This paper presents a novel approach to estimating the State of Health (SOH) of lithium-ion batteries using a hybrid model based on the Artificial Bee Colony (ABC) algorithm and Bidirectional Gated Recurrent Unit (BiGRU). The study addresses the challenges of traditional methods, such as low accuracy and high computational costs, by leveraging the black-box characteristics of deep learning models to capture intrinsic correlations in historical cycling data. The authors select health indicators (HIs) from battery cycling data that significantly impact SOH through Pearson correlation analysis. The ABC-BiGRU model is then applied to predict SOH, demonstrating superior performance compared to other recursive neural network models. The model's effectiveness is validated using the NASA battery cycle dataset, showing maximum root mean square error and mean absolute error of 0.016799317 and 0.012626847, respectively. The paper also discusses the advantages of the proposed model, including its ability to capture bidirectional dependencies in time series data and its robustness in handling complex non-linear systems. The experimental results highlight the model's stability and reliability in predicting SOH, making it a promising tool for improving battery management and safety in new energy vehicles.This paper presents a novel approach to estimating the State of Health (SOH) of lithium-ion batteries using a hybrid model based on the Artificial Bee Colony (ABC) algorithm and Bidirectional Gated Recurrent Unit (BiGRU). The study addresses the challenges of traditional methods, such as low accuracy and high computational costs, by leveraging the black-box characteristics of deep learning models to capture intrinsic correlations in historical cycling data. The authors select health indicators (HIs) from battery cycling data that significantly impact SOH through Pearson correlation analysis. The ABC-BiGRU model is then applied to predict SOH, demonstrating superior performance compared to other recursive neural network models. The model's effectiveness is validated using the NASA battery cycle dataset, showing maximum root mean square error and mean absolute error of 0.016799317 and 0.012626847, respectively. The paper also discusses the advantages of the proposed model, including its ability to capture bidirectional dependencies in time series data and its robustness in handling complex non-linear systems. The experimental results highlight the model's stability and reliability in predicting SOH, making it a promising tool for improving battery management and safety in new energy vehicles.