kth-nek5000/deepturbulence ? reverse-engineered prompt

Build me a small research project that can generate data from the Moehlis nine equation turbulence model and use that time series to train neural networks that predict the future evolution of the nine coefficients.

I want scripts for creating a single trajectory, generating many trajectories into one dataset, and simple plotting or visualization so I can inspect the amplitudes and flow fields. Then add training for two model types, a basic feedforward network and an LSTM, plus a prediction workflow that takes an initial sequence and forecasts new time steps.

Keep it easy to run and easy to change the number of samples, sequence length, and training settings. If you need to check current library docs online, do that. Also include a small example or default setup so I can get from data generation to training to prediction without a lot of extra setup.