juel33 / ev-sdg

Synthetic data generator for EV sessions data

EV-SDG

Electric vehicle (EV) - Synthetic data generator(SDG): Parametric models to generate synthetic samples of EV sessions data.

This repository has (i) Trained models that reflect real world dataset, which can be used to generate a synthetic sample of EV sessions, and (ii) Code to train SDG models from new data.

References

[1] Lahariya, Manu and Benoit, Dries and Develder, Chris (2020). Defining a Synthetic Data Generator for Realistic Electric Vehicle Charging Sessions. Proceedings of the Eleventh ACM International Conference on Future Energy Systems, pp. 406–407.

[2] Lahariya, Manu and Benoit, Dries and Develder, Chris (2020). Synthetic data generator for electric vehicle charging sessions: Modeling and evaluation using real-world data. Energies 2020, vol. 13.

Citation

@inproceedings{10.1145/3396851.3403509,
author = {Lahariya, Manu and Benoit, Dries and Develder, Chris},
    title = {Defining a Synthetic Data Generator for Realistic Electric Vehicle Charging Sessions},
    year = {2020},
    publisher = {Association for Computing Machinery},
    doi = {10.1145/3396851.3403509},
    booktitle = {Proceedings of the Eleventh ACM International Conference on Future Energy Systems},
    pages = {406–407}
}

@article{lahariya2020synthetic,
    title = {Synthetic data generator for electric vehicle charging sessions: Modeling and evaluation using real-world data},
    journal = {Energies},
    author = {Lahariya, Manu and Benoit, Dries and Develder, Chris},
    volume={13},
    number={16},
    pages={4211},
    year={2020},
    doi = {10.3390/en13164211},
    publisher={Multidisciplinary Digital Publishing Institute}
}

Developed at UGent, IDlab.

EV sessions data generation

A sample of EV sessions data can be generated using the python script SDG_sample_generate.py. Generate_sample() can be used to generate and save samples to /res/generated samples.

from modeling.generate_sample import generate_sample

AM,MMc,MMe = SDG[0],SDG[1],SDG[2]   # SDG = synthetic data generator

generate_sample(AM=AM, MMc=MMc, MMe=MMe,
                horizon_start=horizon_start,horizon_end=horizon_end)

AM, MMc and MMe are the SDG models that needs to be passed as inputs. Models fitted on a real world data will be used as default SDG models for generation. These models are saved in modeling/default_models/saved_models. Generated EV sessions data will be saved on the res/generated_samples folder. Please refer to SDG_sample_generate.py for further details.

Command line arguments for SDG_sample_generate.py

optional arguments:
  -h, --help            Show this help message and exit
  -start_date START_DATE
                        First date of the horizon for data generation format:
                        dd/mm/YYYY.
  -end_date END_DATE    Last date of the horizon for data generation format:
                        dd/mm/YYYY.
  -use USE              Which kind of models to use. "default" for using the
                        default models "latest" for using the lastest trained
                        models.
  -model MODEL          Modeling method to be used for modeling arrival times:
                        * AC for arrival count models;
                        * IAT for inter-arrival time models.
  -lambdamod LAMBDAMOD  Method to be used for modeling lambda:
                        * AC: has two options, poisson_fit/neg_bio_reg;
                        * IAT: has three options, mean/loess/poly.
  -verbose VERBOSE      0 to print nothing; > 0 values for printing more
                        information. Possible values: 0, 1, 2, 3. (integer)

Default model specifications:

• SDG:

  • AM: Arrival models can be either inter-arrival time models (IAT) or arrival count (AC) models.

    • IAT models: possible modeling methods for lambda are (i) mean, (ii) poly, and (iii) loess.
    • AC models: possible modeling methods for lambda are (i) poisson_fit (ii) neg_bio_reg

  • MMc: Connected time mixture model.

    • Mixtures = GMMs
    • Optimization = EM (MLE)

  • MMe: Energy required mixture model.

    • Mixtures = GMMs
    • Optimization = EM (MLE)

Default models for all possible methods of modeling for AM are provided and can be used.

Model fitting

To train the models, a (real world) dataset is required. Following columns are necessary in this EV sessions data.

Column name	Description	Data format
Started	Starting date and time of the EV session	datetime (dd/mm/YYY HH:MM:SS)
ConnectedTime	Connection time of the EV session	Hours (float)
TotalEnergy	Requested energy of the EV session	kWh (float)
ChargePoint	Charging station	Categorical (str)

Preprocessing

We clean the data and prepare preprocessed data using preprocess module. The command line script SDG_preprocessing.py can be used to create the preprocessed datasets. Running this script will generate a 'slotted data' and a 'preprocessed data' file.

• Save the raw data file in a folder
• Run SDG_preprocessing.py (supporting module ./preprocess )
• Please see /res/preprocess for generated plots and pre processed data.

Command line arguments for SDG_preprocessing.py

optional arguments:
  -h, --help            Show this help message and exit
  -Year YEAR            Year for modeling (integer)
  -Slotmins SLOTMINS    Minutes in each timeslot (integer)
  -create_plots CREATE_PLOTS
                        Indicator for creating plots
  -Sessions_filename SESSIONS_FILENAME
                        Name of the file contaning raw data. This file must be
                        present in the res_folder (str)
  -res_folder RES_FOLDER
                        Location for raw data file. Default is "./res". Inside
                        this directory EV session files must be present. 
                        (string)
  -verbose VERBOSE      0 to print nothing; > 0 values for printing more
                        information. Possible values: 0, 1, 2, 3. (integer)

NOTE: Please do not forget to give the file name and file location while calling this script

NOTE: This file should be run before ./SDG_fit.py (used for training models)

NOTE: Don't forget to install the packages in requirements.txt (pip install -r requirements.txt)

Training SDG models

SDG_fit.py can be used to fit the SDG models.

Command line arguments for SDG_fit.py

optional arguments:
  -h, --help            Show this help message and exit.
  -model MODEL          Modeling method to be used for modeling arrival times:
                        * AC for arrival count model;
                        * IAT for inter-arrival time model.
  -lambdamod LAMBDAMOD  Method to be used for modeling lambda, depending on MODEL: 
                        * AC: has two options, poisson_fit/neg_bio_reg;
                        * IAT: has three options, mean/loess/poly.
  -verbose VERBOSE      0 to print nothing; > 0 values for printing more
                        information. Possible values: 0, 1, 2, 3. (integer)