The rapid advancement of energy storage systems has necessitated the development of robust methodologies for managing and analyzing operational data. Energy storage batteries, particularly lithium-ion variants, generate vast amounts of fragmented data during charge-discharge cycles, often characterized by low quality, multi-source heterogeneity, and incomplete sequences. This paper addresses these challenges by proposing a novel framework for energy storage battery operation data splicing and reconstruction. Our approach leverages gradient descent optimization, empirical modeling, and boundary condition analysis to ensure continuity, accuracy, and adaptability in reconstructed datasets.
1. Introduction
Energy storage batteries are pivotal in modern power grids, renewable integration, and electric vehicles. However, their operational data, collected by battery management systems (BMS), often suffer from fragmentation due to dynamic load profiles, intermittent usage, and varying sampling rates. Disjointed data hinder accurate state-of-health (SOH) estimation, capacity analysis, and fault prediction. Existing studies primarily focus on reconstructing specific curves (e.g., SOC-OCV, EIS) but overlook the holistic splicing of multi-parameter operational fragments.
Our work bridges this gap by introducing a mechanism-driven methodology that unifies voltage, current, capacity, and transient dynamics into coherent datasets. The contributions include:
- A gradient descent-based algorithm for minimizing splicing errors.
- Empirical equations and boundary conditions governing parameter continuity.
- Validation through hybrid pulse power characterization (HPPC) and reference performance test (RPT) data.
- Real-world applications in grid-scale energy storage systems.
2. Methodology
2.1 Gradient Descent Optimization
The core of our method lies in minimizing discontinuities at splicing points. Let X={x1,x2,…,xm}X={x1,x2,…,xm} represent a dataset of operational parameters. The loss function J(X)J(X) quantifies deviations between adjacent data segments:J(X)=∑i=2mmax(0,∣xi−xi−1∣−p)2J(X)=i=2∑mmax(0,∣xi−xi−1∣−p)2
where pp is the permissible boundary value. The gradient is computed as:∂J∂xi=2(∣xi−xi−1∣−p)⋅sign(xi−xi−1)∂xi∂J=2(∣xi−xi−1∣−p)⋅sign(xi−xi−1)
Iterative updates refine the splicing points:xi(t+1)=xi(t)−α∂J∂xi(t)xi(t+1)=xi(t)−α∂xi(t)∂J
where αα is the learning rate.
2.2 Boundary Conditions
Critical parameters and their constraints are summarized below:
| Parameter | Constraint | Rationale | ||
|---|---|---|---|---|
| Current (IdId) | ( | I_{\text{front}} – I_{\text{back}} | \leq 5 , \text{A} ) | Prevents abrupt current shifts, ensuring voltage/capacity continuity. |
| Capacity (CdCd) | ( | C_{\text{front}} – C_{\text{back}} | = 0 ) | Enforces energy conservation and seamless capacity integration. |
| Voltage (UdUd) | ( | U_{\text{front}} – U_{\text{back}} | \leq 0.005 , \text{V} ) | Maintains voltage continuity for accurate state estimation. |
| Voltage Rate (kdkd) | ( | k_{\text{front}} – k_{\text{back}} | \leq 0.0001 ) | Ensures smooth transitions in dynamic behavior (e.g., internal reactions). |
| Transient Duration (TT) | T≥96 sT≥96s | Skips transient phases during工况 shifts to stabilize data. |
2.3 Workflow
The reconstruction process involves six stages:
- Data Acquisition: Collect voltage, current, and capacity data from BMS sensors.
- Fragment Segmentation: Divide data into stable,工况-homogeneous segments.
- Preprocessing: Clean outliers, interpolate missing values, and filter noise.
- Splicing: Apply gradient descent to align fragments under boundary conditions.
- Optimization: Refine curves using physics-informed models (e.g., RC circuits) or machine learning (e.g., XGBoost).
- Validation: Evaluate using RMSE, MAE, and visual comparisons.
3. Validation and Results
3.1 HPPC Test Validation
Using HPPC discharge-phase voltage data (280 Ah battery, 140 A pulse), our method reconstructed fragmented curves with high fidelity. Transient durations during工况 shifts were analyzed (Table 1):
| Rank | Duration (s) | Rank | Duration (s) |
|---|---|---|---|
| 1 | 61 | 7 | 82 |
| 2 | 67 | 8 | 85 |
| 3 | 69 | 9 | 86 |
| 4 | 72 | 10 | 87 |
| 5 | 77 | 11 | 88 |
| 6 | 80 | 12 | 96 |
The reconstructed voltage curves exhibited RMSE < 20 mV and MAE < 1.3%, validating the method’s precision.
3.2 RPT and Real-World Applications
RPT charging data (75 Ah battery, 0.2 C rate) were split into five fragments and reconstructed. Key results included:
- Voltage Consistency: Ud≤0.003 VUd≤0.003V.
- Capacity Alignment: Cd=0Cd=0.
Field tests on grid-scale energy storage batteries (271 Ah, 135 A) further demonstrated robustness. Reconstructed datasets enabled accurate incremental capacity analysis (ICA) for SOH estimation:SOH=QnowQstart×100%SOH=QstartQnow×100%
where QnowQnow and QstartQstart represent current and initial mid-section capacities derived from ICA curves.
4. Engineering Applications
4.1 Peak Shaving and Frequency Regulation
In a 240-cell energy storage battery cluster (5 s sampling), our method reconstructed frequency modulation voltage curves with RMSE<0.01 VRMSE<0.01V. Parameter ranges aligned with empirical values:
| Parameter | Range |
|---|---|
| Voltage | 3.307–3.68 V |
| Current | 2.5–136.4 A |
| State of Energy | 41–83% |
| Duration | 100–1,220 s |
4.2 SOH Estimation via ICA
By integrating spliced data with ICA, SOH estimation errors were reduced by 40%. This synergy enhances predictive maintenance strategies for energy storage batteries.
5. Conclusion
This study establishes a comprehensive framework for energy storage battery operation data splicing and reconstruction. Key outcomes include:
- Mechanistic Insights: Gradient descent optimization and boundary conditions ensure parameter continuity.
- Validation: HPPC, RPT, and real-world datasets confirm RMSE < 20 mV and MAE < 1.3%.
- Applications: Enhanced SOH estimation and grid-scale operational efficiency.
Future work will explore deep learning models for real-time reconstruction and multi-battery synchronization. Our methodology not only elevates data quality but also advances the reliability of energy storage battery systems in sustainable energy infrastructures.