<p>Continual learning aims to enable models to continuously acquire new knowledge without forgetting previously learned information—a fundamental yet challenging task in dynamic environments. Although recent memory selection strategies, such as those based on gradients and uncertainty, have advanced this field, they still suffer from sampling bias and limited generalizability across tasks. To address these limitations, we propose the Energy Alignment Sampling Strategy (EASS), an active continual learning framework that integrates energy-based modeling into both memory replay and active sample acquisition. The core principle of EASS lies in selecting memory and unlabeled samples based on their free energy, uncertainty, and diversity, and jointly training them using a dual-loss function that combines cross-entropy with free energy alignment loss to preserve old knowledge while adapting to new tasks. This approach helps mitigate the forgetting of old knowledge and boosts the efficiency of new knowledge acquisition. Our experimental results on the CIFAR-10 dataset demonstrate the effectiveness of our method. Additionally, we have applied our framework to two structural damage classification datasets—the Container Damage Dataset and the Building Structural Damage Dataset—with promising outcomes. This study advances the field of continual learning and highlights its practical applications in real-world scenarios, particularly in structural damage assessment.</p>