to request a review.
  6. Distinguish software checks, numerical examples and results measured at a real installation.

How moderation works

KYAMOS administrators can edit, hide and lock discussions, review private reports, and suspend accounts. Hidden content and edit history remain private for moderation review. A named forum moderator has not yet been announced; public posting remains subject to the launch gate.

To report a concern or appeal an action, use Report content or info@kyamosmultiphysics.com. Include the public topic address. Do not post the report as a public reply.

What becomes public

Published topic titles, messages, generic author labels and publication times are public. Your account email and private reports are not displayed. Discussion records and a limited edit history are retained for the project record and moderation; privacy requests go to KYAMOS.

Read the forum privacy notice

D6 status: prepared for discussion testing and moderator assignment. This page does not certify completion or funder acceptance of the deliverable.