An accurate estimation of the Remaining Useful Life (RUL) is crucial for ensuring operational safety and mission success. This study proposes a hybrid deep learning approach that combines Convolutional Neural Networks (CNN) and Neural Networks (NN) to predict the RUL of batteries. The model was trained and evaluated on a telemetry dataset across two years of satellite operation of a nickel–hydrogen (NiH \(_2\) ) battery, including various operational modes (charging, discharging, and idle). The proposed CNN+NN framework was benchmarked against several standalone models, including XGBoost, Support Vector Regression (SVR), CNN, NN, and Long Short-Term Memory (LSTM), as well as hybrid LSTM-based combinations (LSTM+CNN, LSTM+NN, and LSTM+XGBoost). The hybrid CNN+NN model consistently outperformed the comparative approaches in terms of predictive accuracy and stability. To further evaluate generalization capability, the model was validated on an additional NiH \(_2\) battery from the same mission and subsequently tested on the NASA lithium-ion battery dataset (B0005, B0006, B0007, and B0018). Moreover, a combined dataset evaluation using Leave-One-Battery-Out (LOBO) cross-validation was conducted to assess robustness across batteries. The proposed approach achieved competitive MAE and strong \(R^2\) values across both nickel–hydrogen and lithium-ion datasets, demonstrating reliable cross-chemistry generalization and stable forecasting performance. This study highlights CNN+NN as a promising contribution to space battery prognostics, enabling reliable RUL estimation for LEO remote sensing satellites.
