Application of Data-Driven Online Prediction and Localization Method for Weak Batteries in Energy Storage Stations

With the rapid advancement of renewable energy, energy storage stations are increasingly vital for balancing supply and demand and enhancing grid stability. In battery energy storage systems, variances in manufacturing and operational environments can lead certain battery units to underperform, known as weak batteries. This condition diminishes charge and discharge efficiency, shortens lifespan, and potentially introduces safety risks. Therefore, the precise identification and timely maintenance of weak batteries are essential. This paper presents a novel prediction and localization method for weak batteries by combining Deep Convolutional Neural Networks (DCNN) with ensemble learning and transfer learning. By leveraging long-term monitoring data, this approach reduces data dependency through transfer learning and enhances predictive accuracy via ensemble learning. Experimental results indicate that the proposed model effectively maintains high prediction accuracy on small datasets, significantly improving industrial application efficiency and reliability.