gpu_se

Contains GPU accelerated state estimators, an MPC implementaton, and a bioreactor model. This code forms the practical part of my masters thesis.

Instillation

The following steps are a guide to getting the code running. Version numbers shown in brackets, are the version numbers used for development. Other versions might work as well, but have not been tested. Where exact versions are needed, this is explicitly stated.

1. Install conda (v4.8.3)

2. Install the CUDA 10.2 toolkit (Note: the exact version is needed)

3. Clone this repository to your computer

   git clone https://github.com/darren-roos/gpu_se

4. From within the gpu_se directory, clone the cache repository

   git clone https://github.com/darren-roos/picklejar

   Ensure that it is in the gpu_se directory:

   gpu_se
   |
   +--picklejar\
   

   The results available in the cache are from runs performed on a machine with an AMD Ryzen 5 2400G, 32 GB of RAM and a GeForce GTX 1070 (8 GB of on board memory). The machine was running Ubuntu 18.04.4 LTS with a Python 3.8.2 environment. Further details about the Python environment can be found in the environment.yml file in the source code repository.

5. From within the gpu_se directory, create the conda environment:

   conda env create -f environment.yml

6. Add the gpu_se directory to your PYTHONPATH

Running

1. Activate the conda environment

   conda activate gpu_se_cuda102

2. Add gpu_se to the PYTHONPATH environment variable. On linux systems the following command should work if you are in the gpu_se directory:

   export PYTHONPATH=$PYTHONPATH:$pwd

3. Run any script using:

   python path/to/script

The following scripts produce the results found in the thesis document:

1. Open loop bioreactor

   • results/bioreactor_openloop/batch_production_growth - Shows the batch, production and growth phases of the bioreactor
   • results/bioreactor_openloop/ss2ss - Shows the open loop transition between two steady states
   • results/bioreactor_openloop/step_tests - Shows open loop step tests of the system

2. Closed loop bioreactor only

   • results/bioreactor_closedloop/no_noise - Shows a closedloop simulation with no noise
   • results/bioreactor_closedloop/with_noise - Shows a closedloop simulation with noise
   • results/bioreactor_closedloop/performance_vs_control_period - Shows the effect of control period on performance
   • results/bioreactor_closedloop/mpc_run_seq - Benchmarks the MPC code

3. Open loop particle filter

   • results/pf_openloop/pf_run_seq - Benchmarks the run time of the openloop particle filter
   • results/pf_openloop/pf_power - Benchmarks the power consumption of the openloop particle filter

4. Closed loop particle filter

   • results/pf_closedloop/bioreactor_performance_pf - Shows the performance of the particle filter in a closed loop

5. Open loop Gaussian sum filter

   • results/gsf_openloop/pf_run_seq - Benchmarks the run time of the openloop Gaussian sum filter
   • results/gsf_openloop/pf_power - Benchmarks the power consumption of the openloop Gaussian sum filter

6. Closed loop Gaussian sum filter

   • results/gsf_closedloop/bioreactor_performance_pf - Shows the performance of the Gaussian sum filter in a closed loop

Building the docs

Complete documentation for the code can be built. It requires a working instillation of latexpdf. It requires the activation of the conda environment:

   conda activate gpu_se_cuda102

Navigate to gpu_se/docs and run:

make clean latexpdf

The required documentation will be found in docs/build/latex/gpuacceleratedstateestimators.pdf.