Causal-Based Spatio-Temporal Graph Neural Networks for Industrial Internet of Things Multivariate Time Series Forecasting

Spatio-temporal data forecasting is a challenging task, especially in the context of the Internet of Things (IoT), due to the complicated spatial dependencies and dynamic trends of temporal patterns between different sensors. Existing frameworks for spatio-temporal data forecasting often rely on pre-defined spatial adjacency graphs based on prior knowledge for modeling spatial features. However, these methods may not effectively capture the hidden connections between components of complex industrial systems. To overcome this challenge, this paper proposes a new approach called Causal-based Spatio-Temporal Graph Neural Networks (CSTGNN) for multivariate time series forecasting. The CSTGNN model uses a causality graph to discover hidden relationships between sensors and comprises three main modules: causality graph, temporal convolution, and graph neural network, to handle spatio-temporal data features effectively. Experimental results on industrial datasets demonstrate that the proposed method outperforms existing baselines and achieves state-of-the-art performance. The proposed approach offers a promising solution for accurate and interpretable spatio-temporal data forecasting.