Research On The Performance Of Functional Liquid Electrolyte Based On Data-driven Method

With the decrease in energy resources and the worsening air pollution caused by traditional fuel vehicles,electric vehicles are becoming increasingly popular.Currently,lithium-ion batteries have been widely used as power sources for electric vehicles and other electric tools.However,to meet the application requirements of electric vehicles,such as long endurance,long cycle life,high rate capability,and adaptability to low-temperature extreme environments,further improvement of the energy density,cycle life,and rate capability of lithium-ion batteries is necessary.Traditional experimental methods are both time-consuming and expensive.To improve the research and development efficiency of lithium-ion batteries,this project employs machine learning methods to study liquid electrolytes.In this paper,based on high-throughput experimental design,high-throughput data acquisition,database construction,and machine learning analysis and prediction,a data-driven solution for functional liquid electrolyte performance optimization screening is proposed,and the specific research contents are as follows:Firstly,a high-throughput liquid electrolyte preparation apparatus and data acquisition system were established.The system can quickly prepare 20 different formulations of liquid electrolytes and collect performance data.Furthermore,a battery design database was created,which can store,add,modify and provide data of liquid electrolyte formulations and corresponding battery data for machine learning algorithms to use.Secondly,over 300 low-temperature liquid electrolyte formulations were prepared and the basic physical properties of the electrolytes were collected through high-throughput experiments.The data of self-measured low-temperature liquid electrolytes were preprocessed and imported into the database.Four machine learning models（Support Vector Regression,XGBoost,Random Forest,and Multi-Layer Perceptron）were constructed,and the data in the database were used for model training,new formulation recommendation,and optimization verification.After comparison and validation,the Multi-Layer Perceptron model was found to have the best predictive performance,with a predicted accuracy for ionic conductivity with Mean Square Error（MSE）:0.37,Root Mean Squared Error（RMSE）:0.60,and R2:0.78,and for viscosity with MSE:0.01,RMSE:0.13,and R2:0.51.Furthermore,the predicted formulations by the Multi-Layer Perceptron model were experimentally verified,and the consistency of ionic conductivity was 88.9%and that of viscosity was 57.9%at-40℃,fully demonstrating the accuracy of the model.Finally,based on the study of low-temperature liquid electrolyte performance,and utilizing the high-throughput equipment and optimized multi-layer perceptron neural network,a recommended fast-charging liquid electrolyte formulation was proposed.The NCM811 button cell battery assembled with the recommended formulation（Li PF₆/Li TFSI/Li ODFB/DMC/EMC/EC/FEC/TMB）exhibited excellent long-cycle and fast-charging performance.Under 1 C/1 C charge-discharge current,the capacity retention rate was 75.28%after 400 cycles,and under 5 C/0.5 C charge-discharge current,the reversible specific capacity was 148.3 m Ah/g,with a capacity retention rate of 64.8%after 190 cycles.These results demonstrate the practicality and effectiveness of the data-driven functional liquid electrolyte performance optimization screening solution.