EPREVAEngineer Paper
Research explainedExplainerElectricalEnergy

Predicting a battery's surface temperature from one charge cycle

A 2025 preprint fits a simple equivalent-circuit thermal model with ordinary least squares, using data from a single cycle — and tests it on NASA, MIT and Oxford battery datasets.

The paper explained here
Equivalent-Circuit Thermal Model for Batteries with One-Shot Parameter Identification (arXiv:2503.12616)
Myisha A. Chowdhury, Qiugang Lu · submitted 16 Mar 2025 · CC BY 4.0
This is Engineer Paper's own explanation, written from the open paper. The original is the primary source; read it before you rely on anything here.
a single battery cell (cylindrical or pouch) on a lab bench, thermocouple or temperature sensor attached to the cell surface, test rig wiring, circuit board or data logger, believable research-lab setting
AI-generated image

Battery management systems need to know how hot a cell is getting, but full electrochemical-thermal models are too heavy to run on a small microcontroller. Chowdhury and Lu propose a lighter route: treat the cell's heat flow like an electric circuit.

The model

The surface temperature T_s follows a first-order thermal circuit with thermal resistance R_T, thermal capacitance C_T, ambient temperature T_a and internal heat generation h:

dTs(t)dt=−1RTCTTs(t)+1RTCTTa(t)+1CTh(t)\frac{dT_s(t)}{dt} = -\frac{1}{R_T C_T} T_s(t) + \frac{1}{R_T C_T} T_a(t) + \frac{1}{C_T} h(t)
(1)

Discretised with a sampling step, it becomes a simple recursion, with ε set by the time constant:

Ts[k]=ϵTs[k−1]+(1−ϵ)Ta[k−1]+(1−ϵ)RTh[k−1]T_s[k] = \epsilon T_s[k-1] + (1-\epsilon) T_a[k-1] + (1-\epsilon) R_T h[k-1]
(2)

The heat generation is approximated from measurable quantities — current I, terminal voltage V and a polynomial in state of charge that stands in for the open-circuit voltage:

h=I[V−∑j=0nηjSOCj]h = I \left[ V - \sum_{j=0}^{n} \eta_j SOC^j \right]
(3)

The trick: make it linear in the parameters

Substituting (3) into (2) gives an expression that is linear in the unknown parameters, so they can be identified with ordinary least squares from one cycle of data — no iterative optimisation. The authors call this one-shot identification.

What they tested

The model was validated on public NASA, MIT and Oxford battery datasets under different degradation states and ambient conditions.

Try the physics

At steady state, (1) reduces to a temperature rise ΔT = R_T · h. Explore it:

Playground · Temperature rise from thermal resistance
solve for
Temperature rise
10 K
Plot temperature rise against
1,00000.01200 K/W (log)

For a worked, step-by-step version of this model with illustrative numbers, read our technical paper on a first-order thermal model of a battery cell.

Use it

Discussion

Technical questions and corrections are welcome. Comments are signed and checked before they appear.

No comments yet.

Sign in to add a typed review comment. Reviews are signed with your name.

More explained