Definition
AMR (Automatic Meter Reading) remotely collects and reports water, electricity, gas, and heat meter data to the meter reading platform through meter data collectors and concentrators. It supports meter protocols such as Modbus, M-Bus, DL/T 645, LoRa, suitable for residential communities, commercial buildings, and industrial parks, and supports energy management and sub-metering.
System architecture
An AMR system typically consists of meters, collectors, concentrators, communication networks, and a meter reading platform. Meters (water, electricity, gas, heat) have built-in pulse outputs or communication modules; collectors connect to multiple meters nearby and collect readings via M-Bus, RS485, pulse lines, etc.; concentrators aggregate data from multiple collectors and upload it to the meter reading platform via 4G, NB-IoT, LoRa, fiber optics, etc. The power industry commonly uses the DL/T 645 protocol, while water, gas, and heat meters mostly use M-Bus or Modbus. On the platform side, functions such as data storage, billing, reporting, exception alerts, and energy analysis are realized.
Communication Methods
The connection between meters and collectors uses wired buses (M-Bus, RS485, MBus) or wireless (LoRa, Zigbee, RF, etc.); the connection between concentrators and the platform uses public networks (4G, NB-IoT, fiber optics) or private networks. NB-IoT is suitable for meters directly connecting to the platform without the need for concentrators, offering wide coverage and low power consumption; LoRa is suitable for self-organizing networks within buildings, with controllable costs; 4G is suitable for concentrator uplinks with sufficient bandwidth. When choosing, coverage, cost, power consumption, and operational convenience need to be considered.
Application Value
AMR replaces manual meter reading, improving efficiency, reducing errors, and supporting real-time billing and prepaid services. Sub-metering can distinguish electricity use for lighting, air conditioning, power, etc., supporting EMS energy management and energy-saving optimization. Abnormal consumption analysis can detect leakage, theft, and equipment failure. Data accumulation provides the basis for load forecasting, peak-valley analysis, and demand-side management. It is widely used in residential communities, commercial complexes, industrial parks, public buildings, and other scenarios.
Frozen values and tariff windows
Automatic meter reading needs the frozen quantity at the billing instant, not the instantaneous value flashing on the display when the collector happens to poll. A multi-function electricity meter usually writes forward active energy, and the energy in each tariff, into freeze registers on the hour or at the day boundary. Some water and heat meters keep only a cumulative volume; others also keep an hourly freeze. If the collector takes the instantaneous quantity before the freeze is written, daily usage on the platform differs from the meter's billed value by one interval. Meter rooms in a residential building are dense and the supply is stable, so a concentrator on a wired bus fits. Power cabinets in a plant usually have multi-function meters spread out. Each meter needs its own communication address, and the poll interval must be longer than that meter's response time. Time-of-use pricing and prepayment also require the collector clock and the meter clock to stay within about the same minute. Once the peak and off-peak windows drift apart, energy is booked to the wrong tariff. A successful exchange does not mean the bill is right.
Water, electricity, gas, and heat should not share one wait time. An electricity meter answers frame by frame. A gas or water meter on M-Bus wakes more slowly, and a concentrator timeout copied from the electricity meters will mark the slow meters offline. A heat meter should be read for both cumulative heat and instantaneous flow. If only cumulative heat is stored, a stopped meter cannot be told from a user who is not taking heat. Whether the uplink is a cellular module or the campus network, the concentrator should keep at least one day of frozen copies locally, so a short platform outage does not drop that day's billing window. A catch-up read must carry the freeze time. The time of the catch-up must not overwrite the original timestamp.
Meter address and time-base mismatch
The meter number, the communication address, and the asset code are often different fields. The address field in DL/T 645 is the communication address. The asset number on the nameplate is not necessarily the address to poll. The M-Bus primary address space is small, and the factory number sits in the secondary address. If the reading software uses the asset number as the slave address, the whole bus only times out, which looks like a dead concentrator. Another error is the ratio. If the current-transformer or voltage-transformer ratio is never written on the platform, energy is scaled too high or too low while the frame check is perfectly valid. For a pulse meter, confirm the volume or energy each pulse represents, and whether a collector power loss clears the pulse buffer. A missing terminator at the end of the bus, A and B swapped, or mixed baud rates on the same line all show up as occasional missing readings, not as a calculation error on the platform.
If a replacement concentrator keeps the old meter archive, the new unit polls the old addresses and a whole community suddenly has no data. Time-set commands should go out only outside the freeze window, and energy read during a time set should not be stored as a billing value. For scattered meters already reached through a DTU, transparent mode does not assemble freeze frames. The master still has to poll in the meter protocol. Treating the DTU as a collector that has already parsed the protocol leaves the platform with a partial frame or frames stuck together. When the uplink is congested, keep the daily and monthly freezes first. Instantaneous values can be slowed. The billed values should not be the ones dropped.
Catch-up reads and bus occupancy
If catch-up reads are inserted into the daytime instantaneous poll, slow meters push the billing window aside. The concentrator should place load profiles in a low-usage period and limit how many frames one meter may send in a single catch-up run. A water meter wakes more slowly than an electricity meter, so the electricity meter's response time must not be used as the water meter's timeout. When an address in the archive has to change, stop polling that meter before writing the new address. Changing it while polling records the new meter's answer against the old archive. Store records with freeze time and communication address as the key. A later catch-up fills gaps only. It does not overwrite a billing value already confirmed.
A manual correction should keep a digest of the original frame and the reason for the change, so reconciliation can tell a meter reading from a later edit. After a concentrator power cycle, read the last freeze time it saved and continue from the gap, rather than reading again from midnight the intervals that already succeeded. When the uplink is congested, drop instantaneous values and keep the daily and monthly freezes. A meter that fails on the bus should be given a longer interval for a while. Otherwise one meter that never answers fills the whole poll cycle, and the other meters in the same pit go missing with it.