readme.md

Time Series to Image Converter for Energy Data

Notebook uses pyts to transform time series image data to:

• Gramian angular summation fields
• Gramian angular difference fields
• recurrence plots

For easier data handling NILMTK was used.

The converter offers a variety of parameters to set such as:

• input time series window size
• output image size
• resample period
• stacking n images as a video
• adding brightness to the GAF images
• choosing between Gramian angular fields or recurrence plots
• using only a selected building
• manually selecting the appliances
• saving the source time series

environment

Works with all datasets supported by NILMTK. Tested on:

• REFIT
• UKDALE
• iAWE
• REDD
• ECO

Example Output

Examples below are Gramian angle summation filed images transformed using UKDALE dataset over a 60-minute window.

Computer monitor and washing machine examples:

Examples for selected appliances:

Install Instructions

❗️ If possible, install on a Linux machine.

1. Install Anaconda by following instructions here.

2. Clone this repository and change directory to conda:

       git clone https://github.com/jenkoj/ts2img && cd ts2img/conda

3. Create a new environment by running:

       conda env create --name ts2img --file=ts2img.yml 

4. Activate the newly created environment:

       conda activate ts2img

5. Get hold of a dataset converted to NILMTK format or convert your dataset.

   • Check if your dataset has supported converter link, if not you can write your own dataset converter by following instructions here.

   • Convert your dataset by following notebook here.

6. Place converted dataset in datasets directory.

7. Set parameters.

8. Finally, run:

       ipython -c "%run converter.ipynb"

When adjusting the parameters start with iAWE. Since it is small, it is easy to handle.