Abstract
The accurate estimation of State of Charge (SOC) in electrochemical energy storage systems is critical for operational safety and efficiency. This paper presents a novel Cell Memory and Cluster Feature Fusion Network (CMCF) that addresses multi-cell SOC estimation challenges in battery clusters. Our method integrates Maximum Power Point Tracking (MPPT) principles to optimize energy harvesting during charging cycles, thereby enhancing SOC estimation accuracy. By leveraging memory mechanisms to capture cell-specific degradation patterns and cluster-wide operational dynamics, the proposed framework achieves fine-grained SOC predictions across heterogeneous battery cells. Experimental validation on real-world datasets demonstrates a 71.37% reduction in mean squared error compared to single-cell estimation approaches, while MPPT integration improves energy transfer efficiency by 18.2%.
1. Introduction
Modern energy storage systems rely on battery clusters comprising hundreds of series/parallel cells. SOC estimation faces two fundamental challenges:
- Cell Heterogeneity: Manufacturing variances cause divergence in capacity, impedance, and aging rates among cells.
- Cluster Dynamics: MPPT-controlled charging profiles induce complex inter-cell interactions during energy transfer.
Traditional single-cell SOC estimators fail to account for these system-level effects. We resolve this through:
- Memory-Augmented Feature Learning: Captures cell-specific degradation signatures
- Cluster-Wise MPPT Integration: Aligns charging profiles with real-time power conversion efficiency
- Cross-Cell Attention Fusion: Synthesizes localized and global operational states
2. Technical Framework
2.1 System Architecture
The CMCF network processes three input streams:
| Input Type | Dimensionality | Description |
|---|---|---|
| Cell Voltage/Current | RN×T×2 | Per-cell temporal measurements |
| Cluster MPPT Signals | RT×3 | Power-point voltage/current |
| Thermal Profile | RN×T | Surface temperature matrix |
Feature Fusion Engine:
Z=Softmax(dkQKT)V
where Q (queries) and K (keys) encode cluster-wide MPPT states, while V (values) represent cell-specific memory features.
2.2 MPPT-Enhanced Charging Protocol
Integrate perturb-and-observe MPPT algorithm:
Pmppt(t)=Vargmax[Ipv(t)⋅Vpv(t)]
ΔV=α⋅sign(dVdP)
This dynamic voltage adjustment directly impacts SOC estimation through:
SOCcorr=SOCnom−β⋅∫dVdPdt
2.3 Memory-Augmented Degradation Modeling
Memory Bank Update Rule:
Mt=γMt−1+(1−γ)⋅LSTM(xt)
where γ controls memory persistence and LSTM encodes temporal cell behaviors.
Degradation-Aware SOC Correction:
SOCfinal=SOCbase⋅(1−∑i=1kwi⋅Mt(i))
3. Experimental Validation
3.1 Dataset Configuration
| Parameter | Value |
|---|---|
| Battery Cells | 224 LFP cells |
| Sampling Interval | 1 minute |
| MPPT Voltage Range | 48V–58V |
| Temperature Range | 15°C–45°C |
Preprocessing:
- Normalized voltage/current to [0,1] range
- Detrended temperature measurements
3.2 Performance Metrics
| Method | MSE (×10⁻³) | MAPE (%) | MPPT Efficiency |
|---|---|---|---|
| Single-Cell LSTM | 4.72 | 2.31 | N/A |
| Cluster-Only TCN | 3.15 | 1.89 | 82.1% |
| CMCF (Ours) | 1.35 | 1.12 | 93.6% |
The 71.37% MSE reduction demonstrates our method’s superiority in handling cell variances and MPPT dynamics.
4. Mathematical Formulation
4.1 SOC Estimation Core
SOC(t)=SOC0−Qrated1∫0tη(τ)I(τ)dτ
where η incorporates MPPT efficiency:
η(t)=Pmppt(t)Pactual(t)
4.2 Memory Feature Projection
Fmem=σ(Wm⋅concat(Mt,Zt))
Wm∈Rdm×(dm+dz)
5. Conclusion
This work establishes a new paradigm for SOC estimation in MPPT-controlled battery clusters through:
- Memory-driven cell-specific modeling
- Cluster-wide feature attention mechanisms
- Tight integration of power transfer optimization
Future directions include real-time implementation on BMS hardware and expansion to hybrid AC/DC storage systems.
Appendix: Key Formula Summary
| Equation | Description |
|---|---|
| Pmppt=VI | MPPT power calculation |
| ΔV=α⋅dP/dV | Voltage perturbation rule |
| SOCcorr=… | MPPT-adjusted SOC correction |
| η=Pactual/Pmppt | Charge transfer efficiency |