This repository contains the final submission code for Master's thesis: Developing a Reinforcement learning interface for partial differential equation control.
Define the following methods specific to each problem
obs_shape(): define the shape of your observation spaceaction_shape(): define the shape of your action spacephysics: instantiate the physics object located insrc/env/physics
build_obs(): calculate the observation vector in this method.build_reward(): define reward calculation in this method.
action_transform(): define the transformation according to the problem, e.g. if you want to applying actions only at certain parts of the domain.
experiments/burgers_equation_experimentsandexperiments/heat_equation_experimentscontains the PDE experiments for uncontrolled simulation, baseline agent, MPC agent and RL agent individually. The file ending in_evalcompares all the three agents using multiple random initial states.