① What’s the Story
The remaining battery percentage on an EV dashboard is not a directly measured value but an estimate that the battery management system (BMS) computes from voltage, current, and temperature data using algorithms. The focus of patent competition around this estimation is shifting from “making one algorithm more precise” to “how to cross-check and weight multiple estimation results.”
Starting with the industry situation, state of charge (SOC) is a key indicator that underpins driving-range guidance and protection against overcharge and over-discharge. The tension is that estimates that are relatively accurate when a battery is new become shaky as usage time lengthens, because of the gap between actual chemical degradation and the model the algorithm assumes. Even for the same cell, load patterns differ depending on where it is used — a car, a two-wheeler, or an ESS — so the nature of the error differs too.
This article answers two questions. First, on what principles is SOC estimated, and what are the weaknesses of each method? Second, with what rights structure did two published patents (KR20150109643A and KR20230109916A) address these weaknesses, and where can later entrants look for design-around options?
② Basic Principle
Because battery cells behave nonlinearly, values such as SOC, SOH (State of Health), and SOP (State of Power) cannot be measured directly. The BMS feeds measurable data such as voltage, current, and temperature into respective estimation models to calculate them. In eco-friendly vehicles, battery performance is directly tied to vehicle performance, so how accurate and fast the estimation is becomes an indicator that separates BMS capabilities.
Source: reconstructed by the editorial team from the configuration described in published document KR20150109643A
By analogy, SOC estimation is like gauging the remaining fuel in a car without a fuel gauge. One way is to keep subtracting what you have used from what you filled (current integration); another is to park the car for a while and measure the level in the tank (rest voltage). The first accumulates meter error; the second requires stopping the car. That is why real BMSs combine several methods so each covers the others’ weaknesses.
The academic literature organizes it the same way. A study published in Energies in 2022 divides SOC estimation into four branches: characteristic-parameter-based (the open-circuit voltage, OCV, method), integral estimation (Coulomb counting), physical-model-based (Kalman filters, etc.), and data-driven (machine learning). It explains that Coulomb counting estimates the initial SOC from OCV and then integrates current over time, and so fundamentally depends on the accuracy of the initial SOC and of current measurement.
- SOC: how full is it now (remaining charge)
- SOH: how much has it aged compared with a new battery (degradation)
- SOP: how hard can it be discharged or charged right now (power limit)
③ Comparison
The table below sets out the principles and weaknesses of representative SOC estimation methods side by side. Quantitative figures such as error rates vary widely with cell chemistry and test conditions, so they are not included here.
| Method | Input data | Principle | Main weakness |
|---|---|---|---|
| OCV-based | Voltage (V) | Uses the correspondence between rest voltage and SOC | Needs rest time, so hard to use alone while driving |
| Coulomb counting | Current (A) and time | Adds or subtracts integrated charge (Ah) from an initial SOC | Initial SOC error and current sensor error accumulate |
| Model-based (Kalman filter, etc.) | Voltage, current, temperature | Corrects equivalent-model predictions with measurements | Error grows when model parameters shift with degradation |
| Data-driven (machine learning) | Large volumes of operating data | Estimates SOC directly from learned relationships | Reliability outside the training range needs verification |
Source: reconstructed by the editorial team from the classification in Zine et al., Energies 15(21), 8172 (2022)
What this table says: The four methods have different weaknesses, so none can be called superior outright. A structure that combines multiple estimates becomes the natural solution, and the two patents below are cases that turned exactly such combinations into rights.
④ Patent Perspective
KR20150109643A (battery pack degradation state estimation device and method): The configuration identified for this patent document is a multi-stage computation pipeline. A status information sensing unit measures voltage and current, a conversion computation unit converts them into open-circuit voltage (OCV), and an estimation unit calculates SOC from the OCV. This SOC value is used in two branches. One branch uses the SOC itself to estimate a primary SOH; the other derives a degradation factor from the SOC and applies it to a pre-built lookup table to obtain a secondary SOH.
A filtering unit combines the two SOH values into a final value. The core is a dual-estimation structure that derives SOH along two paths and filters them as if cross-verifying. However, the applicant, filing date, grant status, and legal status of this document could not be confirmed in this research because access to Google Patents was restricted, and they are left as items requiring verification below.
KR20230109916A (battery state prediction method and battery system providing the same): According to Google Patents, the applicant is LG Energy Solution; it was filed on January 14, 2022 (same as the priority date), published on July 21, 2023, and granted on February 25, 2026 as KR102931475B1, and it is Active. Paraphrasing claim 1, it consists of a battery including multiple cells; a communication unit that receives identification information of the upper system via wireless communication; a monitoring unit that collects voltage, current, and temperature; multiple SOC estimation models based on predetermined algorithms; a storage unit holding weights for each piece of identification information; and a control unit that applies the weights to the results of the multiple models to calculate an averaged SOC for each cell.
Source: reconstructed by the editorial team from the claims of KR20230109916A (Google Patents)
This claim structure reads less as a framework broadly bundling “SOX” (a collective term for SOC, SOH, and SOP) and more as SOC estimation limited to changing weights according to “which upper system the battery is mounted in.” The specification explains that it raises the reliability of estimation results by using weights optimized for different applications such as vehicles, two-wheelers, and ESS. The heart of the scope of rights lies not in the estimation models themselves but in the chain “receive identification information → select weights per identification information → weighted average.”
Placing the two patents side by side reveals the filing trend. KR20150109643A raised “accuracy” by producing estimates along two paths from the same data, while KR20230109916A, filed in 2022, brought contextual information — “where is this same battery used?” — into the estimation. This supports the interpretation that the axis of competition is shifting from algorithms to application data and weight management.
There is also a point to note on enforcement. Both patents follow the typical approach of claiming the algorithm as a device configuration made up of “units,” which sidesteps questions of patent eligibility for software inventions. However, because weight tables and filtering logic are hard to observe from outside the product, proving infringement may not be easy for the rights holder. Conversely, rather than relying on that difficulty of proof, implementers are better placed in a dispute if they keep claim-comparison evidence at the level of source code and design documents.
Design-around scenario (counterargument): A business seeking to avoid KR20230109916A could consider a design that omits some claim elements, for example by using weights fixed at the manufacturing stage instead of receiving upper-system identification information wirelessly, or by selecting a single model instead of a weighted average. There is a counterargument, however. If identification information is received via wired communication or another route and ultimately performs the same function, it could become subject to a doctrine-of-equivalents assessment. Whether a design-around succeeds must therefore be judged by comparing the actual implementation of the communication method and weight application against the claims, and it is safer to conduct the FTO review before the software specification is finalized.
⑤ Limitations and Outlook
SOC estimation is relatively accurate when a battery is new, but the fundamental limitation is that as usage time lengthens, error accumulates between actual chemical degradation and the model the algorithm assumes. For this reason, the patent trend is moving away from refining a single estimation algorithm and toward multi-model structures that cross-verify multiple estimates and correct them to fit the situation. The fact that KR20230109916A claims SOC calculation per cell rather than for the pack as a whole can also be read as a design reflecting the reality that degradation differences between cells widen with longer use.
There is also a counter-scenario. Multi-model and per-application weight structures carry heavier validation burdens as the number of models and weight tables grows, and badly set weights can even produce worse results than individual models. And since weights must be relearned when cell chemistry changes, multi-model structures cannot be assumed to always win on a cost-benefit basis.
The strategic alternative is to manage “the data and procedures that produce the weights,” rather than the estimation algorithm itself, as an asset. Protecting the method of deriving weights from per-application operating data, the weight-update procedure, and validation criteria separately through distinct filings or as trade secrets may preserve a competitive edge even if algorithm claims are designed around.
Practical action: Those planning BMS software should check whether their SOC logic has a “function that changes model weights according to the identification information of the host system,” and if so, first draw up an element-by-element comparison chart against claim 1 of KR20230109916A (granted as KR102931475B1).
Items requiring verification
- The applicant, filing and publication dates, grant status, and current legal status of KR20150109643A
- Whether the actual wording of claim 1 of KR20150109643A matches the multi-stage pipeline described above
- Whether the granted claims of KR102931475B1 are the same as claim 1 at the publication stage (KR20230109916A)
- The extent to which the term “SOX” is used in the specification of KR20230109916A
Sources
- Google Patents, KR20230109916A – Battery state prediction method and battery system providing the same (LG Energy Solution)
- Zine, B. et al., “Experimentally Validated Coulomb Counting Method for Battery State-of-Charge Estimation under Variable Current Profiles,” Energies 15(21), 8172 (2022)
- KR20150109643A – Battery pack degradation state estimation device and method (published document)