Causal-Based Spatio-Temporal Graph Neural Networks for Industrial IoT Multivariate Time Series Forecasting

This repository contains the code for our paper:

Causal-Based Spatio-Temporal Graph Neural Networks for Industrial Internet of Things Multivariate Time Series Forecasting
✍ Authors: Amir Miraki, Austėja Dapkutė, Vytautas Šiožinys, Martynas Jonaitis & Reza Arghandeh
📚 Conference: Explainable Artificial Intelligence (xAI 2023), World Conference on Explainable AI
📖 Book Series: Communications in Computer and Information Science (CCIS, Volume 1903)

In this work, we introduce a framework that enhances the efficiency of spatio-temporal graph neural networks (GNNs) for Industrial IoT (IIoT) multivariate time series forecasting by integrating causality-based learning.

Our approach improves model interpretability by discovering hidden relationships between sensors, ensuring that deep learning decisions in industrial time series forecasting are both accurate and explainable. This is particularly crucial for predictive maintenance, anomaly detection, and process optimization in industrial settings.

---

📌 Key Features

✔ Causal-Based Learning: Incorporates a causality graph to uncover hidden sensor relationships.
✔ Spatio-Temporal Graph Neural Networks: Integrates graph-based spatial modeling with temporal convolution for improved forecasting.
✔ Explainability: Enhances interpretability by leveraging causal inference in spatio-temporal forecasting.
✔ Industrial Applications: Predictive maintenance, anomaly detection, and operational efficiency in Industrial IoT systems.

---

🧠 Model Architecture

Our framework consists of three main modules:

• Causality Graph Module: Discovers hidden relationships between sensors using causal inference.
• Temporal Convolution Module: Captures temporal dependencies in multivariate time series data.
• Graph Neural Network Module: Models spatial dependencies between sensors using graph neural networks.

---

📝 Citation

If you find this work useful, please cite our paper:

@inproceedings{miraki2023causal,
  title={Causal-Based Spatio-Temporal Graph Neural Networks for Industrial Internet of Things Multivariate Time Series Forecasting},
  author={Miraki, Amir and Dapkutė, Austėja and Šiožinys, Vytautas and Jonaitis, Martynas and Arghandeh, Reza},
  booktitle={Explainable Artificial Intelligence},
  pages={120--130},
  year={2023},
  publisher={Springer}
}