flex_heat

This repository contains implementations of five algorithms: upper bound, basic control strategy, linear program, mixed-integer nonlinear program and reinforcement learning for exploring pipeline energy storage for combined heat and power (CHP) economic dispatch.

Modules

• ./data: Data module contains the data used in experiments. These data include heat demand, electricity price, season and time of the day. Data is divided on training and testing dataset.
• ./src:
  • ../data_processing: Creation of training and testing datasets by parsing the data into four separate datasets: heat demand, electricity price, season, and time of the day, and data discretization.
  • ../optimizers: The implementation of five optimization algorithms:
    • .../upper_bound.py: Creation of the upper bound on profit for CHP economic dispatch.
    • .../basic_strategy.py: Calculation of profit by following the control based on constant inlet temperature of 90°C.
    • .../abdollahi_2015.py: Linear program of CHP economic dispatch without grid dynamics.
    • .../li_2016.py: Implementation of mixed-integer nonlinear program for CHP economic dispatch. This implementation constains operation on day-ahead and real- time electricity market.
    • .../rl:
      • ..../q_agent.py: Implementation of Q-learning agent. Updates to Q-matrix based on environment information, and training and testing of Q-agent.
      • ..../environment.py: Environment of Q-learning agent. This environment generates the state, and reward based on observations from simulation environment.
      • ..../state.py: Two state spaces of Q-learning agent: full state space and partial state space.
  • ../simulator: The simulator is the property of collaborator company Flex-Technologies. The code will be made available with the technical report (manuscript in preparation).
• ./tests: Feasibility evaluation of outputs of optimizers by the simulator.
• ./results: This module stores experimental results of algorithms and simulator evaluation.
• ./plots: Replication of experimental plots. The code reads experimental results from ./results module, and generates plots related to algorithms' profit, feasibility, and optimization stability.
• ./util: util package contains mathematical constants, thermodynamics physics constants and district heating network parameters used in all experiments.

Installation

Mathematical optimizers use SCIP optimization suite. To install the util package:

python setup.py install

Usage

To create and solve mathematical models, and to train and test Q-learning algorithm:

python -m src.optimizers.main

To evaluate outputs of the optimizers by the simulator:

python -m tests.main

To replicate experimental plots:

python -m plots.main