Related Experiment Video
Updated: Dec 27, 2025
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
938
A Dual-Attention Recurrent Neural Network Method for Deep Cone Thickener Underflow Concentration Prediction
Zhaolin Yuan1, Jinlong Hu1, Di Wu2
1School of Computer and Communication Engineering University of Science & Technology Beijing, Beijing 100083, China.
Sensors (Basel, Switzerland)
|March 1, 2020
Summary
This study introduces a dual attention neural network for predicting deep cone thickener underflow concentration, crucial for mining operations. The model enhances prediction accuracy by integrating spatial and temporal data with domain knowledge.
Area of Science:
- Chemical Engineering
- Data Science
- Industrial Process Control
Background:
- Deep cone thickeners are vital for industrial sedimentation processes.
- Accurate prediction of underflow concentration is critical for optimizing downstream mining operations.
- Existing time series models may not fully capture complex spatial and temporal dynamics in thickener data.
Purpose of the Study:
- To develop and evaluate a novel dual attention neural network for predicting underflow concentration in deep cone thickeners.
- To improve the accuracy and reliability of underflow concentration predictions by modeling both spatial and temporal features.
- To incorporate domain knowledge through supplementary features to enhance predictive performance.
Main Methods:
- A dual attention neural network architecture comprising an encoder and decoder was implemented.
- The model processes multi-sensor data, capturing spatial and temporal dependencies.
- Supplementary features derived from domain knowledge were integrated with raw sensor data.
Main Results:
- The proposed dual attention model demonstrated superior prediction accuracy compared to other time series models, reducing common error indices by over 10%.
- Ablation studies indicated that enhanced features reduced fitting error by approximately 5%, while dual-attention modules reduced error by approximately 11%.
- The model's feasibility and efficiency were validated using industrial case data from an FLSmidth Tailings Thickener within an Industrial Internet of Things (IIoT) framework.
Conclusions:
- The dual attention neural network is an effective method for predicting deep cone thickener underflow concentration.
- Integrating domain knowledge and dual attention mechanisms significantly improves prediction accuracy.
- This approach offers a promising solution for optimizing industrial sedimentation processes and subsequent mining operations.
