Symmetric spatiotemporal learning network with sparse meter graph for short-term energy-consumption prediction in...

Area of Science:

• Manufacturing Systems Engineering
• Artificial Intelligence
• Energy Management

Background:

• Short-term energy prediction is crucial for manufacturing operations like anomaly detection and energy saving.
• Existing single-node prediction methods face challenges with parameter collection and modeling.
• Current multi-node prediction methods lack sufficient knowledge guidance and struggle with complex spatiotemporal relationships.

Purpose of the Study:

• To develop an advanced method for accurate multinode energy-consumption prediction in manufacturing systems.
• To improve prediction performance by incorporating process knowledge and addressing spatiotemporal complexities.
• To provide a robust framework for real-time energy management and control.

Main Methods:

• Introduction of a Sparse Meter Graph (SMG) to represent process knowledge, including production nodes and material/energy flows.
• Development of a Symmetric Spatiotemporal Learning Network (SSTLN) using graph convolutional networks (GCN) and gated linear units (GLU).
• SSTLN is designed to balance spatial and temporal feature extraction, detail capture, and noise suppression.

Main Results:

• The proposed SSTLN-SMG method achieved lower prediction errors in multinode energy consumption compared to state-of-the-art methods.
• Experimental validation was conducted using datasets from an aluminium profile plant.
• The method demonstrated superior performance over approaches using deformed meter graphs or learning networks.

Conclusions:

• The SSTLN-SMG provides a significant advancement in multinode energy-consumption prediction accuracy.
• The integration of process knowledge via SMG effectively guides the SSTLN in capturing complex spatiotemporal dynamics.
• This approach offers a promising solution for enhancing energy efficiency and operational control in manufacturing environments.