To maximise the service life of costly traction batteries and gain customer confidence in battery electric vehicles, long-term monitoring of internal battery states is essential. Since these states cannot be measured directly, estimation methods are required. While the state of health (SOH) reflects the battery’s remaining service life, precise knowledge of the state of charge (SOC) is vital for optimal operation — affecting charging strategies, cell balancing, range estimation, and thermal management. This thesis proposes a virtual simulation test, derived from the conventional capacity test, offering a scalable approach for SOH estimation that does not interfere with vehicle operation. A digital battery twin, based on a recurrent encoder-decoder network architecture, is used in place of the physical battery. This novel model learns long-term battery behaviour solely from measurable signals. Due to transfer- and meta-learning existing digital twins can be tailored to specific battery characteristics using a few logged driving cycles. Accurate SOC estimation for modern high-energy cell chemistries requires modelling open-circuit voltage (OCV) hysteresis, with its pronounced directionality and complex trajectories. Existing models often fall short in this regard, whereas the developed Trajectory Correction Hysteresis (TCH) model succeeds. Its associated transfer-fit method ensures practical, data-efficient application. After advanced ageing of the battery, the TCH model can be effectively adapted using only a few OCV/SOC points along the envelope curves. The proposed SOC correction framework enables robust OCV-based SOC estimation under the influence of hysteresis. Reliable SOC error correction and accurate hysteresis modelling in uncertain real-world conditions are achieved only through the combined application of the Trajectory Correction Hysteresis model (TCH), the transfer-fit method, and the SOC correction framework.