Capacity prediction of K-ion batteries: a machine learning based approach for high throughput screening of electrode materials

Abstract

Machine learning (ML) techniques have revolutionized the field of materials science in recent decades. ML has emerged as an excellent tool to accelerate the screening of electrode materials for alternative metal ion batteries, particularly K-ion batteries, which can outperform the conventionally used lithium-ion batteries with drawbacks of low abundance and high reactivity in air. Since specific capacity is an important metric to estimate the performance of a battery, hereby we attempt to predict the specific capacity for potassium battery electrode materials using ML based on compositional features for the first time. We have employed various ML models and Kernel Ridge Regression is identified as the most reliable model for our dataset, considering mean absolute percentage error as the performance metric. From the obtained specific capacity values, we have also determined the number of K ions that can be intercalated in the formula unit of considered electrode compounds. DFT calculations have been performed to confirm the stability of intercalated electrode materials. Our results show that the application of ML algorithms can circumvent the huge computational cost associated with DFT-based screening studies for identifying suitable electrode materials with high specific capacity, which is crucial for efficient battery technology.