The decline in lithium battery capacity can compromise its safety and stability, emphasizing the importance of accurate capacity estimation for better decision-making. However, prevailing black-box data-driven models face challenges in safety-critical applications due to their lack of interpretability. Additionally, these models often rely on fixed operating conditions for feature extraction, limiting their suitability for real-world scenarios with variable conditions. To address these issues, this paper presents an enhanced adaptive neural fuzzy inference system (ANFIS) designed to accommodate random operating conditions. Firstly, the factors influencing capacity degradation are analyzed, and relevant features are extracted and refined from battery measurement data. Subsequently, an activation mechanism simplifies the system structure, while an attenuation coefficient is introduced to tailor the model to battery cell characteristics. Further refinement is achieved through continuous optimization of fuzzy clustering centers using an adaptive particle filter algorithm. Validation of the system is conducted using the NASA random walk battery dataset, resulting in a capacity estimation root mean square error (RMSE) of 3. 73% . Comparative analysis demonstrates that the proposed system offers superior accuracy and a degree of interpretability when contrasted with other methods.