World Model for Battery Degradation Prediction Under Non-Stationary Aging

Kai Chin Lim; Khay Wai See

cs.LGMar 11, 2026

World Model for Battery Degradation Prediction Under Non-Stationary Aging

Authors: Kai Chin Lim, Khay Wai See

Organizations: Independent Researcher, Australia

Abstract

Degradation prognosis for lithium-ion cells requires forecasting the state-of-health (SOH) trajectory over future cycles. Existing data-driven approaches can produce trajectory outputs through direct regression, but lack a mechanism to propagate degradation dynamics forward in time. This paper formulates battery degradation prognosis as a world model problem, encoding raw voltage, current, and temperature time-series from each cycle into a latent state and propagating it forward via a learned dynamics transition to produce a future trajectory spanning 80 cycles. To investigate whether electrochemical knowledge improves the learned dynamics, a Single Particle Model (SPM) constraint is incorporated into the training loss. Three configurations are evaluated on the Severson LiFePO4 (LFP) dataset of 138 cells. Iterative rollout halves the trajectory forecast error compared to direct regression from the same encoder. The SPM constraint improves prediction at the degradation knee where the resistance to SOH relationship is most applicable, without changing aggregate accuracy.

Figures & tables

Figure 1: World model architecture. Raw V/I/T time-series per cycle are encoded by a shared 1-D CNN, then processed by PatchTST to produce latent state

\mathbf{z}(k)

. The dynamics module rolls out future latent states and a shared head decodes SOH at each step. The physics constraint operates during training only.

Stage	SOH range
Stage 1	1.00–0.95
Stage 2	0.95–0.90
Stage 3	0.90–0.85
Stage 4	below 0.85

Table 1: Aging stage definitions.

Model	Rollout	Physics	MAE	RMSE	MAPE%	S1 MAE	S2 MAE	S3 MAE
PIWM	Yes	Yes	0.0063	0.0097	0.66	0.0056	0.0080	0.0185
WM	Yes	No	0.0063	0.0090	0.65	0.0054	0.0098	0.0135
CNN-PatchTST	No	No	0.0078	0.0112	0.81	0.0071	0.0103	0.0149

Table 2: Architecture ablation on the test set. All variants use single pass joint training. Lowest values per column are in bold.

Model	h5	h10	h20	h50
PIWM	0.0067	0.0072	0.0081	0.0109
WM	0.0065	0.0068	0.0075	0.0096
CNN-PatchTST	0.0136	0.0135	0.0133	0.0134

Table 3: Future SOH trajectory MAE at forecast horizons

h\in\{5,10,20,50\}

cycles ahead.

Figure 2: Per-cell SOH trajectory performance. Top row: three representative cells showing rollout advantage on cell 127, PIWM+EWC failure mode on cell 57, and short lifecycle limitation on cell 45. Forecast wicks in purple show 80 cycles ahead predictions from PIWM, 20 wicks per cell. Bottom: MAE heatmap for all five methods across all 14 test cells, sorted by degradation depth.

Phase	Validation MAE	Early stop epoch
Batch 1	0.023	18
Batch 2	0.057	18
Batch 3	0.014	20
Test	0.021	—

Table 4: PIWM + EWC batch-staged training progression. Validation MAE at the end of each batch phase.

Model	Training	MAE	RMSE	MAPE%	S1 MAE	S2 MAE	S3 MAE
PIWM	Joint	0.0063	0.0097	0.66	0.0056	0.0080	0.0185
PIWM + EWC	Batch-staged	0.0210	0.0377	2.19	0.0167	0.0413	0.0341
LSTM	Joint	0.0209	0.0286	2.20	0.0164	0.0344	0.0598

Table 5: Control tests compared with PIWM under joint training.

Panel of Figure 3 — Figure 3: PCA of latent vectors $\mathbf{z}(k)$ , coloured by SOH value. The rollout objective produces a unified SOH-ordered curve (b, c), while the encoder without rollout retains per-cell structure (a) and batch-staged training distorts the manifold (d).

Explore similar work

PiDDM: Physics-Informed Differentiable Degradation Modeling for Lithium-Ion Battery State-of-Health Prediction
Jul 31, 2026Zeping Chen, Ruda Jian, Sachin Sigdel +3RUL Estimation Physics-Informed ML
Department of Chemical and Biomolecular Engineering, University of Notre Dame, Notre Dame, IN, USA · Department of Mechanical Engineering, The University of Texas at Dallas, Richardson, TX, USA · Department of Aerospace and Mechanical Engineering, University of Notre Dame, Notre Dame, IN, USA +1
BatteryMFormer: Multi-level Learning for Battery Degradation Trajectory Forecasting
May 26, 2026Ruifeng Tan, Jintao Dong, Weixiang Hong +3Multivariate Time Series Forecasting RUL Estimation
School of Computer Science and Engineering, Central South University · Material Genome Institute, Shanghai University
Real-Time State-of-Health Estimation and Online Degradation Prognosis from Partial Battery Discharge Using Physics-Informed Neural Networks
Aug 14, 2026Begoña Ispizua, Serio Gil-López, Leire Arrizabalaga +1Lithium-Ion Batteries RUL Estimation
Tecnalia - Basque Research Technology Alliance (BRTA), Derio, 48160, Spain · Unibersity of the Basque Country (EHU), Leioa, 48940, Spain · Mondragon Unibertsitatea, Bilbao, 48001, Spain

Jul 31, 2026J/K move · Enter open · S save

PiDDM: Physics-Informed Differentiable Degradation Modeling for Lithium-Ion Battery State-of-Health Prediction

Zeping Chen, Ruda Jian, Sachin Sigdel, Guoping Xiong+2

Department of Chemical and Biomolecular Engineering, University of Notre Dame, Notre Dame, IN, USA · Department of Mechanical Engineering, The University of Texas at Dallas, Richardson, TX, USA · Department of Aerospace and Mechanical Engineering, University of Notre Dame, Notre Dame, IN, USA+1

RUL Estimation Physics-Informed ML Lithium-Ion Batteries

Open paper AskPDF arXiv

Accurate prediction of lithium-ion battery state of health (SOH) is essential for reliable energy storage operation. However, purely data-driven models may generalize poorly across cycling protocols and produce physically implausible behavior during long-term extrapolation. We developed a physics-informed differentiable degradation modeling framework (PiDDM) for battery SOH prediction. PiDDM incorporates empirical Arrhenius degradation kinetics associated with solid electrolyte interphase growth and loss of lithium inventory into the training objective, encouraging physically consistent capacity fade under diverse operating conditions. The framework was evaluated using a public dataset of 55 batteries cycled under six operating protocols. PiDDM achieved the lowest average prediction error among the evaluated models and substantially reduced mean squared error relative to a multilayer perceptron and a baseline physics-informed neural network. For extrapolation, the models were trained on the first 90% of each battery's cycle life and evaluated on the unseen final 10%. PiDDM captured accelerated end-of-life degradation while avoiding the nonphysical capacity regeneration produced by the baseline models. These results show that incorporating degradation physics into neural network training improves predictive accuracy and physical consistency, providing a promising approach for practical battery health monitoring.

Abstract

Explore similar work

PiDDM: Physics-Informed Differentiable Degradation Modeling for Lithium-Ion Battery State-of-Health Prediction

BatteryMFormer: Multi-level Learning for Battery Degradation Trajectory Forecasting

Real-Time State-of-Health Estimation and Online Degradation Prognosis from Partial Battery Discharge Using Physics-Informed Neural Networks