Soft Sensor Modeling Method Based on Improved KH-RBF Neural Network Bacteria Concentration in Marine Alkaline...

Area of Science:

• Biotechnology
• Chemical Engineering
• Artificial Intelligence

Background:

• Marine alkaline protease (MAP) fermentation is a complex process with unmeasured parameters affecting protease quality.
• Online detection of critical parameters in MAP fermentation is challenging.
• Existing soft sensing models may have limitations in accuracy and computational efficiency.

Purpose of the Study:

• To develop a novel soft sensing model for online prediction of bacterial concentration and relative active enzyme in MAP fermentation.
• To address the difficulties in online detection of unmeasured parameters affecting MAP quality.
• To improve the accuracy and efficiency of soft sensing models for complex fermentation processes.

Main Methods:

• Development of an improved Krill Herd algorithm (LKH) incorporating Levy flight and enhanced location update formulas.
• Integration of the LKH algorithm with a Radial Basis Function (RBF) neural network (LKH-RBFNN).
• Application of adaptive RBF neural network algorithms and control laws to approximate unknown parameters.

Main Results:

• The LKH-RBFNN model achieved lower root mean square error (0.938) and maximum absolute error (0.569) compared to traditional KH-RBFNN and PSO-RBFNN models.
• The improved Krill Herd algorithm demonstrated enhanced global search ability and avoided local optimization.
• The optimized RBF neural network parameters reduced overcorrection and computational load.

Conclusions:

• The proposed LKH-RBFNN soft sensing model provides accurate online prediction of key parameters in MAP fermentation.
• This method effectively overcomes the limitations of online detection for complex fermentation processes.
• The LKH-RBFNN model offers a promising solution for improving the control and quality management of MAP fermentation.