Deep-DeePC

Deep-DeePC is a data-enabled predictive control framework that leverages deep neural networks for real‐time greenhouse climate management. By learning the complex mapping from past measurements to future control actions, Deep-DeePC dramatically reduces the online computational load of traditional MPC schemes. To enforce physical and actuator constraints, we embed a differentiable quadratic programming (QP) layer directly into the end-to-end learning architecture.

In a numerical case study using outdoor climate data from Columbus, OH, we benchmark Deep-DeePC against a nonlinear MPC (NMPC) baseline and the standard DeePC algorithm. Our results on lettuce cultivation demonstrate superior climate regulation and highlight the approach’s scalability to other greenhouse crops and operating scenarios.

Repository Structure

• Deep-DeePC/

  • requirements.txt – Pinned package versions

  • README.md – Project overview

  • main_code.ipynb – Main code for simulation

    • GreenhouseParams() – Greenhouse parameters
    • CropClimateModel() – Nonlinear crop-climate model
    • DiffQPLayer() – Differentiable QP layer
    • DeepDeePC() – Deep DeePC framework
    • DeepDeePCDataset() – Deep DeePC dataset generation
    • DNNPredictor_LSTM() – DNN Predictor using LSTM
    • FigurePlotter() – Figure plotting

  • input/ – Input data

    • Colombus_OH_30days.pkl – DeeP DeePC training dataset using outdoor weather in Columbus, OH

  • figures/ – Figures

  • trained_model/ – Trained model

    • checkpoint_lstm.pth – DeeP DeePC training dataset using outdoor weather in Columbus, OH

Codes

This folder contains the Python scripts used in the study. Details of each file or folder is provided below:

Requirements

This project is designed for Python 3.9 and depends on the core scientific and optimization libraries:

• PyTorch
• NumPy
• CVXPY
• CVXPYlayers
• SciPy
• Matplotlib

All exact, tested versions are pinned in requirements.txt. To install them, simply run:

pip install -r requirements.txt

Citation

Please use the following citation when using the data, methods or results of this work:

Kim, J., You, F. Energy-Efficient, Crop-Aware Intelligent Control for Smart Greenhouse with Deep DeePC Framework. Submitted to Applied Energy.