This paper proposes an ABC-BiGRU model for estimating the State of Health (SOH) of lithium-ion batteries. The model combines the bidirectional gated recurrent unit (BiGRU) with the artificial bee colony (ABC) algorithm to enhance the accuracy of SOH prediction. Traditional methods for SOH estimation, such as physical and electrochemical models, are limited by their complexity and sensitivity to environmental factors. In contrast, the ABC-BiGRU model leverages the black-box characteristics of deep learning to analyze historical cycling data and extract health indicators (HIs) that significantly impact SOH. The model uses Pearson correlation analysis to select HIs and applies the ABC algorithm for hyperparameter optimization, improving the model's performance. The ABC-BiGRU model demonstrates superior predictive performance compared to other recurrent neural network models, with a maximum root mean square error (RMSE) of 0.016799317 and a mean absolute error (MAE) of 0.012626847. The model is validated using the NASA lithium battery cycling dataset, and experimental results show that the ABC-BiGRU model achieves higher accuracy and robustness in SOH prediction. The study highlights the effectiveness of the ABC-BiGRU model in accurately estimating the SOH of lithium-ion batteries, which is crucial for extending the lifespan of new energy vehicles and ensuring their safety. The model's ability to capture bidirectional dependencies in time series data and its efficient hyperparameter optimization through the ABC algorithm contribute to its superior performance. Future research will focus on improving the selection of HIs and exploring the impact of additional environmental factors on battery capacity.This paper proposes an ABC-BiGRU model for estimating the State of Health (SOH) of lithium-ion batteries. The model combines the bidirectional gated recurrent unit (BiGRU) with the artificial bee colony (ABC) algorithm to enhance the accuracy of SOH prediction. Traditional methods for SOH estimation, such as physical and electrochemical models, are limited by their complexity and sensitivity to environmental factors. In contrast, the ABC-BiGRU model leverages the black-box characteristics of deep learning to analyze historical cycling data and extract health indicators (HIs) that significantly impact SOH. The model uses Pearson correlation analysis to select HIs and applies the ABC algorithm for hyperparameter optimization, improving the model's performance. The ABC-BiGRU model demonstrates superior predictive performance compared to other recurrent neural network models, with a maximum root mean square error (RMSE) of 0.016799317 and a mean absolute error (MAE) of 0.012626847. The model is validated using the NASA lithium battery cycling dataset, and experimental results show that the ABC-BiGRU model achieves higher accuracy and robustness in SOH prediction. The study highlights the effectiveness of the ABC-BiGRU model in accurately estimating the SOH of lithium-ion batteries, which is crucial for extending the lifespan of new energy vehicles and ensuring their safety. The model's ability to capture bidirectional dependencies in time series data and its efficient hyperparameter optimization through the ABC algorithm contribute to its superior performance. Future research will focus on improving the selection of HIs and exploring the impact of additional environmental factors on battery capacity.