PSCAD_Fortran_RL_Program

A simple case for PSCAD-Fortran RL program

Why this?

A project of power system needs to be simulated in PSCAD, and it utilizes some advanced Refincement Learning algorithms to control the simulation. Since PSCAD does not provide a API that can some advanced languages like Python can directly use, so try to write RL algorithm in Fortran.

Basic structure

PSCAD can create a self-defined component, in which a like-Fortran code can be written to call a Fortran subroutine. The key is how to transfer the data between those softwares.

Document introduction

A typic second order system, refer to PSCAD_DDPG.pscx The main RL algorithm, refer to ddpg.f90

Environment setting

The multiple programing is based on PSCAD 4.6.3, Intel Fortran compiler, VS 2019.

Progress:

Almost run through all the procedures, however, for the training of actor network in DDPG algorithm, do not have a idea of how to update it since we do not have a auto-differentiation tool in this Fortran-based NN framework.

(Put this aside)

Credits: The neral network structure: Neural Fortran https://github.com/modern-fortran/neural-fortran