Deep Learning-based IndustrialNILM

This repository provides tools for preprocessing and modeling power consumption data for Non-Intrusive Load Monitoring (NILM) using deep learning. It is tailored for industrial environments where multiple machines of the same type operate concurrently.

📦 Modules

data_extraction.py

This module handles data preprocessing:

• Loads datasets in NILMTK format (.h5 / .hdf5)
• Aligns aggregate and appliance signals to a common sampling rate
• Extracts only overlapping time intervals
• Outputs a .csv file with the following columns:
  • timestamp
  • aggregated_signal
  • machine1
  • machine2

Note: machine1 and machine2 are two instances of the same machine type, as the dataset includes duplicate machines per category.

main.py

This module manages the modeling pipeline:

• Loads the preprocessed CSV data
• Applies windowing:
  • Sliding windows for the training set
  • Non-overlapping windows for the testing set (validation used as test in this example)
• Initializes the selected deep learning model
• Supports both training and inference, controlled via a configuration flag
• Implements several architectures inspired by residential NILM literature:
  • CNN
  • CRNN
  • TCN
  • WaveNet
  • BERT
  • LSTM
• Evaluation metrics:
  • Mean Absolute Error (MAE)
  • Signal Aggregated Error

📁 Dataset

This tool is designed to work with the following dataset:

**[Industrial Machines Dataset for Electrical Load Disaggregation](https://ieee-dataport.org/open-access/industrial-machines-dataset-electy:

Requirements.txt is the environment for processing the data, with the libraries required by NILMTK. Architecture for WAVENET (https://github.com/picagrad/WaveNILM) and BERT (https://github.com/Yueeeeeeee/BERT4NILM) integrate original codes and they refer to https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=8682543 and https://dl.acm.org/doi/10.1145/3427771.3429390 respectively; TCN derives from https://ieeexplore.ieee.org/document/8911216, LSTM from https://arxiv.org/abs/1507.06594, CNN from https://dl.acm.org/doi/abs/10.5555/3504035.3504353 and CRNN from https://ieeexplore.ieee.org/document/9831435.