GPU-Accelerated Direct Transcription-Based Nonlinear Model Predictive Control

This repository contains the source code used for the case studies presented in the manuscript "GPU-Accelerated Direct Transcription-Based Nonlinear Model Predictive Control" by Evelyn Gondosiswanto and Joshua L. Pulsipher. Note that running the GPU benchmarks require an NVIDIA GPU with CUDA support.

Building the repository

git clone --recursive https://github.com/infiniteopt/GPU-NMPC-paper
cd GPU-NMPC-paper
mkdir -p \
    distillation/logs \
    pde/logs \
    results/{Ipopt,MOI,ExaModelsCPU,ExaModelsGPU,MPCGPU}

Running the Julia Benchmarks

First, you'll need to install Julia (available at https://julialang.org/downloads/). Then to configure the required packages, create and instantiate a Julia environment using the Project.toml file as follows:

julia> cd("GPU-NMPC-paper/")

julia> ]

(@v1.12) pkg> activate .

(GPU-NMPC-paper) pkg> instantiate

The JuMP.jl and InfiniteExaModels.jl benchmarks are run using the run_cases_nmpc.jl file.

Case Study 1: Optimal Control of a Distillation Column

The source code for this case study is contained in:

• distillation.jl (JuMP.jl and InfiniteExaModels.jl)
• distill-OptimalControl.jl (OptimalControl.jl)

Case Study 2: Temperature Control of a PDE Heated Plate

The source code for this case study is contained in pde.jl.

Running the MPCGPU Distillation Benchmark

The source code for this case study is contained in the distillation/mpcgpu subfolder. As this GPU-based case study is written in CUDA C/C++, you will need to install CUDA Toolkit (see https://developer.nvidia.com/cuda-toolkit-archive) and a compatible C++ compiler.

To build and compile the code from the root directory, run the following command:

nvcc -std=c++17 \
    distillation/mpcgpu/distillation_pcg.cu \
    -Idistillation/mpcgpu \
    -IMPCGPU/include \
    -IMPCGPU/include/common \
    -IMPCGPU/include/dynamics \
    -IMPCGPU/include/pcg \
    -IMPCGPU/GLASS \
    -IMPCGPU/GLASS/include \
    -IMPCGPU/GBD-PCG \
    -IMPCGPU/GBD-PCG/include \
    -lcublas \
    -o distillation_pcg

Then the case study itself can be run using:

./distillation_pcg

The benchmark results are written to their respective folders in the results subfolder.