Deep Learning-Based Approach for Heat Transfer Efficiency Prediction with Deep Feature Extraction.

Area of Science:

• Boiler engineering
• Predictive maintenance
• Artificial intelligence in industry

Background:

• Ineffective fixed soot blowing schedules lead to reduced boiler heat transfer and safety risks.
• Accurate ash accumulation prediction is crucial for optimizing boiler operations.
• Current methods lack the precision needed for dynamic industrial environments.

Purpose of the Study:

• To develop a high-precision ash accumulation prediction model for industrial boilers.
• To improve heat transfer rates, energy efficiency, and operational safety.
• To enable predictive maintenance for boiler heated surfaces.

Main Methods:

• Established a dynamic fouling model and a clearness factor (CF) health index.
• Employed Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) and Kernel Principal Component Analysis (KPCA) for deep feature extraction and multiscale time series analysis.
• Utilized an adaptive sliding window and an encoder-decoder architecture with an attention mechanism (EDA) for enhanced time series information mining.

Main Results:

• The proposed deep learning model demonstrated superior prediction accuracy and reduced training time compared to Long Short-Term Memory (LSTM).
• Validation on a 300 MW boiler dataset confirmed the model's effectiveness in multistep and different starting point predictions.
• The model successfully performed predictive maintenance tasks on economizer datasets under variable working conditions.

Conclusions:

• The fused deep learning model offers a significant advancement in ash accumulation prediction for industrial boilers.
• The approach provides valuable operational guidance for enhancing heat transfer, saving energy, and reducing consumption.
• This predictive maintenance strategy contributes to safer and more efficient industrial operations.