Performance Prediction Of Lithium And Sodium Ion Cathode Materials Based On Machine Learning Algorithms

In the current energy storage battery system,the most widely used lithium batteries,low-cost lithium-ion batteries alternative to sodium-ion batteries,are of high economic value.Among them,the relationship between structural characteristics and performance of cathode materials is crucial to improve the electrochemical performance,based on which the cyclic discharge characteristic capacity of doped sodium-based nickel-manganese(NMMn)oxide cathodes,and the six properties of lithium electrode materials were investigated:Gravimetric Capacity(G＿C),Volumetric Capacity(V＿C),Stability Charge(S＿C),Stability Discharge(S＿D),Gravimetric Energy(G＿E),Volumetric Energy(V＿E).The current traditional experimental method is long in period and not wide in scope,and cannot predict the mapping relationship between structure and performance systematically.For this reason,the following work is done in this thesis to address the above problems.(1)In this thesis,two machine learning(ML)methods,including gradient-boosting machine(GBM)and random forest model(RF),are employed to analyze 112 data set samples of doped NMMn cathode materials.The initial discharge capacities(IC)and 50th cycle end discharge capacities(EC)of the electrode materials were predicted by using the basic system properties such as molar mass and crystal structure size.The results show that the GBM algorithm has higher accuracy in predicting IC and EC,with the root mean squared error(RMSE)of 23.21 m Ahg-1and 17.04 m Ahg-1,respectively,and the coefficient of determination(R~2)of 0.68 and0.60,respectively.The importance analysis of variables based on SHAP(SHapley Additive ex Planations)also showed that NMMn cathode material structures with specific sodium content(0.75<x<1.25),doping element content(x<0.2)and nickel content(x<0.4)were more likely to have higher IC and EC.Further studies showed that the role of nickel content in the 50th cycle end-discharge capacity of sodium ion batteries was higher.cycle end discharge capacity of sodium ion batteries than that of lithium ion batteries.These result parameters provide new guiding ideas for the design of high performance cathodes for sodium ion batteries.(2)A LightGBM(LightGradient Boosting Machine)based lithium-ion battery performance prediction algorithm is proposed using 2281 Li-ion battery datasets,and the relevant performance parameters are used to generate descriptors of the materials by the Matminer python package.The LightGBM used on the dataset was cross-validated ten times with R~2scores of 0.58,0.54,0.61,0.62,0.68,and 0.69 for weight capacity,capacity capacity,weight energy,volume energy,stability charge,and stability discharge predictions,respectively.Thus,machine learning can be an effective tool for predicting Li electrode battery performance.Finally,SHAP analysis was performed to identify the most important descriptors.The results show that density(2.0<x<5.0)mainly affects G＿C and G＿E;density atomic(8<x<15)primarily affects V＿C and V＿E,and secondarily affects G＿C and G＿E(8<x<15);while the main influences of S＿C and S＿D are energy＿above＿hull(0<x<0.2),C＿a(2<y<12)also has a significant effect on V＿C and V＿E.