The etch process plays a critical role in CD (Critical Dimension) control for patterning, and as technology nodes shrink, the importance of accurately predicting etch bias has escalated. Recent years have witnessed remarkable advancements in etch bias modeling research, with the emergence of various innovative methods, including machine learning and deep learning, for predicting etch bias. Nevertheless, the current machine learning approach remains an empirical model, relying heavily on the rationality of manually designed features. Deep learning methods, despite their popularity, often rely solely on Convolutional Neural Networks (CNNs) and lack a comprehensive integration of the fundamental principles of the etching process. To address these limitations, we introduce a novel loading effect adaptive etch bias model. This model uniquely determines its input range based on the specific characteristics of the pattern, enabling it better to capture the intricate nuances of the etching process. Compared to previous CNN-based models, our approach demonstrates a superior ability to perceive the loading effect by optimal FOV (Field of Vision), resulting in significantly higher accuracy in etch bias prediction. This innovative model not only enhances the precision of etch bias estimation but also contributes to the overall advancement of patterning technology.