Remote Terminal Unit RTU

The box, and the thing that shares its initials

Functionally an RTU is two halves in one enclosure. One half is input and output: contact inputs that read a breaker or a lockout, analogue inputs that read a transducer or a meter, counters that accumulate energy pulses, and control outputs that drive an interposing relay.

The other half is a protocol stack that packages those points for a master station and accepts commands from it. Classic operation is polled — the master asks, the outstation answers — and both DNP3 and the IEC 60870-5 family also support reporting by exception, so a point that has not moved does not consume bandwidth on a link that may be narrow.

Because the box lives in a substation or a plant control building, it is normally specified against substation-hardened environmental and immunity requirements: IEEE 1613 and IEC 61850-3 for the equipment class, IEEE C37.90.1 for surge withstand on the wiring that reaches it. It is also fed from the station DC system rather than from an AC outlet, the same 110 or 125 V DC supply the trip circuits use, which is one more reason the charger and battery deserve their own line on a commissioning check sheet.

One naming collision is worth clearing up before it costs somebody a week. Modbus RTU is a serial transmission mode — binary framing with a CRC, as opposed to the ASCII mode — and a device speaking it is almost never a remote terminal unit in the sense of this entry.

The two live side by side on real datasheets: a Power Electronics controller lists its comm ports as RS-485 Modbus RTU and Ethernet Modbus TCP/IP, and Huawei's SmartLogger3000B lists Modbus-RTU together with IEC 60870-5-103 and DL/T645 on its RS-485 port and Modbus-TCP with IEC 60870-5-104 on Ethernet. Read those lines as transmission mode plus application protocol, not as a claim about what the box is for.

Where the RTU function physically lives varies, and the variation is a procurement question rather than an engineering one. Some projects install a dedicated utility RTU; many put the telecontrol stack inside the plant controller or a substation gateway that already concentrates the plant's data. Vendor sheets make the option structure visible. Power Electronics' PPC PRO lists its protocols as Modbus TCP and ADS for inverter control plus DNP3 with grid operators — device-side and utility-side protocols named separately on one line.

SMA's plant controllers go further and sell the telecontrol interface as an ordering choice: a protocol-converter option with selections for none, IEC 60870-5-104, DNP3 or IEC 61850-7-4, and a supported-protocol list where IEC 60870-5-104 is standard on some models while IEC 61850 and IEEE 1815 appear as upon request. A utility-facing protocol assumed rather than ordered is a change order later.

The point map is the deliverable

A point map is a table, and every row of it is an agreement. Each row carries an index number, a point type, what the point means in words, its engineering units, a scale factor, a deadband if it is analogue, and a direction. DNP3 organises those types as object groups: binary inputs in groups 1 and 2, analogue inputs in 30 and 32, counters in 20 and 22, control relay output blocks in group 12 and analogue outputs in 40 and 41, with static data available as class 0 and events assigned to classes 1, 2 and 3 so the master can poll urgent changes more often than background ones.

IEC 60870-5-104 does the same work with type identifications rather than object groups — a measured value in short floating point, a single command, a setpoint — and runs the 60870-5-101 application layer over TCP/IP. The mechanics differ; the obligation does not. Both ends must hold the same table.

Scaling is the field that gets people. Many points travel as integers, so a plant exporting 87.42 MW may be sending 8742 with a documented scale of 0.01, and the map is where that factor is written down. The failure mode is not that the plant and the utility disagree about physics — it is that one end loaded a different revision of the table than the other, so a number arrives correctly and is interpreted by a factor of ten or a sign.

Deadbands belong in the same document, because an analogue point configured to report only on a meaningful change is the reason the link stays inside its bandwidth, and a deadband set too wide is a display that looks alive while lagging reality. Sign convention deserves its own column and an explicit sentence about which direction is positive at the boundary; the sign-convention entry covers why that is never obvious.

Treat the map as a controlled document with a revision number, not as a spreadsheet somebody emails. Adding a point on site does nothing at all until the control centre loads it at the other end, and removing or renumbering one silently repoints everything after it. The practical rule is that a map revision is a joint action with a date, a version and a verification step, exactly like a settings sheet. What flows through those points once they are agreed — cadence, quality, what the market wants to see — is the telemetry entry's subject; this entry stops at the table and who signs it.

Hardwired, serial, Ethernet: three interfaces on one panel

Hardwired I/O is the oldest layer and the one that survives because it fails visibly. A status point arrives as a dry contact wetted from the station DC supply, and a command leaves as a contact closure into an interposing relay rather than as a message. Analogue points that are hardwired normally use a 4-20 mA loop, whose useful property is the live zero: a broken loop reads 0 mA, which is outside the signal range and therefore distinguishable from a genuine reading of zero.

Real equipment ships this layer as standard. SMA's MV power station sheet lists remote signalling as two dry contacts, NC/NO, for alarm and tripping, one of them a watchdog contact. Huawei's JUPITER station enumerates the signal types running through its control logic as digital inputs, digital outputs, analogue inputs, RS-485, PT100 and fan control, each with its own cabling. A plant single-line that shows only a network drawing is missing half of what has to be terminated.

Serial is the middle layer, and it persists for reasons that are partly technical and partly regulatory. RS-485 multidrops several devices on one twisted pair, tolerates long runs where an Ethernet segment would not, and carries DNP3 or IEC 60870-5-101 to a utility over a leased line, a radio or a fibre modem.

SMA's Power Plant Manager datasheet is representative of the scale involved: an RS-485 serial interface alongside two digital inputs and four digital outputs, which is a small, deliberate set of signals rather than a firehose. In North America there is a second reason serial links have stayed at utility boundaries, which is that NERC CIP requirements draw a line between cyber assets connected by a routable protocol and those that are not, so the connectivity choice at the demarcation has compliance consequences beyond bandwidth.

Ethernet is where new work lands, with DNP3 over TCP conventionally on port 20000, IEC 60870-5-104 on port 2404 and IEC 61850 MMS on port 102 — worth knowing because those numbers are what a firewall rule and a network diagram argue about. Fibre is the usual medium across the plant-to-utility boundary, since it isolates the two grounding systems as well as the two networks; Power Electronics' PPC PRO, for instance, carries six RJ45 ports plus two multimode fibre SC connectors on its communication switch.

One caution about reading interface lists: several are alternatives rather than additions. EPC Power's CAB1000 documentation notes that Modbus TCP, CAN and Modbus RS-485 are standard but only one is usable at a time. Count the ports you can actually run simultaneously before promising the utility a path that shares hardware with the plant's own.

Who owns the box, and where the plant's data stops

Two arrangements are common and both are defensible. In one, the interconnecting utility furnishes and owns the RTU, installs it in the plant's control building or in its own substation, and the plant delivers signals to a demarcation panel — a terminal strip, test switches, and a documented list of what appears on which terminal.

In the other, the plant owns a gateway that presents the agreed points to the utility's master, and the demarcation is a network port and a protocol rather than a row of terminals. Which one applies is set by the interconnection agreement and by local utility practice, so do not assume it from the last project. What changes with ownership is who may open the cabinet, who holds the credentials on the device, who is allowed to change a configuration, and who pays when a card fails at two in the morning.

The point map is the demarcation for data even when the sheet metal is not. Once a signal appears in that table, the utility operates on it: a dispatcher acts on the MW it displays and on the breaker status it shows, and the operating log records what the point said rather than what the plant knew. That makes several ordinary-sounding questions contractual. Who is responsible for the accuracy of a telemetered value and its scaling. Who is responsible for the availability of the path it travels.

What the plant is expected to do when the utility's master is down, as opposed to when the plant's own network is down. Get these into the division-of-responsibility matrix alongside the points list, because they are cheap to write and expensive to argue about after an event. Settlement is a separate question with a separate device — the revenue meter entry covers why an operational telemetry point is not a billing determinant even when both report the same megawatts.

Commissioning is where the map stops being a document. Point-to-point verification means somebody at the plant forces or produces a real condition while somebody at the control centre confirms what appeared on their display, point by point, in both directions, including the control outputs.

It is slow and it is the only test that catches a transposed index or an inverted contact. Schedule it with the utility rather than around them, record the result against the map revision that was tested, and keep the record with the site acceptance test package; the site-acceptance-test entry covers the wider ITP this sits inside.

Remote trip, remote close, and a megawatt limit

Supervisory control through an RTU is deliberately deliberate. The standard sequence is select-before-operate: the master selects a specific control point, the outstation acknowledges the selection and starts a short timer, and only then does the master issue the operate — DNP3 does this with a control relay output block, IEC 60870-5 with a select-and-execute qualifier on the command.

The output drives an interposing relay rather than a trip coil directly, and it is subordinate to the local hardware. A local/remote selector at the switchgear, device 43 in the ANSI/IEEE C37.2 numbering, blocks remote operation while people are working, and permissive interlocks sit in front of a close command so that a remote close cannot be issued into a condition the plant would not allow locally. If a remote command has ever appeared to be ignored, the selector is the first thing to check.

This path is supervisory, not protective, and the distinction is measured in time. An RTU command completes in the second or seconds range, which is fine for opening a breaker on an operator's instruction and useless for clearing a fault.

Protection-grade signalling — direct transfer trip from the utility substation, for example — rides a dedicated teleprotection channel and is engineered in cycles, independent of the telecontrol link; the protection-relay entry owns that scheme and the anti-islanding obligations that sit beside it. If a project drawing routes a trip that must operate in cycles through the same box that carries a 2-second telemetry scan, the drawing is wrong.

Curtailment arrives here too, and it arrives as data rather than as a contact in most modern schemes: an active-power limit or setpoint written to an analogue output point, sometimes a small set of discrete steps for coarser control, occasionally a simple enable that hands control to a market signal. The RTU accepts it and passes it to the plant controller, which is the thing that actually enforces the number at the point of interconnection net of losses; the power-plant-controller and automatic-generation-control entries own that loop. Jurisdiction decides both the obligation and its shape.

IEEE 1547-2018 requires each distribution-connected DER in North America to provide a local interface supporting one of IEEE 2030.5, IEEE 1815 (DNP3) or SunSpec Modbus; IEEE 2800-2022 addresses transmission-connected inverter-based resources; in Europe, where a member state applies Regulation (EU) 2016/631 to storage, larger units must be able to receive and act on a remote active-power instruction, with the national grid code fixing the interface.

ISO and RTO telemetry standards then add their own requirements on top, and some markets — CAISO's non-generator-resource model is the usual example — expect state of charge among the points. Confirm against the executed agreement rather than a generic list.

Two failure behaviours belong in the specification before they are needed. First, loss of link: what the plant does when the command path drops must be a pre-agreed state, whether that is holding the last limit, ramping to zero, or reverting to a local default, and it belongs in the same document as the point map because the utility has to know what to expect.

Second, stale-but-plausible data: DNP3 and IEC 60870-5 both carry quality flags — online, restart, communications lost — precisely so a dispatcher can tell a frozen value from a live one, and a map that discards the flags produces a display that looks healthy while showing a number from ten minutes ago. A heartbeat point that must change on a schedule, or a watchdog contact of the kind SMA's station sheet lists, is the cheap version of the same defence.