Battery knowledge · State estimation
Data-driven SOC estimation
Estimating SOC with a model learned from data — LSTM neural networks, support vector regression, Gaussian processes — that maps histories of voltage, current and temperature to charge. It needs no physical model, but is only as good as its training data across temperature and aging, and is harder to verify for functional safety than a Kalman filter.
- Also called
- machine-learning SOC estimation, neural-network SOC estimation, support vector machine, SVM, LSTM
Example
State estimation: an example situation
Current-sensor bias accumulates in coulomb counting. Exact model SOC does not establish real estimator accuracy.
Read Data-driven SOC estimation alongside Coulomb counting in this context. Estimating state of charge by integrating the measured current over time. Simple and fast, but sensor offset accumulates, so it is corrected against the OCV–SOC curve at rest or fused with it in a Kalman filter.
Use in Lemonergy
Distinguish internal states from noisy measurement estimates.
This is a reference concept. No executable Lemonergy block has been verified for this term.
References & further reading
Further reading on this topic
These sources provide broader context; they are not presented as citations for every statement in this entry.
- Extended Kalman filtering for battery management systems of LiPB-based HEV battery packs: Part 3. State and parameter estimation
Gregory L. Plett · Journal of Power Sources · 2004
Topic background · Model-based battery state-of-charge and parameter estimation with extended Kalman filtering, including uncertainty and model dependence.
- Extended Kalman filtering for battery management systems of LiPB-based HEV battery packs: Part 2. Modeling and identification
Gregory L. Plett · Journal of Power Sources · 2004
Topic background · Battery charge balance, open-circuit voltage, ohmic loss, polarization dynamics and model identification; parameters require cell-specific data.