Electrification-driven circular economy with machine learning-based multi-scale and cross-scale modelling approach

Community energy systems, integrating electricity storage, smart transportation, and flexible energy interactions can mitigate renewable energy intermittency and uncertainty, and stabilize local grids. The battery circular economy, involving cascade use, reuse and recycling, aims to reduce energy storage costs and associated carbon emissions. However, developing multi-scale and cross-scale models based on physical mechanisms faces challenges due to insufficient expertise and temporal discrepancies among subsystems. This study proposes a general machine learning approach for a source-grid-demand-storage community with a battery circular economy, streamlining data preparation, model training, validation, and testing. Machine learning models replace traditional physical models for energy consumption, renewable energy supply, and battery cycling ageing, improving computational efficiency. Furthermore, a battery circular economy principle is proposed, which suggests reusing 80 % capacity retired electric vehicle batteries in buildings and recycling at 60 % capacity. The study validates machine learning-based surrogate models against traditional physical models. Results show an 82.5 % reduction in computational time, from 40 to 7 min without significant deviation in techno-economic-environmental performance (relative difference ≤3 %). This study offers a framework for machine learning in the 'source-grid-demand-storage' system within a battery circular economy, enhancing computation efficiency and accuracy for low-carbon transitions and informing multi-scale, cross-scale model advancement.