An online intelligent detection method for slurry density in concept drift data streams based on collaborative...

Area of Science:

• Industrial Process Monitoring
• Data Stream Analysis
• Machine Learning

Background:

• Slurry density detection models in industrial settings frequently experience performance degradation.
• Concept drift, a change in data distribution, is a primary cause of this degradation in dynamic industrial environments.

Purpose of the Study:

• To propose an intelligent detection method for slurry density in concept drift data streams.
• To enhance the accuracy and adaptability of slurry density detection models in industrial applications.

Main Methods:

• Utilized Gaussian process regression (GPR) with regularized stochastic configuration for initial model building.
• Implemented a sliding window-based online GPR for linear model parameter updates.
• Employed a forgetting mechanism for nonlinear model recursive updates, alongside network pruning and stochastic configuration for dynamic structure adjustment.

Main Results:

• The proposed method effectively mitigates concept drift by prioritizing recent data.
• Enhanced mechanistic and data-driven models improved dynamic relationship capture and reduced reliance on outdated information.
• Experimental results on industrial data demonstrated superior performance over existing algorithms in all density estimation metrics.

Conclusions:

• The developed intelligent detection method significantly improves slurry density detection accuracy in industrial concept drift scenarios.
• The approach ensures real-time detection and model adaptability through collaborative computing in industrial settings.
• This method offers a robust solution for maintaining high performance in dynamic industrial data streams.