RPSPP: Recurrence Plot Spacial Pyramid Pooling Network for Appliance Identification in Non-Intrusive Load Monitoring

Implementation of high sample rate appliance identification using recurrence plots and a spacial pyramid pooling neural network.

A parameter free appliance identification algorithm for NILM using a 2D representation of time series known as unthresholded Recurrence Plots (RP) for appliance category identification. One cycle of voltage and current (V-I trajectory) are transformed into a RP and classified using a Spacial Pyramid Pooling Convolutional Neural Network architecture. The performance of this approach is evaluated on the three public datasets COOLL, PLAID and WHITEDv1.1.

The full publication can be found here: Parameter_Free_Recurrence_Plots.pdf

Preprocessing

1. Voltage-Current Zero Crossing Alignment
2. Select one cycle
3. Piecewise Aggregate Approximation (PAA)

Network Architecture

----------------------------------------------------------------
        Layer (type)               Output Shape         Param #
================================================================
   RecurrenceLayer-1        [-1, 1, 2000, 2000]               0
   RecurrenceLayer-2        [-1, 1, 2000, 2000]               0
         MaxPool2d-3          [-1, 2, 200, 400]               0
            Conv2d-4          [-1, 2, 200, 400]              38
              ReLU-5          [-1, 2, 200, 400]               0
         MaxPool2d-6           [-1, 2, 66, 133]               0
            Conv2d-7          [-1, 32, 66, 133]             608
              ReLU-8          [-1, 32, 66, 133]               0
            Conv2d-9          [-1, 32, 66, 133]           9,248
             ReLU-10          [-1, 32, 66, 133]               0
      BatchNorm2d-11          [-1, 32, 66, 133]              64
           Conv2d-12          [-1, 32, 66, 133]           9,248
             ReLU-13          [-1, 32, 66, 133]               0
      BatchNorm2d-14          [-1, 32, 66, 133]              64
           Linear-15                    [-1, 9]          24,489
================================================================
Total params: 43,759
Trainable params: 43,759
Non-trainable params: 0
----------------------------------------------------------------
Input size (MB): 0.02
Forward/backward pass size (MB): 81.98
Params size (MB): 0.17
Estimated Total Size (MB): 82.16
----------------------------------------------------------------

Requirements

Software is only tested on Linux using an NVIDIA GPU.

pip install -r requirements.txt

Datasets

• COOLL
• PLAID 2017
• WHITEDv1.1

COOLL Structure

Argument: --cooll-path

The expected directory structure:

- configs
 |- scenario1_1.txt
 |- scenario1_2.txt
 |- ...
- data
 |- scenarioC1_1.flac
 |- scenarioV1_1.flac
 |- ...
- appliances_and_action_delays.txt

PLAID Structure

Argument: --plaid-path

The expected directory structure:

- CSV
 |- 1.csv
 |- 2.csv
 |- ...
- meta_2017.json

WHITED Structure

Argument: --whited-path

The expected directory structure:

- AC_Electrolux_r5_MK2_20151031065948.flac
- AC_Electrolux_r5_MK2_20151031070257.flac
- ...

Usage

python -m rpspp --cooll-path=<path-to-dataset>

Multiple dataset path can be passed

python -m rpspp --cooll-path=<path-to-dataset> --plaid-path=<path-to-dataset>

Run as Docker container

Create .env file

LOG_PATH=/host/path/logs/
COOLL_PATH=/host/path/logs/cooll
PLAID_PATH=/host/path/logs/plaid
WHITED_PATH=/host/path/logs/whited

Build images

docker-compose build

Run RPSPP experiments for all datasets

docker-compose up rpspp

Run WRG experiments for all datasets

docker-compose up wrg

Results

Leave One Group Out

|Algorithm|COOLL|WHITED|PLAID| |---|---|---|---| | De Baets et al.** | N/A | 0.7546 | 0.7760 | | WRG (our result) | 0.4447 | 0.3954 | 0.8921 | | RPSPP | 0.5329 | 0.4310 | 0.8942 | ** De Beats results are as published by the authors

5-Fold

|Algorithm|COOLL|WHITED|PLAID| |---|---|---|---| | WRG (our result) | 0.8957 | 0.9984 | 0.8082 | | RPSPP | 0.9213 | 0.9924 | 0.8456 |

Citation

Please cite this work as:

@inproceedings{RPSPP_2021,
  author = {Wenninger, Marc and Bayerl, Sebastian P. and Maier, Andreas and Schmidt, Jochen},
  title = {Recurrence Plot Spacial Pyramid Pooling Network for Appliance Identification in Non-Intrusive Load Monitoring},
  booktitle = {20th IEEE International Conference on Machine Learning and Applications - ICMLA 2021},
  year = {2021},
}