Performance Prediction Models with Improved Accuracy and Generalizability for Organic Cathode-Active Materials of Lithium-Ion Battery

Development of organic energy storage requires enhancing performances of active materials. In particular, reaction potential and specific capacity of cathode-active materials have significant impact on energy density of organic lithium-ion battery. However, discovery of new compounds for active materials based on professional experience and intuition meets the limitation of huge search space of organic molecules. The performance predictors enable efficient discovery of new potential compounds. Although the predictors of potential, capacity, and energy density (models G1) are prepared in the previous work, these become older and have problems. In the present work, the updated models G2 have been constructed to improve the accuracy, usability, and generalizability. The models G2 are prepared by sparse modeling for small data combining machine learning and chemical insight on the training data set with adding new data. The updated models are validated using a new test data set and data-scientific methods. The improved predictors contribute to efficient exploration of new cathode-active materials to realize high-performance batteries.