Why Parallel Batteries Do Not Share Current Equally

Parallel batteries share a terminal voltage, but they do not necessarily share current equally. Each branch responds to its own internal voltage and resistance. As charge moves, the states of charge change, so the current split changes too. A branch carrying less current now may therefore behave differently near the end of charging or after the external load stops.

This matters when battery capacity is expanded in modules: the total ampere-hour rating says little about the current through each branch. The physics also explains an apparent contradiction in battery research. Parallel connection can help some differences shrink, yet that does not mean every cell carries the same current, ages at the same rate or recovers lost capacity.

One terminal voltage can hide two different currents

Open-circuit voltage, or OCV, is the voltage associated with a cell’s electrochemical state when current is absent. Under load, the measured terminal voltage also includes voltage drops. In a simple resistance model, V = U − IR: terminal voltage V equals OCV U minus branch current I multiplied by resistance R, with positive current defined as discharge. Two parallel cells can have the same V while their U, R and I differ. Weng and colleagues use this distinction to derive the current-sharing equations for two cells. [1] (pp. 38–40).

For cells a and b, the model gives I_a = (U_a − U_b + R_b × I_total) / (R_a + R_b), with I_b = I_total − I_a. Voltage is in volts, resistance in ohms and current in amperes. The expression contains two effects: a share of the external current set by resistance, and a transfer driven by the difference between the cells’ internal voltages. [1] (p. 38, equations 4–9).

If the two OCVs are equal at that instant, the lower-resistance branch carries more of the applied current. Once the OCVs differ, the voltage-difference term can strengthen or oppose that split. Dividing total current by the number of branches therefore describes an average; it does not establish the current in any particular branch.

The last row is especially useful. At zero external current, the equation becomes I_a = (U_a − U_b) / (R_a + R_b) and I_b = −I_a. Charge can circulate between the branches even though a sensor on the common output reads zero. This describes the mechanism, not a safe connection procedure or an allowable voltage mismatch. [1] (p. 38).

Why the current split changes during a cycle

State of charge, or SOC, describes the fraction of a cell’s present charge capacity that remains. Its rate of change depends on current divided by capacity: the same ampere current moves a smaller-capacity cell through its SOC range faster. That change can alter OCV, which then feeds back into current sharing. Resistance sets the immediate response; capacity helps determine how the response evolves. [1] (pp. 38–39).

Weng’s analytical model assumes constant capacity and resistance, and identical straight-line OCV–SOC relationships for the two cells. Under a constant external current, the SOC difference approaches a steady value. That steady value can be nonzero: the cells may move through their charge ranges at the same rate while retaining an offset between them. A stable difference is not the same as no difference. [1] (pp. 39–40).

For two cells with fixed capacities and resistances, identical linear OCV–SOC curves and a sufficiently long constant-current interval, the model predicts that current is shared in proportion to capacity: I_a = Q_a × I_total / (Q_a + Q_b). Here Q_a and Q_b are the cells’ present charge capacities in ampere-hours. The cells then experience equal current relative to their own capacities, even if their ampere currents differ. Equal capacities give equal steady currents in this particular model despite a resistance mismatch, because the SOC-driven contribution offsets the resistance-driven contribution. [1] (p. 40, equation 21).

The time needed to approach this state matters. For the linear model, the characteristic time is τ = [(R_a + R_b) / α] × [Q_aQ_b / (Q_a + Q_b)], where α is the slope of OCV against fractional SOC. With capacity in ampere-hours, resistance in ohms and α in volts, τ is in hours. The paper uses three time constants as an approximately 95% approach to the steady value. A short charge or discharge interval may end before that condition is reached. [1] (p. 39).

Charging also changes character when the charger reaches its voltage limit. During the constant-voltage phase, total current falls, and the SOC differences can evolve differently from those during constant-current charging. The paper’s example shows the SOC difference reducing during the voltage hold, followed by a smaller current-imbalance spike when discharge begins. The preceding charging history therefore matters to the next current transient. [1] (pp. 39–40, figure 2).

The shape of the voltage curve changes the answer

The straight-line OCV assumption makes the mechanism understandable, but its exact cancellations are not universal battery properties. Weng and colleagues compare the analytical result with a nonlinear NMC/graphite OCV curve. The linear model underestimates SOC imbalance in the illustrated case and misses a change in the current-imbalance trajectory near the end of charge. A piecewise-linear voltage curve follows the nonlinear simulation more closely. [1] (p. 42, figure 4).

This distinction is particularly relevant when discussing LFP batteries. In a flat part of an LFP OCV curve, a substantial SOC difference can produce only a small voltage difference. In the linear model, a smaller OCV slope weakens the voltage-driven response and lengthens the rebalancing time; a steep region near the end of charge changes that behavior. The NMC/graphite numerical example cannot be reused as an LFP module prediction. [1] (pp. 39 and 42). The related LFP SOC article explains why voltage alone can leave charge state uncertain.

Two practical interpretations follow from the model. Equal current at one operating point does not prove equal SOC. Likewise, closely matched SOC does not prove equal current when capacities differ. The quantity being compared—voltage, SOC, current or capacity—must stay explicit, because agreement in one does not establish agreement in the others. [1] (p. 41, figure 3).

Capacity differences can shrink while every cell keeps aging

Long-term aging asks a different question from current redistribution within one cycle. Song and colleagues coupled electrical, thermal and aging models to study capacity variation in parallel-connected cells. Their five-cell LFP simulations began with the same total capacity but different distributions among the cells. Under the modeled conditions, larger-capacity cells lost capacity faster and the differences narrowed. [2] (pp. 7–9, figures 4–6).

The result is convergence through unequal wear, not restoration. In the simulated capacity curves, all cells continue losing capacity while their trajectories draw together. A more uniform group can therefore be an older group with less total charge capacity remaining. This is why the term “self-balancing” needs a named quantity: convergence of capacity, redistribution of SOC and equality of current describe different outcomes. [2] (pp. 8–9).

The table separates two effects that are easily confused. Initial variation diminished in the first two simulated groups, yet the initially uniform group still lost slightly less total capacity. The modeled benefit of matching was modest in this comparison; the study does not establish a universal penalty for every mismatched battery bank. Its approximately 300-hour convergence under aggressive simulated cycling is also not a field timetable. [2] (pp. 7–9).

The authors link convergence to the assumed degradation behavior and similar thermal conditions. Their conclusion identifies convex or linear remaining-capacity trajectories as supporting convergence under the analyzed assumptions. The paper also states that laboratory validation of the progression was future work. Its numerical curves should therefore be read as model results, with the aging law and thermal assumptions attached. [2] (pp. 6–10).

Temperature can upset the apparent self-regulation

The thermal assumption is consequential. Song and colleagues explain that a hot cell can have lower resistance and attract more current, while elevated temperature also accelerates the modeled degradation directly. If one location remains much hotter than the others, the shared electrical connection does not guarantee that aging differences will shrink. Similar cooling coefficients in a simulation are a specific boundary condition, not a description of every physical rack. [2] (p. 10, section 5.2).

Current alone is also an incomplete aging indicator. Weng and colleagues discuss a model scenario in which an aged cell takes less charging current but reaches a higher SOC than the fresher cell. They argue that lower current could reduce one degradation driver while the higher charge state increases another. Which effect dominates depends on the degradation mechanisms included; their current-sharing equations do not by themselves settle the lifetime outcome. [1] (pp. 42–43, section 4.3).

Taken together, the papers explain why a low branch current is not automatically evidence of a healthier branch, and why a stable pack-average temperature does not establish uniform cell conditions. Electrical redistribution and thermal aging interact. Neither study turns that interaction into a blanket permission to combine batteries with different histories.

What this means for a modular HUA-X battery system

A complete battery module adds BMS behavior, switching, protection and communications to the cell-level physics. The two-cell model does not include those system functions. Nor does it describe every series/parallel arrangement inside a module. Its value is to explain why a bank’s total current and nominal capacity cannot establish the operating conditions of every branch.

The HUA-X PowerHom-5R rack battery provides a concrete example of the quantities that remain distinct: its published specifications are 51.2 V, 100 Ah and 5.12 kWh nominal energy, with maximum continuous charge and discharge currents of 50 A and 100 A respectively. Those current limits apply to the module; nominal energy describes a different constraint. [3]

For an approved multi-module configuration, arithmetic addition of nominal capacity does not demonstrate equal current sharing or establish the bank’s allowable current. The limiting branch, protective switching and inverter/BMS coordination can matter before the arithmetic total is reached. The PowerHom-5R product page therefore leaves module count and expansion topology dependent on the supplied configuration. CAN/RS485 interfaces alone do not establish compatibility with a particular inverter or a mixture of battery revisions. [3]

The useful distinction is between adding nameplate energy and establishing how a battery bank behaves. Parallel connection changes the way differences interact; it does not erase them. For a PowerHom-5R project, the technical document request and project inquiry provide the route to the model-specific expansion and interface information. Cell-level research does not specify a PowerHom-5R branch-current tolerance or establish mixed-age expansion capability.