fivetuple/nilm_sig

Code for 'Using path signatures for non-intrusive load monitoring.'

Supporting code for PES conference paper, "Automatic feature generation for non-intrusive load monitoring using path signatures."

PJM 23-APR-21

---

results scripts

predict_reference_features.m - Replication - predict reference features from signature.

predict_cooll_labels_using_reference_features.m - Predict COOLL labels using reference features.

trainEnsemble.m - Output from classification app, amended.

trainKNN.m - Output from classification app, amended.

trainSVM.m - Output from classification app, amended.

reference_predictor_importance.m - Predictor importance analysis for reference features.

predict_cooll_labels_using_signatures.m - Predict COOLL labels using path signatures.

signature_predictor_importance.m - Predictor importance analysis for log signatures.

predict_cooll_labels_using_selected_reference_features.m - Replication - prediction using 5 features.

trainEnsembleSelected.m - trainEnsemble model for 5 features.

predict_cooll_labels_using_selected_signatures.m - Prediction using 7 signature features.

trainEnsembleSignatureSelected.m - trainEnsemble model for 7 features.

---

data creation scripts

save_reference_features.m - Generate and save Bruna Mulinari's proposed features.

save_signatures.m - Generate and save path signatures of trajectories.

---

auxilliary scripts

next_upward_zero_crossing.m - Returns the index of the next upward zero crossing in the vector.

errperf.m - Error metrics (amended from original).

errperf_license.txt - Licence file for errperf.

esig_shell.py - Python signature script.

matlab_esig_shell.m - MATLAB signature script.

---

directories

data - Data folder holding labels.txt, cooll_on_off_times.mat, and generated features.

V-I_trajectory - From https://github.com/brunamulinari/V-I_trajectory

---

Additional data and code needed if recreating features:

1. COOLL data - https://coolldataset.github.io/.

2. https://github.com/brunamulinari/V-I_trajectory.

3. Python esig package - https://pypi.org/project/esig/.

The training labels (./Data/labels.txt) are found from the Cool Config files (cooll/Configs/scenario1_*.txt) using,

grep Outlet1 * > ./labels_raw.txt

and from it generating labels.txt using a text editor. A first try with this approach generated some labels "Vacuum_cleaner_" owing to a space at the end of the line. These labels were corrected with a global search and replace.