The Phase Imbalance Diagnostic: Managing Voltage Vector Asymmetry in Multi-Charger Residential Feeder Lines

The Hidden Assumption of Balance in the Low-Voltage Network

The UK low-voltage distribution network is a three-phase, four-wire system, designed on the premise that the load connected to it will be, on average and over a sufficient number of customers, approximately balanced across the three phases. Houses are connected singly to one phase, and the connection of successive properties along a street is rotated between the phases by the installer, so that the statistical sum of many independent, small and uncorrelated demands is close to balanced. The premise has worked well for domestic loads of a few kilowatts: a kettle, a cooker or an immersion heater on a single phase has little effect on the aggregate when it is one of hundreds of loads of similar size.

The electric vehicle charger changes the statistics. A 7.36 kW single-phase charger at 32 A is a load that may be five to ten times the diversified demand of the house to which it is connected, it runs for hours rather than minutes, and its timing is correlated with those of its neighbours, because tariffs and commuting patterns concentrate charging into the same hours. When a cluster of such loads happens to be connected to the same phase, the premise of balance fails. This audit sets out the vector mathematics of the resulting asymmetry, quantifies the voltage, thermal and transformer consequences with a worked example, and describes the control architecture by which an edge-computed allocation scheme can restore balance. It is written for distribution planners, low-voltage engineers and the specifiers of multi-charger commercial and residential schemes.

The Vector Mathematics of Phase Asymmetry

The three line currents of a three-phase system are phasors, complex quantities with a magnitude and a phase angle. In a balanced system the three currents have equal magnitudes and are displaced from each other by 120 degrees. The neutral current is the phasor sum of the three: I_N = I_A + I_B + I_C. When the three have equal magnitude and are 120 degrees apart, they sum to zero and no current flows in the neutral. When they differ in magnitude, they do not cancel, and the difference flows in the neutral conductor.

The method of symmetrical components, introduced by Fortescue, resolves any unbalanced set of three phasors into three balanced sets: the positive sequence (the same rotation as the supply), the negative sequence (opposite rotation) and the zero sequence (three phasors in phase with one another). Defining the operator a = e^j120°, so that 1 + a + a² = 0, the sequence currents are I₀ = (I_A + I_B + I_C) / 3, I₁ = (I_A + a × I_B + a² × I_C) / 3 and I₂ = (I_A + a² × I_B + a × I_C) / 3. The neutral current is therefore I_N = 3 × I₀, so that the neutral current is a direct measure of the zero-sequence content of the load. The negative-sequence current I₂ is the quantity that drives the voltage unbalance, and the voltage unbalance factor is defined as VUF = |V₂| / |V₁| × 100%.

Worked Cases

Consider a feeder serving six 32 A single-phase chargers, at unity power factor for simplicity, neglecting other loads. If all six are connected to phase A, I_A = 192 A at an angle of 0 degrees, and I_B = I_C = 0. Then I_N = 192 A, I₀ = I₁ = I₂ = 64 A. The neutral carries a current equal to that of the loaded phase, and the negative-sequence current is one third of it.

If the chargers are distributed four, one and one, then I_A = 128 A, I_B = 32 A at minus 120 degrees and I_C = 32 A at plus 120 degrees. The neutral current is 128 + 32 × (−0.5 − 0.5) = 96 A, and I₂ = (128 + 32 × (a² + a)) / 3 = (128 − 32) / 3 = 32 A. If they are distributed two, two and two, then I_A = I_B = I_C = 64 A, I_N = 0 and I₂ = 0. The same total energy is delivered in all three cases, since the total real power is the same, but the neutral and negative-sequence currents vary from zero to the full phase current. The neutral current can also be reduced by the cancellation which occurs when chargers are connected to different phases, though it does not go to zero unless the phase currents are equal in magnitude.

It should be noted that the single-phase EV charger is a largely resistive, linear-in-effect load at the fundamental frequency, with a power factor close to unity, so that the neutral current is dominated by the fundamental-frequency imbalance and not by triplen harmonics, which are the usual concern in commercial installations with switched-mode loads. The on-board chargers of vehicles do generate some harmonic current, but at the levels permitted by the relevant product standards the effect on the neutral is secondary to the imbalance.

Voltage Drift, Ohmic Heating and Cable Degradation

The voltage at a customer's terminals depends on the voltage at the source minus the drops along the conductors that carry the current. For a single-phase load, the current flows out along the phase conductor and returns along the neutral, so that the drop is the sum of the drops in both. For a load with a power factor close to unity, the drop in each conductor is approximately I × R, where R is the resistance of the conductor over the length of the run. In a balanced three-phase system the return currents cancel at the star point, the neutral carries no current, and the drop on each phase is simply I_phase × R_phase. In an unbalanced system the neutral carries a current I_N and contributes a drop I_N × R_N, which displaces the neutral potential at the load end of the feeder away from that at the source. The phase-to-neutral voltage of each phase at the load end is the source phase voltage minus its own phase drop minus the neutral displacement, all as vectors.

Numerical Illustration

Take a 300 m feeder in 185 mm² four-core aluminium cable, with an operating resistance of approximately 0.2 ohms per kilometre, giving 0.06 ohms for each conductor over the run, and the neutral of the same cross-section as the phases. With all six chargers (192 A) on phase A, the drop in phase A is 192 × 0.06 = 11.5 V and the drop in the neutral is 192 × 0.06 = 11.5 V, so the phase A voltage at the load end falls by 23 V, from 230 V to approximately 207 V, a reduction of 10%. This is below the 216.2 V lower statutory limit, which derives from the declared voltage of 230 V and the tolerance of plus 10% and minus 6% in the Electricity Safety, Quality and Continuity Regulations. The neutral at the load end is displaced by 11.5 V in the direction of phase A, which has the effect of raising the voltages on phases B and C. Taking the phasor 230 V at minus 120 degrees less the displacement of 11.5 V at zero degrees yields a magnitude of approximately 236 V. The lightly loaded phases therefore rise to about 236 V, an increase of 2.6%, while the loaded phase falls by 10%, a spread of roughly 29 V between phases at the end of the feeder.

Calculating the sequence components of the three load-end voltages (207 V, and 236 V on each of the other two at the appropriate angles) gives V₁ of approximately 226 V and V₂ of approximately 3.8 V, a voltage unbalance factor of 1.7%. This corresponds to the negative-sequence current of 64 A multiplied by the phase impedance of 0.06 ohms. The value is within the limit of 2% for the voltage unbalance commonly applied to the low-voltage network, in line with EN 50160, but only just, and it scales in proportion to the feeder length and the negative-sequence current. On a 500 m feeder with the same loading it would be approximately 2.8%. The planning limits applied by distribution network operators, which are set out in Engineering Recommendation P29, are more stringent than the compatibility limit, since they must leave headroom for other sources of unbalance. The voltage unbalance matters because three-phase motors, including heat pump compressors and the three-phase on-board chargers of many vehicles, are sensitive to it: an unbalance of 2% can raise motor currents by many times that percentage and cause overheating.

With the 4-1-1 distribution the drop on phase A is 128 × 0.06 = 7.7 V plus the neutral contribution of 96 × 0.06 = 5.8 V, a total of 13.4 V or 5.8%, and the voltage unbalance factor falls to approximately 0.85%. With the 2-2-2 distribution, the drop on each phase is 64 × 0.06 = 3.8 V or 1.7%, with no neutral contribution and no unbalance.

Ohmic Heating

The heating loss in a conductor is I²R, so it is quadratic in the current, and a given total current carried unevenly produces more heat than the same current carried evenly. For the same example, the all-on-one-phase case dissipates 192² × 0.06 = 2.21 kW in the phase conductor and a further 2.21 kW in the neutral, a total of 4.42 kW over the 300 m. The 4-1-1 case dissipates (128² + 2 × 32²) × 0.06 = 1.11 kW in the phase conductors and 96² × 0.06 = 0.55 kW in the neutral, a total of 1.66 kW. The 2-2-2 case dissipates 3 × 64² × 0.06 = 0.74 kW and no neutral loss. For the same delivered energy, the unmanaged cluster therefore wastes six times the energy in the cable of the balanced arrangement. The loss also increases with temperature, since the resistance of aluminium rises by approximately 0.4% per kelvin, creating a feedback that amplifies the loss.

The heating has consequences beyond efficiency. Cable ratings assume specified conditions of burial depth, soil thermal resistivity and loading pattern, usually a balanced three-phase load, and a current in the neutral adds a fourth heat-generating conductor to the thermal calculation, so the rating of the cable must be reduced when it carries significant neutral current. Cross-linked polyethylene (XLPE) insulation is rated for continuous operation at 90 degrees Celsius, and older paper-insulated or PVC cables at lower temperatures. The ageing of polymeric insulation follows an Arrhenius-type law, with the life reduced by approximately half for each rise in operating temperature of the order of 8 to 10 K, so that a cable repeatedly held a few kelvin above its design temperature by several hours of evening charging loses life disproportionately. Joints, which have higher resistance and poorer heat dissipation than the cable, are the most vulnerable points, and a failure in a joint is a more likely outcome than a failure of the cable.

Substation Transformer Thermal Stress and Phase Clamping

The typical distribution transformer feeding a residential area is a delta-star unit, designated Dyn11, with an 11 kV delta primary and a 400 V star secondary with an accessible neutral. The delta winding is significant for unbalance. The zero-sequence component of the secondary current produces magnetomotive force in the secondary windings that is in phase in all three limbs. If there were no delta, this would produce zero-sequence flux that must complete its path outside the core, through the air and the tank, producing stray-flux heating in the structural steel and a distortion of the phase voltages. In the Dyn11 transformer the delta winding provides a path for a circulating current that cancels the zero-sequence ampere-turns, so that no net zero-sequence flux forms in the core, and the unbalanced load is reflected to the primary only as positive- and negative-sequence components. This is the reason that this winding arrangement is standard for distribution transformers supplying four-wire loads.

The transformer is not immune, however. The circulating current heats the delta winding as an addition to its normal loading. The negative-sequence current produces losses in the windings and a negative-sequence flux that induces eddy currents in the tank and clamping structures. Most importantly, the thermal rating of the transformer is determined by the hottest winding, and a winding carrying a disproportionate share of the load runs hotter than the nameplate rating, which assumes balanced loading, would suggest.

A 500 kVA Example

A 500 kVA, 400 V transformer has a rated line current of 500,000 / (1.732 × 400) = approximately 722 A. Assume a background load of 250 A on each phase and a cluster of 30 chargers, each at 32 A, for a total of 960 A. If they are distributed 10-10-10, each phase carries 250 + 320 = 570 A, which is 79% of the rating. If they are distributed 20-5-5, phase A carries 250 + 640 = 890 A, which is 123% of the rating, and phases B and C carry 250 + 160 = 410 A, or 57%. The total apparent power is the same in both cases, about 393 kVA, but the neutral current in the second case is 890 − 410 = 480 A.

The effect on the hot-spot temperature can be estimated using the loading guide approach of IEC 60076-7. The temperature rise of the winding hot-spot above the top oil scales with the load factor K raised to a power of 2y, where the winding exponent y is approximately 1.3 for oil-natural air-natural cooled units. If the winding-to-oil gradient at rated load is taken as 23 K, an illustrative figure, then at K = 1.23 on the heavily loaded winding the gradient is 23 × 1.23^2.6, which is approximately 39 K, an increase of 16 K over the rated condition. At K = 0.79 in the balanced case, the gradient is 23 × 0.79^2.6, or approximately 12.5 K, a reduction of 10 K. The top-oil temperature is a function of total losses and also rises in the unbalanced case, because the sum of the squared phase currents increases from 3 × 570² = 974,700 to 890² + 2 × 410² = 1,128,300, an increase of 15.8% in the load loss for the same delivered energy. The relative ageing rate of cellulose insulation in the standard loading guide doubles for every 6 K increase in the hot-spot temperature above a reference of 98 degrees Celsius. A rise of 16 K in the hottest winding corresponds to an increase in the ageing rate of the order of 2^(16/6), which is approximately 6 to 7 times, in that winding, compared with the rated condition, and the difference from the balanced case is correspondingly greater. These numbers are illustrative and depend on the design of the particular unit, but they demonstrate that the unbalance converts a comfortable loading into an accelerated ageing condition without any increase in the total energy delivered.

The voltage regulation of the transformer is also affected. For a transformer with an impedance of 4.5%, an X/R ratio of the order of 4 to 5 and a load power factor of 0.95, the voltage drop at a per-unit current K is approximately K × (R% × 0.95 + X% × 0.31), which equals K × 2.3%. At K = 1.23 this is about 2.9%, compared with about 1.8% for the balanced case at K = 0.79, and it is superimposed on the feeder drop calculated above. The most heavily loaded phase is therefore also the phase with the least voltage headroom, and the lightly loaded phases operate with a higher voltage. Because the tap changer on a distribution transformer is off-circuit and common to all three phases, the voltage headroom cannot be adjusted for individual phases.

The protective devices respond phase by phase. Each phase of the low-voltage board has its own fuse, rated typically in the range of 300 to 800 A depending on the feeder, and the overloaded phase will present a current that is closer to the fuse rating than the average. A fuse carrying an excessive current continuously ages, and may operate on a sustained overload, interrupting the supply to the customers on that phase while the others remain live. This is the mechanism behind the premature operation of phase fuses in clusters of chargers on a single phase, which the network operator experiences as a series of nuisance outages concentrated on particular streets.

The worked example relies on simplifications: unity power factor, a uniform conductor temperature, equal phase and neutral cross-sections and no background load. Real networks have source impedance, diversified household loads and different conductor sizes. The ordering of the options, however, is robust, and the dominant lever is the distribution of current among the three phases, not the total amount.

Dynamic Algorithmic Mitigation and Phase-Allocation Architecture

A static allocation policy, whereby each new charger is connected to the least-loaded phase at the time of installation, is a considerable improvement over random allocation, but it cannot track the changes in usage that occur after installation. Chargers differ in their duty, the phase most heavily used in the evening may not be the one assumed at the design stage, and household loads, heat pumps and the like add their own imbalance. The more effective approach is dynamic, using measurement and control.

Telemetry and State Estimation

The control loop requires a measurement of the phase currents, and where possible the phase voltages, at the point of interest. At the site level this can be obtained from current transformers on the incoming supply of a car park or a depot. At the street level the source can be a low-voltage monitoring unit at the substation, whose current transformers on each outgoing feeder report the phase currents and voltages, or aggregated smart meter data. At the charger level the charge point itself reports the current drawn and the voltage at its terminals. The controller combines these inputs to form an estimate of the current in each phase, the neutral current and the voltage unbalance factor, with timestamps, because the data from different sources have different delays.

The Allocation Problem

The decision to be made is the assignment of each active charger i to a phase p, and its current limit, in order to minimise a measure of unbalance subject to the limits of the network. In formal terms, using a binary variable x_ip equal to one if charger i is connected to phase p, and a current I_i for each, the phase currents are I_p = I_bg,p + sum over i of (x_ip × I_i). The objective can be to minimise the maximum deviation of any phase current from the mean, or the negative-sequence current magnitude, subject to constraints that each charger is connected to exactly one phase, that each phase current stays below its limit I_p,lim, that the vehicles receive the energy they need by their departure time, and that the number of phase changes is limited. The problem has the structure of a mixed-integer linear program or of a bin-packing problem, which can be solved quickly for the numbers of chargers on a single feeder, and which can be approximated by a greedy heuristic that assigns the largest remaining demand to the least-loaded phase.

Phase switching is not free. A charger that can change phase must open its contactor, move to the new phase and re-establish the session, which interrupts the vehicle charging for several seconds and, for vehicles that do not tolerate a change, may terminate the session. Phase switching should therefore be done at the start of a session or infrequently, with a minimum dwell time on each phase, and the algorithm should penalise changes. Three-phase chargers connected to three-phase on-board chargers are inherently balanced, apart from the effect of the vehicle on the load, and offer a simpler solution where the vehicle supports it.

Curtailment and Communication

Where phase switching is not available, the controller can restore balance by curtailing the chargers on the overloaded phase. The mechanism for doing so is the charge profile. Under the Open Charge Point Protocol, the central system sends a SetChargingProfile message that sets a limit in amperes or watts, for a connector or a whole station, and the charge point enforces it by changing the duty cycle of the control pilot signal presented to the vehicle. In the IEC 61851 standard, the vehicle is required to respond to a change in the pilot within a defined time, of the order of five seconds. Where the central system is local, as in an edge gateway at the site, the message round trip over a local network is tens to hundreds of milliseconds, so the limiting factors are the vehicle's response and the control interval, which is typically one to ten seconds. The description of such a loop as sub-second is accurate for the communication but not for the end-to-end response, and the controller should be designed on the longer figure.

The control law should include hysteresis, rate limits and a minimum-current floor, since the pilot cannot signal a current below 6 A, and a charger cannot be throttled continuously below that value. A charger must either be held at 6 A or more, or suspended. The consequences for the user should be explicit: a curtailed vehicle charges more slowly and the system should ensure by prioritisation or rotation that no vehicle is starved. The deadlines of the vehicles feed back to the constraint set, so that a vehicle with an early departure is given priority over one that will remain for the night.

Fail-Safe and Regulatory Considerations

The system must define its behaviour on loss of telemetry or communication. A charger that has not received an update within a timeout should fall back to a conservative current, in the range of 6 to 16 A, rather than to its maximum, and the fallback must be tested. The UK Smart Charge Points Regulations 2021 require domestic and workplace charge points to be capable of responding to signals, to default to off-peak charging and, as I understand them, to include a randomised delay of up to ten minutes in the restart of charging after an interruption or the start of an off-peak period, which assists by breaking up the synchronisation of many chargers switching on at the same moment. The statutory voltage envelope is relevant to the targets: the supply voltage must stay between 216.2 V and 253 V, and the voltage quality standards limit unbalance. Requirements for connecting generation, such as G98 and G99, apply to exporting equipment, including bidirectional chargers, and not to the import behaviour considered here, though an installation with vehicle-to-grid capability would need to comply with them.

Specification Guidance for Planners and Specifiers

Several points follow from the analysis. First, the unit of analysis for a multi-charger scheme is the phase, not the total. The assessment of capacity should be made for the most heavily loaded phase at the time of peak, including a realistic coincidence factor for the chargers, which for a residential cluster on a cheap-rate tariff may approach one. Second, a static allocation rule that rotates the phase assigned to successive chargers should be the minimum standard for all multi-charger schemes, with a register of the phase assignments that is kept up to date. Third, wherever the cluster is large enough to affect a substation, a dynamic scheme with telemetry, per-phase limits and a fail-safe should be considered, and the opportunity should be discussed with the network operator, since the operator holds the information about the headroom and may offer connection terms that reflect the controllability of the load. Fourth, three-phase chargers and phase-switching hardware should be preferred where they are available, and the scheme should be commissioned and tested, including the failure modes. The aim is the same as in every other case of this series: to treat the supply as a constrained physical system, and to design the control around the constraints that it imposes.