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

In response to the current issues of high cost, low accuracy, and inability to protect the original battery quality in electrolyte concentration measurement methods, a concentration prediction method based on RF coaxial probe and Levenberg-Marquardt Backpropagation neural network is proposed. First, a neural network prediction system for electrolyte concentration based on a coaxial probe is constructed. Then, it is compared with the Backpropagation (BP) neural networks optimized by the Bayesian regularized training method and BP neural networks optimized by the quantitative conjugate gradient method. The results show that this method has higher prediction accuracy. Taking zinc trifluoromethanesulfonate [Zn(CF3SO3)2] electrolyte as an example, a series of concentration measurement experiments are carried out using an RF coaxial probe to verify the method's feasibility. The experimental results show that the RF coaxial probe neural network concentration prediction method can effectively measure the electrolyte concentration, with a maximum relative error of only 4.42% and a maximum absolute error of 4.94%. These data indicate that the proposed prediction method has high accuracy and reliability for measuring water-based electrolyte concentration and can predict the electrolyte concentration in real-time.