Research on neural network prediction method for electrolyte concentration based on RF coaxial probes

Area of Science:

• Electrochemistry
• Materials Science
• Sensor Technology

Background:

• Traditional electrolyte concentration measurement methods suffer from high costs, low accuracy, and potential damage to battery quality.
• There is a need for precise, non-destructive, and cost-effective methods for real-time electrolyte concentration monitoring.

Purpose of the Study:

• To develop and validate a new concentration prediction method for electrolytes.
• To address the limitations of existing electrolyte measurement techniques.
• To enable real-time monitoring of electrolyte concentration.

Main Methods:

• Construction of a neural network prediction system utilizing an RF coaxial probe.
• Implementation of the Levenberg-Marquardt Backpropagation (L-MBP) neural network algorithm.
• Comparative analysis against Bayesian regularized and quantitative conjugate gradient optimized Backpropagation (BP) neural networks.
• Experimental validation using zinc trifluoromethanesulfonate [Zn(CF3SO3)2] electrolyte.

Main Results:

• The proposed RF coaxial probe and L-MBP neural network method demonstrated superior prediction accuracy compared to other BP network optimization methods.
• Experimental measurements of zinc trifluoromethanesulfonate electrolyte concentration showed a maximum relative error of 4.42% and a maximum absolute error of 4.94%.
• The method proved effective for real-time electrolyte concentration measurement.

Conclusions:

• The RF coaxial probe combined with the L-MBP neural network offers a highly accurate and reliable approach for measuring water-based electrolyte concentrations.
• This prediction method is suitable for real-time applications, overcoming the drawbacks of conventional techniques.
• The study validates the feasibility and effectiveness of the proposed sensor system for electrolyte analysis.