Quick Summary (TL;DR)
An appliance energy baseline is a repeatable description of how much electricity a device normally uses under stated conditions—not a single wattage read from its label. For a fridge it may be the median kWh per complete day; for a dishwasher, kWh per finished programme; for a television, watts in standby and kWh per viewing hour. A credible baseline records the measurement boundary, observation period, sample coverage and normal variation, then compares like with like when investigating a change.
In Home Assistant, prefer a genuine cumulative energy entity in kWh from a correctly rated monitoring plug or dedicated meter. Use a circuit CT when the appliance is hard-wired or a plug is unsuitable, but do not pretend a shared circuit reading identifies one device. If only instantaneous power in watts is available, the Integral helper can estimate kWh, with sampling limitations. Start with a simple seven- to fourteen-day log; longer observation is necessary for weather-sensitive or irregular loads.
This guide gives a practical measurement protocol, worked examples, helper configuration, appliance-by-appliance sampling recommendations, uncertainty checks and an anomaly worksheet. It does not assume that a product label, manufacturer app estimate or apparent change in the Energy dashboard proves a saving. It also separates UK Type G and continental European Type E/F hardware: a Schuko monitoring plug is not a suitable direct substitute for a UK fused plug.
Define the baseline before buying a meter
There are at least four different questions hidden in “how much does this appliance use?”. The answer depends on whether the load runs continuously, cycles by thermostat, follows a programme or sits mostly idle. A baseline should have a quantity, a time or activity denominator, and a scope. Write those three down before you start collecting data.
| Baseline | Best unit | Appropriate examples | What it does not tell you |
|---|---|---|---|
| Idle/standby power | W, ideally a range | TV, speakers, printer, coffee machine waiting | Total annual kWh without knowing time in that state |
| Daily energy | kWh/day, preferably median and range | Fridge, freezer, router, network cabinet | Seasonal behaviour or how much one programme uses |
| Energy per completed cycle | kWh/cycle with programme name | Dishwasher, washer, tumble dryer | How often the household runs the programme |
| Active-session energy | kWh/session plus session duration | Gaming PC, dehumidifier, electric oven | Typical month unless session frequency is known |
| Seasonal demand | kWh per heating/cooling degree or per comparable period | Heat pump, air conditioner, electric heating | An appliance efficiency rating without heat output and conditions |
| Circuit residual | kWh for a group of loads | Lighting circuit, kitchen sockets | Energy of a single appliance when others share the circuit |
Do not confuse the rating plate with the baseline. A kettle rated at 3 kW draws roughly that power when heating, but its energy depends on how long it heats and how often it is used. A compressor may draw tens or hundreds of watts while running, yet average much less across the day. A 2 W standby draw is only a meaningful annual number once the number of standby hours is established. The formula is energy (kWh) = power (W) × hours ÷ 1,000, for a constant load; for variable power, integrate the readings or use the device’s accumulated energy register.
Keep the energy boundary consistent. A plug measures the device and anything downstream of that plug, not necessarily the plug’s own upstream consumption. A CT on a dedicated branch may include more equipment than the named appliance. A manufacturer’s internal energy sensor may exclude some auxiliary loads. The whole-home meter includes losses and other loads that are not represented by your monitored plug list. These distinctions matter more than adding decimal places to the result.
Choose a measurement method that matches the appliance
For most plug-in appliances, a properly rated power-monitoring smart plugAliExpress price is the least disruptive instrument. It gives a directly attributable measurement without entering the distribution board. However, it may have a minimum reporting threshold, infrequent energy updates or an internal relay that can unexpectedly turn a critical load off. If a device is hard-wired, runs close to a plug’s limits, or must never be switched remotely, discuss a dedicated meter or circuit measurement with an electrician instead.
| Method | What is measured | Advantages | Important limitation | Typical fitting |
|---|---|---|---|---|
| Energy-monitoring plug | Load downstream of one socket adaptor | Good attribution, usually power and accumulated energy | Plug rating, relay behaviour, low-load accuracy, Wi-Fi/Zigbee reliability | TV, desk PC, dehumidifier within rating |
| Dedicated branch CT plus suitable voltage reference | Current/power of one circuit or phase | Non-invasive current sensing at conductor; useful for fixed loads | Correct phase, voltage reference, calibration and qualified fitting essential | Dedicated water heater or heat-pump feed |
| DIN-rail energy meterAliExpress price | Metered conductor(s) and energy registers | Documented registers, stable wired telemetry on suitable models | Board space, current rating, electrician, gateway and communication setup | Hard-wired appliance or sub-board |
| Manufacturer’s built-in telemetry | Internal electrical or software measurement point | No additional plug or CT; may expose operating modes | Some report estimated energy, cloud-only values or incomplete auxiliary load | Supported smart appliances and inverter subsystems |
| Whole-home meter plus subtraction | Residual after subtracting measured loads | Useful for discovering unmonitored demand | Cannot identify a specific device; errors and overlapping boundaries accumulate | Sanity check only, not per-appliance proof |
For a broader comparison of ESP32AliExpress price pulse, CT, PZEM and Modbus methods, see the existing ESP32 energy-monitoring methods guide. If you are planning a dedicated DIY CT installation, the ESP32 CT clamp guide covers the electronics; it does not replace an electrician’s assessment of mains work.
Current UK and European plug choices: check the exact model
“Smart plugAliExpress price” is not a universal electrical specification. Regional connector geometry, fusing, voltage, current, motor-load capability, metering support and firmware are independent checks. Two similarly named models can have different maximum loads, and an energy-monitoring feature shown in a vendor app does not guarantee that every measurement is exposed by every firmware revision in Home Assistant.
| Example documented model | Region and connection | Verified electrical / metering point | Home Assistant path and qualification |
|---|---|---|---|
| TP-Link Tapo P110 (UK product page) | UK Type G, 220–240 V AC | Manufacturer lists 13 A / 2,990 W maximum, 1/6 HP motor rating and energy monitoring; indoor only | TP-Link Smart Home integration uses local polling; device/firmware features and third-party compatibility must be checked |
| Shelly Plug S MTR Gen3 | Continental Europe Type E/F, 220–230 V AC | Manufacturer lists 12 A and 2,500 W resistive maximum, with power and energy meters | Official Shelly integration is local; not a UK Type G product |
| Shelly EM Gen3 with up to two CTs | Fixed installation, not a socket adaptor | Manufacturer describes two CT channels, local Wi-Fi operation and internal data storage | Shelly integration; correct CT selection and electrician fitting required |
| ESPHome power monitor (appropriate certified hardware) | Depends entirely on enclosure and hardware design | Power and daily-energy components available, but accuracy depends on sensor and configuration | Native ESPHome integration; DIY firmware does not make a mains device certified |
The UK P110 rating and the explicit prohibition on using it for EV charging appear on TP-Link’s UK P110 specification. The EU Shelly figures are from the Plug S MTR Gen3 technical specification. The Shelly EM Gen3 knowledge base describes its CT architecture and local operation. Check the label and current manual of the actual retail SKU; do not transpose the EU current rating or connector to a UK product with a similar name.
The Home Assistant TP-Link Smart Home integration reports local polling (normally every five seconds) and says some firmware needs third-party compatibility enabled. It exposes energy entities where the device supports them. The Home Assistant Shelly integration uses local communication and supports multiple Shelly generations, with generation- and firmware-specific behaviour. These are official Home Assistant integrations, not a guarantee that a particular regional SKU will provide every advertised sensor. Confirm entity availability after commissioning before purchasing multiple units.
Safety boundary: Use a UK-approved Type G monitoring plug in a UK socket and a correctly rated Type E/F model in a compatible continental European installation. Avoid travel adaptors, overloaded extension leads and devices with damaged connectors. A 13 A socket rating is not permission to exceed the lower rating of an inserted smart plugAliExpress price, extension lead, motor-load specification or the appliance manufacturer’s instructions. Electrical Safety First’s UK plug and fuse guidance explains BS 1363 and fused UK plugs; its socket-overloading advice covers thermal risks and extension leads.
For fridges, freezers, life-support equipment, sump pumps, essential networking and similar critical loads, consider a measurement-only solution or a relay device with carefully checked power-return behaviour. A remote-controlled plug is not inherently a safe unattended isolator for every appliance. Do not use a generic smart plug as an EV charger controller or substitute for a compliant EVSE. Any new CT, DIN-rail meter, conductor separation or consumer-unit modification should be specified and installed by a competent electrician under the applicable local rules, including BS 7671 in the UK.
Build a sampling plan instead of collecting random screenshots
A useful baseline requires enough time to observe the appliance’s complete operating pattern. Do not pick a “normal” day after seeing the results; choose the period and exclusion rules first. The durations below are practical starting points, not statistical guarantees. Record at least one complete operating cycle, preferably several, and extend the period when weather, occupancy or programme choices materially affect consumption.
| Appliance class | First useful window | Primary metric | Log alongside energy | Why a short test misleads |
|---|---|---|---|---|
| Router, NAS, network switch | 48–72 h; extend if backups run weekly | kWh/day and low/typical W | Backup jobs, disks, PoE devices, UPS topology | Nightly jobs, updates and UPS losses vary |
| Fridge or freezer | 7–14 full days; repeat in another season | Median kWh/day, spread, 24 h completeness | Room temperature, door opening, defrost, loading | Thermostat and defrost cycles are not uniform |
| TV and audio system | Several real viewing sessions plus standby periods | W standby, kWh/viewing hour | Brightness, streaming source, amplifier use | Standby may change with updates and network activity |
| Dishwasher or washing machine | 5–10 complete runs per programme | kWh/cycle by named programme | Load, water temperature, programme, drying option | A 10-minute power snapshot omits heating phases |
| Tumble dryer or dehumidifier | Several full cycles across conditions | kWh/cycle or kWh per operating day | Humidity, load mass, room conditions | Humidity and moisture change runtime |
| Electric heating / heat pump | Several comparable weather periods, then seasonal review | kWh/day versus outdoor conditions | Setpoints, weather, flow temperature, hot-water demand | A mild day cannot represent winter demand |
| Electric oven or hob | Multiple representative cooking sessions | kWh/session | Preheat, temperature, recipe, duration | Nameplate peak power is not session energy |
Where possible, start each run with the appliance in a defined state and stop only after it has returned to that state. For a dishwasher, record the accumulated-energy counter immediately before pressing Start and after the complete programme has finished, including drying. For a refrigerator, use consecutive 24-hour periods rather than sampling only compressor-on intervals. For a NAS, record whether UPS losses are included: a wall plug ahead of a UPS measures a different boundary from a smart PDU behind it.
Commission the sensors and verify the entity you will analyse
- Confirm the exact hardware variant, local socket type, nameplate current and environmental limits. Keep mains work outside the DIY software procedure.
- Install the vendor integration and update firmware only after reviewing release notes. For TP-Link, verify third-party compatibility settings if local access fails; for Shelly, check local network reachability and integration setup.
- In Settings → Tools → States, locate the device’s
sensor.*_powerandsensor.*_energyentities. Actual entity IDs vary. Record the unit and attributes rather than copying example names blindly. - Observe a known change: a lamp or other suitable small test load should produce a plausible increase in power. Check whether the energy counter rises after a sustained run, and whether it survives a device reboot.
- Watch the energy entity across an overnight period. Record any unavailable interval, resets, unexpected negative deltas or impossible jumps before using the history as a baseline.
- If there is no native energy entity, consider the Home Assistant Integral helper for a stable power sensor. Do not add an integral of power and the same device’s native kWh counter to the dashboard as separate devices.
For the Energy dashboard, Home Assistant requires correctly typed sensor attributes. The official Energy FAQ specifies an energy device_class with state_class: total or total_increasing, and suitable energy units. A power sensor has device_class: power and state_class: measurement. If a vendor supplies a lifetime counter, prefer it to a synthetic counter derived from infrequent power samples. Do not “repair” a mislabelled entity by multiplying a value until you have confirmed whether the source uses Wh or kWh.
Turn cumulative energy into daily, weekly and per-cycle baselines
The simplest repeatable calculation is a counter difference: energy used during interval = closing cumulative kWh − opening cumulative kWh. This only works when both readings are valid, refer to the same meter and have not crossed an unhandled reset. If your energy source reports individual interval deltas instead of a cumulative total, it needs a different configuration; treating deltas as a lifetime counter will produce nonsense.
Home Assistant’s Utility Meter helper creates daily, weekly or monthly counters from an appropriate source. Under Settings → Devices & services → Helpers → Create helper → Utility Meter, choose the appliance’s cumulative energy entity and the desired cycle. The exact UI wording may vary by release. For a resettable plug, review the Periodically resetting option; for an interval-delta source, review Delta values. The Utility Meter documentation explains why those flags are not interchangeable.
For an existing YAML-based installation, the following is an illustrative configuration. Replace the entity ID with your actual energy sensor, validate the configuration, and avoid duplicating a helper already created through the UI. This example assumes the source is a valid cumulative energy sensor; it does not convert watts into kWh.
utility_meter:
fridge_energy_daily:
source: sensor.kitchen_fridge_total_energy
cycle: daily
fridge_energy_monthly:
source: sensor.kitchen_fridge_total_energy
cycle: monthlyIf the device only provides watts, Home Assistant’s Integral helper estimates energy by integrating the power series. Select the actual power entity, time unit h, and a suitable prefix so the result is in kWh; choose an integration method deliberately. The official Integral documentation notes that sample frequency matters, and that the helper can update at a configured maximum sub-interval even if the source does not change. Sparse reports may miss a short heating burst or model a step change poorly. Where available, a properly maintained native energy register usually gives a better starting point for baselining than integrating a few sporadic wattage readings.
The ESPHome Total Daily Energy component can integrate power on an ESPHome device and reset at midnight using a time source. Its default method and restore behaviour are documented; neither makes the upstream sensor accurate by itself. For a long-term cumulative counter, ESPHome also documents the general Integration sensor. Avoid creating two independent integrations of the same raw power feed and summing both.
Worked example: a refrigerator’s normal daily consumption
Suppose a correctly configured monitoring plug recorded the following illustrative full-day values for one refrigerator. These are made-up teaching figures, not a test of any brand or model. The day with the largest value coincides with a particularly warm kitchen; we keep it visible rather than deleting it merely because it is inconvenient.
| Complete day | Energy (kWh) | Observation |
|---|---|---|
| Monday | 0.85 | Usual occupancy |
| Tuesday | 0.91 | Usual occupancy |
| Wednesday | 0.98 | Usual occupancy |
| Thursday | 1.14 | Warmer kitchen; more door openings |
| Friday | 0.90 | Usual occupancy |
| Saturday | 0.94 | Usual occupancy |
| Sunday | 1.02 | Guests / more door openings |
Sorted values are 0.85, 0.90, 0.91, 0.94, 0.98, 1.02 and 1.14 kWh. The seven-day median is 0.94 kWh/day; the arithmetic mean is about 0.963 kWh/day, and the observed range is 0.85–1.14 kWh/day. That range is useful context, not a calibrated uncertainty interval. A rough 30-day projection at the median is 28.2 kWh, but it is only a scenario if the following month resembles the observed week.
Do not multiply 0.94 by 365 and describe the result as measured annual consumption. Summer room temperatures, frost build-up, defrost events, holiday absences and appliance ageing can all shift the baseline. A stronger record uses several weeks in different seasons and reports the actual measured kWh for each month. If the fridge is moved or the thermostat setting changes, label a new baseline rather than quietly mixing pre-change and post-change periods.
For daily series with missing hours, avoid assuming the day was complete because a utility-meter sensor still displays a number. Compare the first and last valid timestamps, inspect availability gaps and, where possible, check whether the source meter continued accumulating while Home Assistant was offline. A meter that stores a cumulative total may recover the total energy used over an outage when it reconnects, but it cannot reconstruct the missing time-of-day profile from one closing reading alone.
Standby loads: watts are not an annual bill
Standby is particularly easy to overstate. First distinguish a device that is truly off from one that is in network standby, actively charging, updating firmware or waiting with a display illuminated. Measure a stable state for long enough to catch periodic wake-ups, and repeat after changing the settings that affect standby. At very low power, a plug may round readings to zero or have a measurement floor; a displayed 0 W is not proof of literally zero consumption.
| Illustrative scenario | Calculation | Energy consequence |
|---|---|---|
| Device draws 2 W continuously | 2 × 24 × 365 ÷ 1,000 | 17.52 kWh/year, if truly constant |
| Same 2 W only 10 hours daily | 2 × 10 × 365 ÷ 1,000 | 7.30 kWh/year, if routine persists |
| One device draws 0.5 W continuously | 0.5 × 24 × 365 ÷ 1,000 | 4.38 kWh/year; hard to validate with a coarse meter |
| Six devices at 2 W each, all continuous | 6 × 2 × 24 × 365 ÷ 1,000 | 105.12 kWh/year as a hypothetical aggregate |
The difference between 17.52 and 7.30 kWh is time in standby, not measurement precision. If the device uses 2 W for 10 hours, 60 W for three hours and 0 W otherwise, the correct daily energy is (2 × 10 + 60 × 3) ÷ 1,000 = 0.20 kWh, assuming those states are stable. A single “standby” figure cannot represent that profile.
The monitoring hardware itself consumes electricity. TP-Link’s UK P110 specification, for example, lists its own consumption as 1.64 W with relay on and 0.64 W with relay off. Those are manufacturer-specified device consumption figures, not a measured correction for your individual unit. Whether a particular exposed load-energy counter includes the adaptor’s own draw depends on its metering boundary; do not subtract the datasheet figure from the plug’s reading without evidence. When calculating net savings from adding a monitoring or switching device, account separately for the device’s overhead, its operating schedule and any required hub.
Per-cycle baselines: washers, dishwashers and tumble dryers
A dishwasher has a very different usage pattern from a refrigerator. Recording a 15-minute average while it heats water misses pumping, drying and the length of the programme. Use kWh per completed programme and keep programme names separate. For example, an Eco cycle, a high-temperature intensive cycle and a short rinse should not be averaged together to form a single “dishwasher efficiency” number.
Imagine five complete Eco cycles recorded as 1.05, 0.92, 1.13, 0.95 and 1.02 kWh. The median is 1.02 kWh/cycle, with an observed range of 0.92–1.13 kWh. If a household actually ran that programme four times every week for a year, the simple scenario would be 1.02 × 4 × 52 = 212.16 kWh. That is a conditional projection, not a savings claim or a prediction of anyone’s bill. The household may run different programmes, and water-inlet temperature and load size can change results.
Define the cycle boundary with the appliance’s real behaviour: some machines finish their main programme but continue fans or anti-crease movement. A switch entity that says “on” may remain on all day and is not a useful cycle detector. Prefer the device’s explicit running status where trustworthy, or combine a power threshold with an appropriate delay and a manual spot check. Use the History Stats helper to measure hours in a verified running state, not to infer energy from an arbitrary on/off label.
If you need an activity estimate, the History Stats integration can count time or occurrences of a particular binary state. It depends on retained recorder history for the requested period, and its count semantics are documented: a state already active at the start of the window can count. Compare detected runs with the appliance’s own completed-programme record before multiplying a “run count” by a median kWh/cycle.
Optional: detect active operation without hiding missing data
A threshold helper can be useful for identifying when a machine is active, but it must not treat an unavailable sensor as “off”. Choose the threshold only after inspecting real power traces. A fridge may draw low standby watts between compressor cycles; a washing machine may sit below a 20 W threshold for part of a valid programme. Hysteresis and delayed transitions help avoid rapid state flapping, but cannot solve ambiguous appliance behaviour.
One simple YAML example is a template binary sensor that marks a device active above 15 W and becomes unavailable when the source is unavailable. This is an illustration, not a universal dishwasher cycle detector. In current Home Assistant template syntax, replace the example entity and tune the threshold from measured data:
template:
- binary_sensor:
- name: "Workshop printer active"
unique_id: workshop_printer_active_power
state: >-
{{ states('sensor.workshop_printer_power') | float(0) > 15 }}
availability: >-
{{ is_number(states('sensor.workshop_printer_power')) }}
delay_on: "00:00:10"
delay_off: "00:01:00"The float(0) fallback above is safe only because the availability template explicitly rejects non-numeric source states; do not copy the numeric fallback without the availability check. Home Assistant documents template binary sensors and delays in its Template integration. The Threshold helper is another UI-based option for basic above/below detection. A manually verified run log remains the reference when the device has multiple quiet phases.
Use the Energy dashboard for context, not for every analytical question
The Energy dashboard is valuable for long-term device totals and for showing where monitored loads fit within the home. Home Assistant’s individual-device energy documentation supports suitable energy sensors and upstream-device hierarchies. It is not a substitute for a programme log, an uncertainty assessment or a carefully chosen median. A device’s contribution to a stacked energy chart does not automatically mean that all other household energy has been assigned to specific appliances.
Create a device hierarchy only when the physical measurement relationship is true. A smart plug downstream of a metered kitchen circuit is a child of that circuit; it is not an additional independent slice of household demand. Adding both the circuit and plug totals as unrelated sources and then summing them double-counts the same kWh. Likewise, a heat pump’s built-in compressor sensor may omit immersion backup or circulation pumps while a branch meter includes them. The boundary should appear in the entity name or documentation.
For anomaly work, keep a simple chart of daily kWh and annotate changes in weather, occupancy, firmware, appliance settings and measurement hardware. Home Assistant’s long-term statistics documentation explains that raw state history is kept for ten days by default, short-term statistics are recorded every five minutes and long-term statistics hourly for eligible sensors. Long-term statistics are not automatically purged, but they are aggregates: they do not preserve every second-by-second compressor or heater event.
The Recorder integration controls which entities are stored and the retention of raw history. Excluding an energy sensor from Recorder also prevents its long-term statistics. For a seasonal baseline, retain suitable energy entities and export a monthly summary rather than assuming an old high-frequency power trace will still be available next year. Before changing retention settings, consider database growth and storage endurance.
Circuit monitors: separate attribution from residuals
A branch CT is useful when the device cannot be plugged into an adaptor, but a shared circuit is not an appliance sensor. If a kitchen circuit includes a fridge, dishwasher and sockets, the circuit’s 5 kWh in one day does not tell you which load used 3 kWh. Subtracting two smart-plug totals can estimate a residual, but that residual also includes unmonitored loads, meter inaccuracies, sampling mismatch and standby losses.
For example, suppose a branch meter shows 6.0 kWh, a fridge plug 0.9 kWh and a dishwasher plug 1.1 kWh for the same day. The arithmetic residual is 4.0 kWh. It is not proof that an oven used 4.0 kWh; lights, sockets, hidden loads, timing and measurement uncertainty all remain. On three-phase installations, the CT’s conductor, voltage phase reference and signed direction must be correct; a current-only clamp with assumed voltage can be especially poor for small nonlinear loads.
For hard-wired monitoring, the existing ESP32 Modbus SDM120/SDM630 guide covers a different hardware approach. For an electricity-supply-level cross-check, the ESP32 electricity-meter pulse reader guide explains pulse-based metering. Both approaches require a clear definition of what their sensor actually measures.
How much uncertainty is acceptable?
There are several independent error sources. Instrument error comes from the metering hardware and its range. Sampling error arises when a fast-changing load is reported slowly. Coverage error comes from missing periods, restarts or sensor resets. Behavioural variation reflects genuine differences between days. Combining these into one unexplained “±5%” claim is misleading unless you have manufacturer accuracy data and a justified method.
| Issue | Symptom | Baseline consequence | Practical response |
|---|---|---|---|
| Low-power measurement floor | Standby reports 0 W or jumps in coarse steps | Underestimates small continuous loads | Check published minimum measurement/accuracy and use a suitable reference meter |
| Sparse polling | Brief heater spike absent from power history | Integral-derived kWh may be low or high | Prefer device cumulative energy and verify update cadence |
| Reset / counter wrap | Negative delta or sudden new zero | Spurious daily kWh or missing energy | Inspect source semantics and Utility Meter reset options |
| Unavailability / Wi-Fi outage | Flat line or gap while appliance still runs | Incomplete timeline; possible missed interval | Compare cumulative register after reconnection and mark uncertain days |
| Season / occupancy change | Different daily distributions | Baseline no longer representative | Segment by conditions; do not hide real variation |
| CT phase or direction error | Implausible negative, near-zero or doubled readings | Wrong measurement boundary | Qualified electrician to verify installation and reference |
| Overlap in dashboards | Plug plus parent circuit both counted | Artificially inflated home-device total | Model parent/child relationship and avoid summing duplicates |
A simple confidence label is more honest than false precision. High confidence might mean a suitable dedicated meter, verified cumulative kWh, complete days and repeated stable conditions. Medium might mean a reasonable sample with seasonal or behavioural variation. Exploratory means estimated energy from sparse watts, missing data or an unverified shared circuit. These labels are editorial judgement, not certification grades. If you need contractual or billing accuracy, use an appropriately approved meter and qualified metrology advice rather than a domestic smart plug.
For comparing before and after a change, record both periods under similar conditions. If a NAS update reduced idle power by 4 W but changed backup schedules, the observed daily kWh may rise despite a lower instantaneous idle reading. A weather-driven dehumidifier should be compared at similar humidity, not merely on consecutive calendar weeks. Repeat measurements, inspect spread and consider whether the difference is large relative to normal variation before attributing it to an optimisation.
Make the baseline useful: identify anomalies without alarm fatigue
The baseline becomes valuable when it answers a specific question: “Has the freezer started using materially more electricity?”, “Did the new router increase the network rack’s daily energy?”, or “Which dishwasher programme is cheapest under the household’s actual tariff?”. For a continuously cycling appliance, compare a rolling seven-day median with an earlier comparable baseline and examine the daily range. For a programme appliance, compare energy per cycle with the same programme, not calendar-day energy.
An anomaly rule should combine magnitude, persistence and context. As an illustrative starting rule only, investigate a fridge when its rolling median daily kWh is more than 20% above its own seasonally comparable baseline for seven complete days. This is not a validated diagnostic threshold: the suitable percentage depends on the normal spread and metering resolution. Review door seals, ambient temperature, loading, defrost activity and sensor health before blaming the compressor. Do not automate safety-critical power isolation from an energy anomaly.
For standby, a 1 W increase might be interesting on a small network device but below the resolution of some plugs. A 50 W increase on a 24-hour rack is 1.2 kWh/day and deserves investigation; it might be a new PoE camera or a disk that no longer spins down. Annotate equipment additions and firmware changes. For a dryer, a higher kWh/cycle may simply reflect wetter clothes or a different load. An actionable baseline is a comparison against a plausible explanation, not a leaderboard of appliances.
Purchase and installation cost: do the sums before instrumenting everything
A monitoring plug is often cheaper to deploy than a dedicated wired meter, but multiplying one plug by twenty devices can make the monitoring system expensive and power-hungry. Compare the whole system: purchase price, socket compatibility, networking, gateway, electricity consumed by monitors, cloud or subscription dependency, possible electrician labour, enclosure and ongoing maintenance. Retail prices and promotions change; this guide deliberately avoids inventing a “current best price” or a guaranteed payback.
| Option | Cost components to check | Recurring burden | Best value when |
|---|---|---|---|
| One movable monitoring plug | Plug, safe regional socket, possible setup app | Its own electricity, occasional firmware/network work | You want a short series of appliance audits |
| Several fixed monitoring plugs | Every plug, Wi-Fi capacity or Zigbee coordinator, socket clearance | Multiple radios, entity maintenance and data retention | Several critical baselines must remain visible |
| Dedicated CT or DIN-rail branch meter | Meter, CTs, protection/enclosure, comms gateway, qualified labour | Network/gateway energy and occasional inspection | One hard-wired or high-value circuit needs ongoing tracking |
| Existing appliance telemetry | Vendor gateway/app if required; API availability | Potential cloud dependency and vendor changes | The device exposes sufficiently reliable energy counters already |
| Whole-home monitor only | Main meter/CTs and qualified installation | Network and storage overhead | You mainly need grid or household trend, not per-device attribution |
Before buying, ask whether the baseline will change a decision. A portable plug used for two weeks on each of ten suitable appliances may be enough to identify the major standby loads. A continuously monitored freezer can justify permanent tracking for fault detection, but a rarely used table lamp may not. Where you already have a reliable energy entity, spending on a second sensor may add complexity rather than insight.
If the goal is financial savings, multiply verified kWh differences by the relevant marginal unit tariff, accounting for time-of-use periods if applicable. Do not multiply by the full bill average when that average includes standing charges or taxes unrelated to incremental consumption. The energy baseline alone does not prove that a smart plug saves money; an always-on monitoring plug may slightly increase consumption unless it enables a meaningful behavioural change.
A repeatable appliance-baseline worksheet
Use the following fields in a spreadsheet, notebook or Home Assistant note. The point is to make a result auditable six months later when the meter has been moved or the appliance replaced. A baseline without the meter identity and dates is difficult to reproduce.
| Field to record | Example of an acceptable entry | Why it matters |
|---|---|---|
| Appliance and exact model | Kitchen fridge; manufacturer/model from rating label | Avoids mixing devices |
| Measurement boundary | UK P110 between wall socket and fridge; no extension lead | Defines what is included |
| Sensor / entity | sensor.kitchen_fridge_total_energy in kWh | Links result to recorded source |
| Hardware/firmware | Meter SKU, firmware version, integration name | Supports later compatibility checks |
| Window and coverage | 1–14 October; 14 complete 24 h periods, no outages | Prevents misleading daily averages |
| Operating conditions | Kitchen 19–23°C; normal occupancy; no holidays | Explains variation |
| Metric and spread | Median 0.94 kWh/day; min 0.85, max 1.14 | Reports more than one headline number |
| Confidence and exclusions | Medium; one warm day retained; no resets | Makes uncertainty explicit |
| Action / next review | Repeat in January; investigate sustained increase | Turns measurement into a decision |
For a fixed daily baseline, the simplest analysis is to collect the completed daily utility-meter values, calculate the median, and plot the individual points. Averages can also be useful, especially for reconciling total kWh, but a median is less sensitive to a single unusually busy day. Do not discard an outlier automatically: investigate whether it was a genuine load, a missing day, a meter reset or a sensor error. Keep both the original reading and any corrected series, with the reason for correction.
Troubleshooting common Home Assistant baseline mistakes
| What you see | Likely checks | Do not do this |
|---|---|---|
| Appliance “uses 2,000 kWh today” | W/Wh/kWh unit mismatch, counter reset, invalid long-term statistic | Blindly divide the displayed total without inspecting source data |
| Energy Dashboard refuses the sensor | Device class, state class, unit, entity domain, statistics errors | Set total_increasing on arbitrary instantaneous watts |
| Daily helper resets at odd time | Home Assistant time zone, source counter semantics, DST boundaries | Compare periods with different start/end times |
| Smart plug says 0 W in standby | Resolution floor, off state, device wake cycle | Assume physically zero consumption |
| After Wi-Fi outage, graph is flat | Missing samples versus accumulated register recovered later | Invent an hourly profile from one closing counter |
| Circuit exceeds expected device total | Unmonitored loads, losses, overlapping measurement boundaries | Call the residual one appliance’s energy |
| Baseline rises after moving the meter | Changed boundary, firmware, placement, device settings | Attribute every difference to appliance deterioration |
When Home Assistant reports statistics errors, use the current Settings → Tools → Statistics workflow described in the Energy FAQ. Review the cause and the affected period before editing historical statistics. A counter reset can legitimately produce a decrease; an implausible giant spike can permanently distort long-term energy charts if left untreated. If you changed the entity ID, confirm that the long-term statistic identity and source association still match your intended appliance.
Frequently asked questions
How many days should I measure a fridge?
Seven to fourteen complete days is a practical first baseline if it captures normal thermostat and defrost behaviour, but it is not an annual estimate. Repeat in different ambient conditions and check for missing data. If the device has a known long defrost interval or the household is away, extend the window and annotate the unusual conditions.
Can I use a smart plug on a washing machine or tumble dryer?
Only when the exact regional plug is suitably rated for the appliance’s current, power, starting characteristics and manufacturer instructions. Some motors, heaters and continuous high loads are inappropriate for particular relays even when the nominal watts appear acceptable. Never assume a plug’s resistive rating covers motor loads. For uncertain or high-power applications, consult the manufacturer and a qualified electrician; a dedicated meter may be preferable.
Can I infer the cost of each appliance from a CT on the incoming supply?
Not reliably from the total alone. A whole-home CT measures aggregate demand, so coincident loads are mixed. Non-intrusive load monitoring algorithms may attempt disaggregation, but that is an inference with model-specific limitations, not an independently metered appliance baseline. Use direct measurements for devices where attribution matters.
Why does a daily utility meter disagree with a smart plug’s app?
Check the time zone, midnight boundary, whether the app shows calendar day versus rolling 24 hours, and whether the source reports lifetime energy or a resettable total. Also check outages, device time synchronisation and rounding. A plug’s app and Home Assistant may aggregate at different times even when they use the same underlying meter.
Should I add every plug to the Energy dashboard?
Add devices whose individual energy totals you want to monitor and whose statistics are sound. Use parent/child relationships when a plug sits inside a separately metered circuit, and avoid presenting the same kWh as two independent household loads. For brief portable audits, a temporary device chart and worksheet may be clearer than a permanently crowded dashboard.
Is a power reading in watts enough?
It is enough to study instantaneous states and, with appropriate sampling and an Integral helper, to estimate energy. A reliable cumulative energy counter is usually preferable for per-day or per-cycle kWh. If power updates are sparse, a short high-power event can be missed. Validate any derived energy against a known meter over a complete interval before treating it as a baseline.
What is the difference between a baseline and a savings estimate?
A baseline describes observed use under stated conditions. A savings estimate compares that baseline with a justified alternative and accounts for changes in conditions, tariffs and monitoring overhead. A lower reading after a change is evidence to investigate, not proof that the change caused the entire difference.
Further reading and first-party references
These links are to official Home Assistant, ESPHome, vendor or electrical-safety documentation. They support the technical distinctions above; product compatibility and firmware behaviour should be rechecked against the exact device at the time of purchase or installation.
- Home Assistant — Individual device energy — individual energy entities and upstream hierarchies.
- Home Assistant — Energy FAQ — sensor requirements and statistics troubleshooting.
- Home Assistant — Integral — energy from power, integration method and update intervals.
- Home Assistant — Utility Meter — reset cycles, tariffs, delta and periodically resetting sources.
- Home Assistant — Statistics — recent measurements, sample limits and time-weighted characteristics.
- Home Assistant — History Stats — duration/count from recorded binary states.
- Home Assistant — Long-term statistics — state history versus long-term hourly statistics.
- Home Assistant — Recorder — retention, database and exclusion behaviour.
- Home Assistant — TP-Link Smart Home — local polling and firmware compatibility caveats.
- Home Assistant — Shelly — local integration, generations and update behaviour.
- ESPHome — Total Daily Energy — power integration, time source and restore behaviour.
- TP-Link UK — Tapo P110 — Type G region, 13 A, own consumption and EV exclusion.
- Shelly — Plug S MTR Gen3 — EU connector and 12 A/2,500 W resistive rating.
- Shelly — EM Gen3 — two-CT measurement and local operation.
- Electrical Safety First — Plugs and fuses — UK BS 1363 and fused-plug guidance.
- Electrical Safety First — Overloading sockets — safe load limits and extension leads.