Learning Hidden Battery States from Terminal Measurements

76 concepts53 papers1 proposal

To safely and efficiently operate modern electric vehicles and grid storage systems, a battery management system (BMS) must track critical internal metrics: State of Charge (SOC) (the remaining energy, akin to a fuel gauge) and State of Health (SOH) (the degree of irreversible degradation).

The fundamental challenge is that these states are completely hidden. You cannot insert a physical dipstick into a sealed lithium-ion cell to measure its remaining chemical reactants or track the degradation of its internal electrodes. Instead, a BMS can only observe a battery's external boundaries: the terminal voltage, the current flowing in or out, and the surface temperature.

Historically, engineers relied on simple current integration (Ampere-hour counting) or rigid, physics-based electrochemical models to estimate these hidden variables. However, current integration accumulates sensor noise over time, and explicit electrochemical models are notoriously difficult to parameterize under highly dynamic real-world conditions. To overcome these limitations, modern research has shifted toward data-driven architectures and adaptive filtering frameworks that learn to map easily observable terminal signals to unobservable internal states.


The Black-Box Challenge and Indirect Indicators

If we cannot measure internal degradation directly, how do we find a proxy for it? The solution lies in extracting Indirect Health Indicators (IHIs)—features derived from terminal measurements during standard operation that correlate strongly with capacity loss or resistance growth.

For instance, instead of relying on a flawed explicit model, Ayça Balkan et al. demonstrated in A Behavioral Algorithm for State of Charge Estimation that a battery's open-circuit voltage can be computed directly from terminal current and voltage measurements using a behavioral framework, bypassing complex parameter tuning entirely.

As batteries degrade, their thermodynamic and kinetic properties shift. These shifts manifest in observable patterns during charging or discharging:


Capturing Temporal Dependencies and Capacity Regeneration

Batteries possess "memory." Their current behavior is heavily influenced by past states, temperature history, and usage patterns. Furthermore, battery degradation is highly non-linear and frequently exhibits a phenomenon known as capacity regeneration, where a battery left to rest temporarily recovers a small portion of its available capacity, creating misleading spikes in the overall downward degradation trend.

To capture these complex temporal dynamics, researchers utilize deep learning architectures designed for sequence modeling.

[Raw Terminal Data: V, I, T] ──> [CNN: Spatial/Feature Extraction] ──> [BiLSTM + Attention: Temporal Memory] ──> [Accurate SOH/RUL Prediction]

By combining different neural network topologies, hybrid models can isolate distinct facets of the data. For example, Yukai Tian et al. developed a hybrid framework in State-of-Health Prediction of Lithium-Ion Batteries Based on CNN-BiLSTM-AM that uses a Convolutional Neural Network (CNN) to extract local features from voltage and temperature curves, a Bidirectional Long Short-Term Memory (BiLSTM) network to track long-term degradation over time, and an attention mechanism to weight the most critical features.

To handle the volatile noise introduced by capacity regeneration, Chunxiang Zhu et al. proposed a sequence-to-sequence model in Prognosis of Lithium-Ion Batteries’ Remaining Useful Life Based on a Sequence-to-Sequence Model with Variational Mode Decomposition. Their approach uses Variational Mode Decomposition (VMD) to deconstruct the raw capacity signal into high-frequency noise (the regeneration pulses) and a low-frequency trend (the true underlying degradation), allowing the underlying LSTM to learn the true lifespan trajectory without being misled by temporary chemical recovery.


Adaptive Filtering and State Co-Estimation

The ultimate realization of these models occurs in real-time, online applications. However, a major hurdle exists: SOC and SOH are deeply coupled. As a battery ages (SOH declines), its total capacity shrinks, meaning that a current draw of 10 Amperes will deplete an old battery much faster than a new one. If the BMS does not update its understanding of the battery's health, its "fuel gauge" (SOC) estimation will rapidly drift.

To solve this, modern frameworks use co-estimation, running parallel or dual-loop algorithms that continuously update the battery's parameters in real time.

┌────────────────────────────────────────────────────────┐
│                                                        │
│   ┌─────────────────┐        Updates        ┌──────┐   │
│   │  State of Health│ ────────────────────> │ SOC  │   │
│   │     (SOH)       │ <──────────────────── │ Loop │   │
│   └─────────────────┘   Corrects Capacity   └──────┘   │
│                                                        │
└─────────────────── Real-Time Battery BMS ──────────────┘

This coupling is effectively managed by combining machine learning with adaptive state observers like Kalman filters. In Co-Estimation of State-of-Charge and State-of-Health for Lithium-Ion Batteries Considering Temperature and Ageing, Xin Lai et al. designed a two-stage framework using an Unscented Kalman Filter (UKF). The algorithm dynamically updates the internal capacity parameters as the battery ages, maintaining an SOC estimation error below 2% even across extreme temperature swings and severe capacity decay.

Similarly, Enguang Hou et al. demonstrated in Combined State of Charge and State of Energy Estimation for Echelon-Use Lithium-Ion Battery Based on Adaptive Extended Kalman Filter that pairing an LSTM network (for parameter optimization) with an Adaptive Extended Kalman Filter (AEKF) allows the system to remain highly accurate even when dealing with the highly unpredictable degradation profiles of second-life, "echelon-use" batteries.

By shifting from rigid physical modeling to adaptive, data-driven feature extraction, these techniques allow a BMS to look inside the chemical black box. They transform noisy, external terminal measurements into highly accurate, real-time predictions of a battery's remaining energy and long-term lifespan.

Key Foundational Work

A new battery capacity indicator for lithium-ion battery powered electric vehicles using adaptive neuro-fuzzy inference system (2003) — contributed 2 concept elements to this direction.
A Behavioral Algorithm for State of Charge Estimation (2012) — contributed 2 concept elements to this direction.
History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_battery_remaining_useful_life_prediction_3194f7df,
  title = {battery remaining useful life prediction},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/battery-remaining-useful-life-prediction/rdr_3194f7df136740fa9a02b1ef8441c1d6},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-19},
}

This exact version (version 1, frozen text):

@misc{lacuna_battery_remaining_useful_life_prediction_3194f7df_v1,
  title = {battery remaining useful life prediction},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/battery-remaining-useful-life-prediction/rdr_3194f7df136740fa9a02b1ef8441c1d6/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-19},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake