<p>The popularity of secondhand new energy vehicles (NEVs) has increased, and the rapid circulation of such vehicles has led to a demand for reliable methods to assess their residual value. Buyers and sellers in the used vehicle market are facing problems in determining fair pricing due to the lack of accurate evaluation models. Research intends to enhance an evaluation index and prediction model for the residual value of secondhand NEVs. The aim is to design a system that can precisely calculate residual values, thus making fair transactions in the second-hand NEV market possible. It includes three steps: the first step involves data source and dimension identification from used car trading websites; the second step includes cleaning the data by handling abnormal values, missing data, and feature engineering; and the third step is constructing a residual value evaluation index system, which involves destination, criterion, and indicator layers. It employs an integration model named CatBoost for extracting critical indicators, and the Bidirectional Long Short-Term Memory model (BiLSTM) is finally utilized to predict residual value. It proposed a hybrid CatBoost-BiLSTM approach, which uniquely combines categorical feature importance with temporal sequence learning capturing both static and dynamic patterns influencing NEV depreciation more effectively than existing models. Experimental results indicate that highest accuracy is realized when ample samples are available during the training process. The CatBoost-BiLSTM method is designed to provide reliable residual value predictions for used NEVs in the automobile industry, thereby fostering a sustainable and legitimate system for predicting used vehicle residual value.</p>