Degradation forecasting for accelerated lifetime studies of organic photovoltaics via symbolic regression

As the stability of organic solar cells improves, traditional long-duration testing methods cannot provide timely feedback, retarding optimization and hindering commercialization. Therefore, accurate degradation forecasting becomes a critical challenge, as the complex, nonlinear degradation kinetics of OPV devices complicate prediction. Here, we present a machine learning pipeline based on Class Symbolic Regression (CSR) that accurately extrapolates device degradation ~10× beyond the measured window, enabling day-scale stability decisions. Our workflow achieves 3-4% test error in quantitatively forecasting lifetime using the first 20 h for preliminary screening (200 h) and the first 100 h for long-term stability testing (1,000 h). This approach, validated across >40 donor–acceptor systems and >100 processing conditions, demonstrates generalization across key stability metrics. Additionally, the workflow is lightweight, requiring few resources and training, making it suitable for various OPV laboratories, thereby compressing OPV stability-testing protocols from month-scale to days. This work accelerates decision-making, enhances R&D throughput, and advances commercialization in the OPV industry.