External Load Measurement for EV Charging

What is being measured, and where

The useful quantity is not what the chargers are drawing. It is what everything else is drawing, so the charging allocation can be set to the difference between that and the connection limit.

1. 1

   At the incoming supply

   Current transformers on the main incomer, upstream of everything including the chargers. This measures total site load, and the charging allocation becomes the connection limit minus what is measured. The correct position for almost every shared supply.

2. 2

   On the building circuits only

   Measuring the building but not the chargers. Arithmetically equivalent and often easier to retrofit, but it depends on nothing else being added downstream later.

3. 3

   On the charging circuit only

   Measures what charging is doing and tells you nothing about headroom. Useful for verification, useless as the input to a load manager, and a surprisingly common mistake.

The third case is worth calling out because it produces a system that appears to work. It shares capacity between chargers correctly and has no idea that the building has just switched on its chillers.

Sampling, averaging, and why fast is not always better

Instinct says measure as fast as possible. In practice a control loop that reacts to every transient produces charging that surges and dips continuously, which vehicles dislike and which achieves nothing.

• The protective device you are trying not to trip has a time-current characteristic. It tolerates a brief overload and not a sustained one, so the averaging window should reflect that rather than the instantaneous value
• Vehicles take time to respond to a changed limit, so commanding faster than they can follow only adds error
• Lift starts, chiller compressors and welding loads produce short peaks that will pass before charging could have responded
• A slower window with a conservative margin is usually more stable and safer than a fast window chasing noise

The design question is therefore not sample rate alone. It is sample rate, averaging window and reserve margin together, chosen against the characteristic of the device being protected.

Three phase, and the imbalance problem

On a three phase supply the limit applies per phase, not to the total. A site with 100 A per phase does not have 300 A to give away.

This matters because single phase charge points load one phase each, and buildings are rarely balanced to begin with. Measuring only the total, or only one phase and assuming the others match, produces a system that trips the most loaded phase while reporting plenty of headroom. Each phase has to be measured, and the allocation has to be made per phase, which also means knowing which phase each charge point is connected to. That mapping is frequently undocumented and has to be established during commissioning.

Retrofitting onto an existing panel

Most of this work is retrofit rather than new build, and the constraints are physical before they are electrical.

• Split core transformers can be fitted without breaking the conductor, which usually decides the approach because an outage on a live building is hard to arrange
• Space around the incomer is often the binding constraint, particularly in older panels
• Getting the signal from the panel to the controller is a cabling route that has to be found, and it is frequently the longest part of the job
• Existing metering may already be present and readable over Modbus, which avoids new transformers entirely
• Direction and phase orientation must be verified at commissioning. A transformer fitted backwards produces a reading that looks plausible and is wrong in the dangerous direction

Losing the measurement

A load manager is only as safe as its behaviour when it cannot see. This has to be designed rather than assumed, and it has to be tested.

• Fall back to a fixed allocation that is safe with no knowledge of building load, which usually means a small fraction of the connection
• Detect loss by timeout, because a failed sensor or a dead bus produces silence rather than an error
• Never hold the last known value. The reason the measurement stopped may be the same event that changed the load
• Alarm it, because a site quietly running at reduced allocation for three weeks is a fault that went unnoticed
• Test recovery, not only failure