pvnet/models/multimodal/readme.md

Multimodal model architecture

These models fusion models to predict GSP power output based on NWP, non-HRV satellite, GSP output history, solor coordinates, and GSP ID.

The core model is multimodel.Model, and its architecture is shown in the diagram below.

This model uses encoders which take 4D (time, channel, x, y) inputs of NWP and satellite and encode them into 1D feature vectors. Different encoders are contained inside encoders.

Different choices for the fusion model are contained inside linear_networks.