Quick Summary (TL;DR)
Safe load shedding is a hierarchy of permissions, not a clever way to trip household loads just before the main fuse operates. Home Assistant is useful for supervisory decisions: it can detect sustained high grid import, recommend pausing a discretionary load, ask a supported EV charger to reduce its target within the manufacturer-approved control interface, or provide the owner with a warning. It must not replace the supply’s overcurrent protection, a certified charger load-management system, an inverter’s battery-management system, or a heating appliance’s internal safeguards. The most important design decision is what Home Assistant must never interrupt.
Best starting point: measure import power and per-phase current accurately, classify loads as essential, conditionally flexible or genuinely discretionary, and observe for two weeks. Use advisory-only automations first. Where automatic control is justified, prefer a device’s supported native pause or current-limit feature over switching its mains supply. Include a manual override, a maximum interruption time, an explicit data-validity requirement and a documented recovery sequence. A smart plugAliExpress price is appropriate only for a compatible, rated, safely restartable plug-in load—not as a universal substitute for an approved contactor or a dedicated electrical protection device.
This guide explains priority tiers, high-load scenarios, control architectures, hysteresis, restoration order, the difference between power and energy, three-phase complications, Home Assistant example configurations, costs and commissioning. It is written for UK and European readers, but electrical approval, supply limits and installation rules vary by country. Installation and modification of fixed mains wiring must be designed, installed and tested by an appropriately qualified professional in accordance with local rules.
What Is Load Shedding, and What Is It Not?
A home typically draws electricity from several different devices at once. An oven, induction hob, immersion heater, EV charger and heat pump can create a relatively brief peak far above the ordinary daytime load. Load shedding means temporarily reducing or stopping selected, suitable loads when a defined constraint is approached. That constraint might be a contractual connection limit, a battery inverter’s continuous output capability, an off-grid generator’s capacity, or a tariff incentive. The constraint needs to be identified first: a high electricity bill and an overloaded service connection are different problems.
Three approaches are frequently confused. Load monitoring measures demand and reports it. Load management adjusts consumption by controlling a supported device or process. Electrical protection disconnects unsafe fault or overload conditions using equipment designed and installed for that purpose. Software monitoring may inform load management, but neither should be claimed to provide certified overcurrent, residual-current, earthing, isolation or thermal protection. You cannot make an unsuitable cable, plug, relay or socket safe by adding a Home Assistant automation.
For example, a charger with approved dynamic current limiting can measure household import and automatically reduce vehicle charging. Its installer-configured mechanism responds independently of whether your dashboard is running. A general smart-home controller may then provide higher-level instructions such as choosing an off-peak schedule. The second layer is useful but does not replace the first.
| Function | What it does | Appropriate lead controller | What failure should mean |
|---|---|---|---|
| Supply protection | Protects conductors and people from electrical faults | Properly designed protective devices and installation | Protective function remains available |
| EV charger dynamic load management | Limits charging to approved service/circuit capacity | Charger and manufacturer-supported CT or meter arrangement | Charger uses its documented safe response |
| Inverter/battery limits | Enforces battery, inverter and discharge constraints | Inverter, BMS and certified transfer equipment | Native protection remains in force |
| Home Assistant load advisory | Warns or prioritises discretionary activity | Home Assistant with valid measurements | No assumption that protection depends on HA |
| Flexible load scheduling | Moves a compatible load to another time | Manufacturer controller or supported local API | Manual use and safe device fallback |
| Emergency grid outage | Changes available sources or disconnects supply if required | Certified switching/protection architecture | Home Assistant is an observer, not a substitute |
Begin with the actual constraint
A household may have a single-phase or three-phase supply. The rated main fuse or service current is not automatically the same as the tariff limit, the inverter’s permitted output, or the target threshold for a scheduling system. The installer must establish the relevant electrical limit and whether simultaneous phase currents, neutral conditions and diversity matter. In a three-phase installation, 12 kW total demand can conceal an unbalanced phase carrying considerably more current than the others. Do not infer spare capacity from summed kW alone.
The useful supervisory quantities include real-time grid import power (W or kW), optionally current per phase (A), the state of controllable loads, and a reliable indication that readings are recent. Cumulative import in kWh is excellent for invoices but is generally too slow and indirect for near-real-time decisions. Some utility integrations update only every few minutes or hours. A cloud account’s historical kWh data should not be mistaken for a certified current-limiting measurement.
Decide Which Loads Must Never Be Shed
The safest shedding algorithm starts as an exclusion list. If losing power creates a safety risk, ruins essential work, interrupts medical equipment, defeats a pump or causes an unsafe restart, it should not be part of a casual automation. Even apparently discretionary loads can become time-critical depending on who is home, which season it is and how the equipment operates.
| Priority | Examples | Typical treatment | Special caveat |
|---|---|---|---|
| P0 — Never shed | Medical equipment, fire/security functions where continuity is required, essential networking for alarms, critical sump or sewage pumps | Exclude from Home Assistant shedding entirely | Separate resilience and backup design may be required |
| P1 — Essential/condition-dependent | Refrigeration, essential lighting, accessibility devices, heating in freezing conditions | Normally exclude; dedicated protective or backup arrangements | Avoid interrupting compressor or frost protection cycles |
| P2 — Flexible under manufacturer rules | EV charging, thermal storage heating, supported hot-water generation, battery charging | Prefer native pause/setpoint interface | Observe minimum-off time, departure and hygiene constraints |
| P3 — Discretionary and safely stoppable | Certain scheduled plug-in noncritical chargers or devices, selected low-risk equipment with verified restart | Candidate for approved device control | Check plug rating, restart and unattended-use risks |
| Not classifiable without assessment | Washing machine, dishwasher, tumble dryer, dehumidifier, portable heater, aquarium equipment | Do not automate power interruption by assumption | Water leaks, heat, pumps, refrigeration or stalled cycles may result |
The classification is a household-specific hazard assessment, not a universal product whitelist. A freezer in a monitored outbuilding, an aquarium heater, a home office workstation or a bedroom heater for a vulnerable occupant may be essential in that household. Build an explicit never_shed register and revisit it whenever people or equipment change.
Safety exclusions need two forms: a policy decision that the automation never issues a control command, and a physical/system architecture in which a failure elsewhere cannot inadvertently remove the load. Merely choosing not to add an entity to a dashboard is not enough if an upstream relay or whole-circuit contactor still feeds that appliance.
The danger of interrupting finished cycles
A washing machine, dishwasher or tumble dryer is not necessarily a useful shedding load merely because its nameplate power is high. Removing power mid-cycle can leave water standing, prevent a fan’s cooling run, cancel a programme or cause unexpected behaviour when supply returns. Unless the manufacturer documents a suitable remote pause/resume function and your model exposes it, a notification to postpone starting the appliance is safer than interrupting it. Smart plugs should not be used to interrupt loads exceeding their current, inrush or duty ratings; contact and enclosure temperature matter as well as the nominal ampere figure.
For hot water, legionella and scalding protection are not optional conveniences. A planned shift to a cheaper period must preserve the manufacturer’s temperature protections and any applicable hygiene programme. For heat pumps, frequent shutdown and restart can increase cycling and violate intended compressor operating constraints. For EV charging, use the charger’s supported control interface with an approved load-balancing arrangement: do not repeatedly cut supply to the wallbox as an improvised current regulator.
The Measurements You Need Before Controlling Anything
A credible decision begins with a meter whose location and units are known. For a grid-import constraint, the measurement must represent the correct boundary—normally total power or per-phase current at the property’s grid connection, not simply the solar inverter’s output or a single branch circuit. Where export and import are signed, confirm the convention with a real load change. Many sites show positive import and negative export; others expose separate nonnegative import and export entities. A wrong sign can make the algorithm shed loads on a sunny day while failing to react to true import.
| Measurement | Why it matters | Minimum validation before use | Common mistake |
|---|---|---|---|
| Grid import power, W | Identifies sustained loading | Compare a known appliance step and confirm sign | Using export-negative power as an always-positive import value |
| Per-phase current, A | Detects phase-specific constraints | Match L1/L2/L3 to documented connection | Dividing total kW by 230 V on three-phase supply |
| Load demand/status | Estimates available controllable reduction | Confirm that reported state matches reality | Assuming a commanded OFF state proves zero watts |
| Device availability | Identifies communication loss | Disconnect network intentionally during tests | Treating cached values as live measurements |
| Meter timestamp or freshness | Distinguishes recent from stale information | Observe actual update cadence | Using a retained MQTT reading indefinitely |
| Native equipment alarms | Detects failed or inhibited control | Read manufacturer documentation for each fault | Clearing an alarm with a generic reset command |
| Cumulative kWh | Validates longer-term effect | Reconcile with utility and appliance meters | Using an hourly energy feed for sub-second decisions |
A split-core CT sensor needs correct ratio, phase assignment and orientation; a wired DIN-rail meter requires suitable installation and isolation by an electrician. Do not ask Home Assistant to compensate for uncertain electrical wiring through arbitrary calibration factors. Our detailed guide to whole-home monitoring without cloud dependencies covers selection and topology; the bill-reconciliation guide explains checking cumulative totals.
Refresh rate is not the same as protection response
Suppose a meter updates every five seconds, the local integration adds one second and the automation requires ninety seconds of sustained loading. The supervisory response will take at least about ninety seconds, potentially more if a sample is missed. That can be acceptable for tariff optimisation or reducing average demand, but cannot be treated as protection against an electrical overload. Even one-second telemetry cannot guarantee bounded response if the device, broker, Wi-Fi, host or automation is unavailable. Supply and device limits must remain enforceable without that software chain.
Load Shedding Architecture: Three Layers, Three Responsibilities
A robust architecture distinguishes hard protection, native equipment management and supervisory scheduling. The installer configures the appropriate supply protection and certified current-limiting devices. The manufacturer’s device controls remain the final authority for compressor, charger, battery and immersion safety. Home Assistant reads operating measurements and requests only documented discretionary actions. This separation keeps the system understandable during faults and future modifications.
| Layer | Example hardware or software | Permitted role | Not permitted as its substitute |
|---|---|---|---|
| Installation | Service protective device, RCD/RCBO, appropriately rated circuits | Electrical fault and overload protection | Smart relayAliExpress price, automation or MQTT watchdog |
| Device controller | EV charger with approved CT/meter, heat-pump controller | Native current, thermal and process limits | Generic contactor cycled by HA |
| Supervisory control | Home Assistant, local MQTT, Modbus data, user dashboard | Advise, schedule and request supported power reductions | Certified safety or protection function |
| Measurement | Certified meter/approved CT installation | Trustworthy input within its rated boundary | Assumed power inferred from a device name |
| Operator | Manual override, physical labelled controls, clear alarm reporting | Recover usable service and understand shedding | Unlabelled app-only control with no fallback |
Use native power modulation before binary disconnection
When a device offers a documented dynamic power setpoint, reducing its demand usually creates fewer problems than cutting its supply. An EV charger might offer charging-current adjustment or pause/resume. A heat-pump interface may permit a manufacturer’s authorised demand-response signal, but an arbitrary power command must never override its protective logic. An immersion diverter may manage the resistive load internally. A home battery inverter may expose import, charging or reserve requests, yet BMS and grid-code constraints remain inviolable.
A home-automation smart plugAliExpress price is a comparatively narrow tool. It should be used only for an appropriately rated, plug-in, noncritical appliance that is expressly suitable for unattended remote interruption and subsequent resumption. For any fixed circuit modification or DIN-rail contactor installation, obtain a competent electrician’s design covering protective coordination, breaking capacity, thermal behaviour, contactor utilisation category, coil supply, manual isolation and testing. This guide intentionally gives no consumer-unit or mains-wiring instructions.
Local control reduces one failure mode, not every failure mode
Locally integrated products avoid dependence on a vendor cloud for routine monitoring. For instance, the official Home Assistant Shelly integration states that it communicates directly with supported devices and does not require the cloud. That still leaves your LAN, switch/router, the device itself and Home Assistant as potential failure points. Local control is useful but it is not inherently fail-safe. Read our EV charger connectivity comparison for practical offline distinctions.
Build a Priority Policy Before Writing Automations
A written policy should fit on one page. Its purpose is to make decisions stable under routine household activity and understandable to somebody other than the person who wrote the YAML. Record the constraint, the loads available to relieve it, how long they may remain deferred, whether they need user consent, and what automatically restores them. Include a clear disabled/monitor-only mode and who can override it.
| Policy field | Example for illustration | Why document it |
|---|---|---|
| Managed constraint | Advisory import threshold, not the service-fuse rating | Avoid false claims of electrical protection |
| Stage 1 | Request reduced EV target through approved API | Largest easily adjustable load first |
| Stage 2 | Defer a scheduled noncritical load not yet running | Avoid mid-cycle appliance interruption |
| Exclusions | Medical, refrigeration, pumps, essential heating, safety systems | Prevent unintended critical interruption |
| Data validity | Accept only recent, plausible readings | Stale MQTT/Modbus values must not count as live |
| Entry dwell | Above advisory threshold continuously for 90 s | Filters brief cooking or kettle spikes |
| Exit dwell | Below lower threshold continuously for 5 min | Prevents oscillation |
| Manual override | User can disable supervisory control | Supports guests, servicing and emergencies |
| Maximum interruption | Per device; e.g. departure deadline for EV | Maintains service requirements |
| Offline state | Native equipment settings continue; HA sends no new command | Keeps software loss predictable |
Start by stating which constraint you are actually solving. If the goal is reducing the highest fifteen-minute billing demand, the control approach differs from a charger that must never exceed a circuit’s current limit. Similarly, a home battery limited to 3 kW continuous discharge cannot be made safe by a supervisory response three seconds later. Native protections and approved changeover/distribution systems must handle fast or safety-relevant events.
Thresholds, Hysteresis and Restoration Order
One threshold that turns a load off at 6 kW and back on at 6 kW will chatter whenever normal power readings fluctuate around that value. Hysteresis uses separate enter and exit thresholds, plus dwell times. A simple advisory system might request Stage 1 when validated import stays above 6.0 kW for ninety seconds, then allow recovery only after import remains below 4.5 kW for five minutes. These are deliberately illustrative numbers, not proposed UK fuse limits or approved design settings. The right values depend on the actual connection, inverter, household loads, sampling delay and available headroom.
| Parameter | Example | Result if too aggressive | Sensible design check |
|---|---|---|---|
| Enter threshold | 6.0 kW import | Unnecessary shedding | Establish normal short peaks first |
| Exit threshold | 4.5 kW import | No stable recovery | Ensure margin exceeds typical measurement noise |
| Entry persistence | 90 s | Responds to transient kettles | Define whether fast action is even needed |
| Exit persistence | 5 min | Repeated on/off cycles | Observe actual household load variation |
| Stage separation | Minimum observation between requests | Several loads stopped simultaneously | Measure realised power reduction first |
| Maximum shed interval | Device-specific | Occupant loses essential service | Include a user-visible deadline |
| Retry limit | Limited unsuccessful actions | Automation floods device/cloud | Alert and stop retrying on failure |
| Default on missing data | No new supervisory actuation | Stale decisions and bad restoration | Require fresh data before resuming |
A reliable control loop should observe the effect of one change before taking another. If charging is reduced by 2 kW, wait for the charger and grid meter to confirm the new operating point. If the command failed, do not assume that 2 kW has been removed. Equally, after loads are restored, wait for the new meter reading before releasing the next load. Using a fixed delay without feedback can produce the very overload or oscillation the automation was supposed to avoid.
Restore in the reverse order of necessity—not necessarily the reverse command order
Priorities and deadlines determine restoration. An EV departing soon may need its charge allocation returned before the dishwasher is allowed to start. An optional heater may remain deferred if the room is occupied and comfortable, but never if safety or frost protection depends on it. Every load needs a documented maximum permissible interruption and either a native autonomous resumption strategy or a clear user action. Silence is not a valid recovery plan: if a manual restart is required, notify the occupant.
A single global switch.turn_on to restore everything defeats staged recovery. Each load’s controller should verify that the circuit is available, the appliance’s process permits resumption, and the load remains inside the agreed discretionary policy. Smart-home scenes that re-enable all switches after a restart are a particular hazard when they include high-power devices.
Worked Example: A Home with EV Charging, Cooking and an Immersion Heater
Consider a fictional single-phase home whose installer has already commissioned the electrical supply and whose EV charger uses its own approved current-management system. This example is about comfort-oriented supervisory prioritisation, not making the connection safe. The homeowner wants the car charged by 07:00 while avoiding unnecessary evening load peaks. Assume the grid meter reports 2.0 kW of ordinary demand, a variable 2.2–3.0 kW hob load, a 3.0 kW immersion heater with its own suitable controller, and up to 7.0 kW charging when the charger permits it. The equipment cannot all be treated as controllable in the same way.
| Load | Example demand | Control owner | Supervisory action |
|---|---|---|---|
| Base household load | 2.0 kW | Occupants and individual appliances | Observe, do not shed |
| Induction cooking | 2.2–3.0 kW | User and hob controls | Do not interrupt |
| EV charger | Up to 7.0 kW | Charger, approved meter/CT and vehicle | Request supported lower charging target |
| Immersion heating | 3.0 kW | Certified immersion/cylinder controls | Defer discretionary heat-up if manufacturer supports it |
| Fridge, freezer, sump pump | Included in base | Native safety/function controls | Never shed |
At a particular moment, cooking and ordinary consumption might total 4.6 kW. If the charger is drawing 3.0 kW, grid import could be approximately 7.6 kW before accounting for any solar or battery contribution. When Home Assistant observes this, it can suggest further reducing EV demand within an approved EV management interface or notify the owner that an optional hot-water schedule has been deferred. It must not instruct the charger to exceed its independently enforced limits, infer per-phase currents from a simple total, or switch off the cooking circuit.
Suppose the house has 2.5 kW of solar production and a fixed 0.5 kW of export at the same time. Then the grid-import measurement is lower than the total gross appliance demand; that is expected. Control must use the boundary relevant to the managed constraint. Grid import, gross household consumption and inverter output are not interchangeable: an inverter might have its own current limit even when grid import looks small.
A rule that merely sees the 7.6 kW reading and switches a 3 kW load off is not automatically a good decision. What if the hot-water heater is not actually on? What if EV charging has already paused? What if solar export was accidentally reported as positive import? Before each action the controller should check current eligibility, device availability and measured or confirmed response. When an action fails, it should report failure rather than pretending the removed load exists.
Why the avoided peak does not equal energy savings
Deferring a 3 kW water heater for thirty minutes removes up to 1.5 kWh from that interval. It does not necessarily remove 1.5 kWh of total daily energy. The heat may simply be supplied later. Load shedding can reduce a peak, improve use of cheap-period electricity or avoid an agreed capacity charge, but household kWh savings only occur when energy use is genuinely reduced or when efficiency changes. For hot-water and space-heating loads, losses or heat-pump COP can make the total energy outcome better or worse depending on when and how heat is produced.
| Quantity | Definition | Example | Interpretation |
|---|---|---|---|
| Instantaneous power | Rate of energy use | 3 kW | Load size, not energy saved |
| Deferred energy | Power multiplied by time | 3 kW × 0.5 h = 1.5 kWh | Energy shifted if heat-up happens later |
| Genuine avoided energy | Demand permanently removed | 0.2 kWh fewer optional tasks | Only this changes total consumption directly |
| Import cost effect | Change in imported kWh by tariff | Shift 1.5 kWh from 30p to 15p/kWh | Nominal saving 22.5p before losses |
| Export opportunity cost | Lost payment when solar is diverted | 1.5 kWh at 8p/kWh | An additional economic consideration |
| Capacity-peak effect | Reduced metered peak for an applicable tariff | 1 kW lower interval peak | Valuable only if the tariff actually charges for it |
The tariff example above uses illustrative prices, not live rates or a savings promise. A household on a conventional domestic tariff without a demand charge may gain little financial benefit from simply suppressing a brief peak. It can still benefit operationally if the manufacturer-approved management system reallocates flexible demand safely.
Worked Example: A Three-Phase Supply with One Busy Phase
Consider a three-phase dwelling with single-phase cooking circuits and an EV charger whose phase use varies with operating mode. A total import of 10 kW might look modest compared with a notional three-phase capacity, yet the individual phase currents can be unequal. For an approximate resistive unity-power-factor load, current on a phase is I ≈ P / V. If Phase L1 carries 6.0 kW at 230 V, its current is about 26 A. That is not an estimate of a whole service rating: it simply demonstrates why phase identification matters. Actual power factor, voltage and meter configuration affect the measured result.
| Phase | Illustrative import | Approximate current at 230 V and PF 1 | Why it matters |
|---|---|---|---|
| L1 | 6.0 kW | 26.1 A | May be near a site-specific threshold |
| L2 | 2.5 kW | 10.9 A | Apparently spare capacity on another phase |
| L3 | 1.5 kW | 6.5 A | Lower loading |
| Total | 10.0 kW | Not a valid per-phase current | Summing power hides phase imbalance |
An automatic reduction request should therefore be based on the equipment’s approved phase-aware logic when supply current is the relevant limit. A Home Assistant dashboard can plot three phase currents and warn about unusual imbalance, but it cannot make phase-unaware load-shedding decisions electrically equivalent to a certified limiter. With power electronics and reactive loads, kVA, kW and amperes can diverge; do not use a simplified kW calculation to size wiring or protective devices.
Comparing Hardware: Smart Plug, Native API, Approved Relay and Meter
There is no universal “load shedding module” that suits every appliance. The correct hardware follows from whether the load is plug-in or fixed, whether a safe remote pause exists, and whether the device’s internal controller requires continuous supply. A cheap smart plugAliExpress price may offer useful power telemetry on a small resistive load, but the cheapest relay cannot be assumed to handle induction motors, large inrush currents, continuous high current or safety-critical equipment. Buy to an approved installation design, not to an attractive current number on a marketplace listing.
| Hardware/approach | Strong fit | Weak fit / exclusion | What to verify |
|---|---|---|---|
| Charger manufacturer’s dynamic limiter | EV charging under supply constraints | Other non-EV household loads | Exact charger, CT, firmware, topology and commissioned limits |
| Native local EV charging API | Scheduling and discretionary target changes | Fast protection in lieu of approved limiter | Offline behaviour, write permissions, acknowledgement and fallback |
| Manufacturer demand-response heating input | Explicitly supported heating curtailment | Arbitrary power cut to compressor | Installer documentation, minimum run/off times and frost protection |
| Appropriately rated smart plugAliExpress price | Suitable, noncritical plug-in device | Fixed circuits, unsafe restart loads, high inrush | Plug/socket rating, load type, thermal conditions and certifications |
| Approved contactor / relay system | Professionally designed fixed-load control | DIY consumer-unit experimentation | Utilisation category, coil control, enclosure, isolation and protection |
| DIN-rail or clamp-on monitoring meter | Read-only demand monitoring | Protection without a proper limiter | Accuracy, phase mapping, permissions and installer access |
| UPS for controller/network | Better logging and connectivity resilience | Main-load backup without a properly designed system | Runtime, safe shutdown and failure modes |
Smart plugs: what “16 A” does and does not tell you
A plug advertised as “16 A” does not establish that it is appropriate for every 16 A load. Switching a resistive load differs from controlling a compressor or motor with starting current. Plug shape, national socket ratings, conductor quality, enclosure temperature, surge current, duty cycle and the manufacturer’s tested use cases matter. In the UK, the plug/socket context can be particularly different from continental sockets. Never treat a smart plug as a replacement for fixed installation protection or place an underrated adaptor in a high-load path.
A good buying question is not merely “does it integrate with Home Assistant?” It is “can I explain why this particular load can be interrupted at this particular time, which device is electrically rated to do it, and what it will do when power returns?” If you cannot answer every part, stick to observation or supported manufacturer control. Smart-plug telemetry can still provide valuable consumption baselines without controlling the outlet. See our energy-monitoring smart plug comparison and the meter technology buyer’s guide.
Contactors: an electrical design choice, not a YAML accessory
A properly chosen contactor can be part of an approved building-controls installation, but its physical installation, overload protection, isolation, coil power, thermal dissipation and auxiliary feedback are matters for the qualified electrician. The contactor’s continuous-current rating alone is insufficient: the manufacturer’s utilisation category and duty rating matter for the load. A mechanical contactor also cannot provide load-current feedback unless monitoring is separately arranged. A relay’s internal switch position, a software-reported state and actual measured power can disagree.
Where equipment provides an accessory control input, such as a certified demand-response or external enable contact, verify that the manufacturer actually permits the proposed supervisory use. A “voltage-free” relay contact does not make arbitrary thermostat, inverter or EV control terminals interchangeable. Avoid bypassing a device’s electronic control board, compressor delays or safety interlocks.
Rough costs and ownership considerations
A basic local monitoring approach may cost less to own than retrofitting several circuits with professional contactors. A more expensive controller can be justified if it has approved current management, good fault reporting and a tested fallback. Since regional pricing and installation scope vary, compare itemised quotations rather than rely on a fabricated universal market price. The following is a decision worksheet, not price guidance.
| Component | One-off cost to request in quote | Recurring consideration | Hidden cost |
|---|---|---|---|
| Grid/phase meter | Device plus installation and commissioning | Calibration/support if applicable | Meter location and enclosure space |
| EV charger with compatible management | Charger, CT/meter, installation | Optional cloud account | Integration/firmware support |
| Home Assistant host | Host, storage, power supply | Electricity, backups | Administration and network uptime |
| Contactor arrangement | Engineer design, components, testing | Inspection and eventual replacement | Safe manual operation, circuit rework |
| Monitoring plug | Properly rated device | Batteries rarely applicable; electricity | Failed sockets, heat and loss of control |
| UPS/network resilience | UPS, switch and battery | Battery replacement, standby power | Runtime testing |
For a site needing only a nonbinding warning, avoid buying control hardware until real measured history proves the need. A local meter and a notification often provide more defensible value than an elaborate shedding scheme that cannot be independently verified.
A Home Assistant Implementation That Starts in Monitor-Only Mode
The examples here are advisory configurations. They create virtual entities and change only a helper that reports whether reducing discretionary consumption might be useful. They deliberately do not switch physical appliances, alter charge limits or command fixed-circuit contactors. Before using them, create/identify the corresponding local grid-import power sensor; the examples assume it is called sensor.grid_import_power and reports watts with positive values for import. If your entity reports kW, uses negative import or can show unknown/unavailable, correct that at the data-mapping stage rather than copy the example blindly.
Example 1: Define an explicit manual enable and advice status
Create two helpers in configuration.yaml (or their equivalents under Settings → Devices & services → Helpers). The automation permission starts OFF and cannot turn itself ON after a restart in this example. The advice string remains separate from any physical switching device. After commissioning, an owner can enable monitor-only advice without authorising load control.
# configuration.yaml — monitor-only helpers
input_boolean:
load_shedding_advisory_enabled:
name: Load shedding advisory enabled
initial: false
input_select:
load_shedding_advice:
name: Load shedding advice
options:
- Monitor only
- Reduction suggested
- Normal demand
- Data unavailable
initial: Monitor only
A globally restored input boolean could resume in its previous state following a Home Assistant restart. Setting initial: false explicitly prevents that for this helper. That is a useful supervisory design choice, not a physical protection feature. Operators should understand that switching off the advisory helper does not disarm their installed charger or electrical protection system.
Example 2: Separate valid, sustained high demand from safe restoration
Use two template binary sensors. The first requires an import reading above the example advisory threshold for ninety seconds. The second requires readings below a lower recovery threshold for five minutes. Both become unavailable when the source is nonnumeric. The thresholds are illustrative and do not correspond to any legal current limit.
# configuration.yaml — advisory-only threshold indicators
template:
- binary_sensor:
- name: Grid Import High Advisory
unique_id: grid_import_high_advisory_076
availability: "{{ is_number(states('sensor.grid_import_power')) }}"
state: "{{ states('sensor.grid_import_power') | float(0) > 6000 }}"
delay_on: '00:01:30'
- name: Grid Import Recovery Advisory
unique_id: grid_import_recovery_advisory_076
availability: "{{ is_number(states('sensor.grid_import_power')) }}"
state: "{{ states('sensor.grid_import_power') | float(0) < 4500 }}"
delay_on: '00:05:00'
Important limitation: is_number() rejects text such as unavailable, but a sensor that continues displaying an old numeric reading has not necessarily become unavailable. Validate freshness separately using your integration’s availability mechanisms or reliable timestamp/heartbeat. MQTT has documented availability and expire_after facilities; retained payloads can replay stale readings. There is no generic freshness guarantee solely because a dashboard shows a plausible number. Our cloud-free metering guide discusses the wider telemetry architecture.
Example 3: Advisory state machine — no switch, relay or charger command
The next automation updates a virtual dropdown in response to the high and recovery indicators. It also revisits the current state on Home Assistant start; on startup it defaults to monitor-only until valid, sustained demand criteria have been satisfied. In this monitor-only example, Reduction suggested is a label for the user, not a demand for any physical control action.
# Add to automations.yaml when using a separate automations file
- id: energy076_load_advisory_state
alias: Energy 076 - monitor-only load advisory
mode: restart
triggers:
- trigger: state
entity_id:
- binary_sensor.grid_import_high_advisory
- binary_sensor.grid_import_recovery_advisory
- trigger: homeassistant
event: start
- trigger: time_pattern
minutes: '/5'
actions:
- choose:
- conditions:
- condition: state
entity_id: input_boolean.load_shedding_advisory_enabled
state: 'off'
sequence:
- action: input_select.select_option
target:
entity_id: input_select.load_shedding_advice
data:
option: Monitor only
- conditions:
- condition: template
value_template: >-
{{ not is_number(states('sensor.grid_import_power')) }}
sequence:
- action: input_select.select_option
target:
entity_id: input_select.load_shedding_advice
data:
option: Data unavailable
- conditions:
- condition: state
entity_id: binary_sensor.grid_import_high_advisory
state: 'on'
sequence:
- action: input_select.select_option
target:
entity_id: input_select.load_shedding_advice
data:
option: Reduction suggested
- conditions:
- condition: state
entity_id: binary_sensor.grid_import_recovery_advisory
state: 'on'
sequence:
- action: input_select.select_option
target:
entity_id: input_select.load_shedding_advice
data:
option: Normal demand
The above demonstration intentionally takes no action when both sensors are between thresholds. That preserves the current advisory state and implements hysteresis. There is an additional real-world requirement: the system must handle a numeric but stale meter correctly. Add a vendor-supported availability/heartbeat entity and classify missing freshness as Data unavailable before deriving any automated hardware action. The sample is suitable for observation and dashboard prototyping, not for controlling the service connection.
When Home Assistant starts, the helpers and automations initialise according to their own documented semantics; for-based dwell periods do not survive an automation restart or reload. Do not treat a restart as proof that a pre-restart ninety-second condition is still true. If you need guaranteed persistence for a non-safety scheduling requirement, document and test the appropriate helper/timer/timestamp design. For fast capacity protection, use independently engineered equipment instead.
Example 4: Visibility without actuation
If you want a compact dashboard indicator, the following template sensor copies the current advice into a read-only text entity. It adds convenience without increasing control authority.
# configuration.yaml — optional read-only dashboard sensor
template:
- sensor:
- name: Load Shedding Recommendation
unique_id: load_shedding_recommendation_076
state: "{{ states('input_select.load_shedding_advice') }}"
icon: mdi:flash-alert
Check the rendered states in Developer Tools, verify that unknown or unavailable meters produce a visible warning, and review automation traces. Never infer electrical safety from a green icon or a low average kW chart. Home Assistant’s automation and template documentation explains trigger semantics, state availability and restart behaviour; use the version matching your installed Home Assistant release.
What a production control integration would still require
To progress beyond advice, a separate design review must establish whether the manufacturer supports supervisory curtailment; which endpoint or input is approved; how it acknowledges requests; minimum/maximum permissible levels; local failure behaviour; user override; event logging; and a rollback plan. That design should be load-specific and tested at the device. A generic action such as switch.turn_off cannot encode those constraints by itself. For an EV charger, follow the charger vendor’s dynamic-management documentation and read our dynamic load balancing hardware guide.
Failure Modes: Test the Bad Days, Not Just the Sunny Ones
An automation tested only when every device is online has not been properly commissioned. The interesting cases occur when a meter stops updating during a high-load period, a cloud API rejects a command, Home Assistant restarts in the middle of an interval, or an appliance unexpectedly returns after a power cut. Test fault scenarios while all real switching actions are disabled. If you later commission a permitted equipment-specific interface, repeat the relevant tests under the manufacturer and installer’s approved process.
| Failure scenario | Misleading symptom | Safe supervisory behaviour | How to test without interrupting power |
|---|---|---|---|
| Internet unavailable | Vendor dashboard freezes | Local sensing and device-native limits remain in force | Disconnect WAN temporarily, observe states |
| Wi-Fi or Ethernet failure | Last numeric reading remains visible | Mark data invalid; stop making new discretionary requests | Disable test integration/device network in isolation |
| MQTT broker restarts | Retained value reappears as fresh | Require supported availability and freshness semantics | Restart test broker and inspect timestamps |
| Meter loses power | Import reading drops to zero or disappears | Never interpret it as new spare capacity | Simulate unavailable entity, do not work in live panel |
| Home Assistant restarts | Pending delays and for triggers disappear |
Start monitor-only and rebuild state from current readings | Restart HA with load control disabled |
| API rate limit or timeout | Command appears sent but has no effect | Retry in a bounded way, alert and stop assuming relief | Use intentionally unreachable test endpoint |
| Device reboot | Switch may restore a previous on/off state | Use device-appropriate restart policy | Test a spare, low-risk device under supervision |
| Native thermostat requests heat | Heat pump restarts unexpectedly after deferral | Native controls remain authority; do not block protections | Exercise simulator or vendor test mode |
| Manual override engaged | Automation immediately undoes user’s change | Suspend automation for that load and alert | Toggle helper, inspect automation traces |
| Solar battery begins charging | Grid import rapidly increases | Control decisions use measured grid/phase data | Replay historical trace or read-only simulation |
| Load doesn’t drop as predicted | Wrong entity mapping or command rejection | Do not count presumed shed watts; notify failure | Inject a simulated no-response state |
| Load returns automatically | Peak returns before next supervisory cycle | Native device must remain inside commissioned limits | Verify manufacturer restart behaviour |
Fail-safe, fail-open and fail-closed are context-specific
People sometimes insist that every relay should turn OFF on a communication failure. That is not a universally safe rule. Switching off an optional decorative load is different from removing power from a sump pump, refrigeration, ventilation or an appliance that needs a cooling cycle. Nor is an always-ON policy automatically safe for a load with a genuine operating constraint. The correct state is an engineering decision for each device, made before choosing a control interface, and backed by native or physical safety controls. For supervisory software, the prudent universal rule is narrower: do not make new high-consequence decisions when inputs cannot be trusted.
An owner also needs to know what a manual override changes. A manual override must not defeat protective equipment; it simply disables Home Assistant’s optional requests. Keep overrides discoverable, label them in plain language and avoid attaching dangerous powers to a decorative dashboard control. If an EV charger has a native pause or override function, check how it interacts with the car’s charging timer and the vendor’s own tariff schedule.
Recovery should be measured rather than assumed
After a loss of communication, the system should reacquire stable measurements, verify device state and allow the manufacturer’s controller to establish safe operating limits. A staged restoration process can then consider the next device’s availability and demand. Do not start five devices together simply because the last recorded state was on. In particular, an EV charging target restored by a vendor app may collide with a Home Assistant plan unless you define which system owns the schedule.
A good event log records (1) the measured constraint and whether the value was valid, (2) the exact requested advisory stage, (3) the available loads at that time, (4) operator overrides, (5) what the device actually acknowledged, and (6) the eventual measured response. It should be possible to reconstruct why a load was paused, when it resumed and why an expected saving did not occur.
Design for Batteries, Solar and Backup Supply
Solar generation and batteries complicate the difference between household power, grid import and inverter output. A grid meter might show only 0.5 kW imported while the home uses 5.5 kW, because solar supplies the rest. If clouds arrive, grid import can rise quickly; if the battery reaches reserve level, available discharge can fall. A home inverter may also have a distinct maximum AC output in backup mode and an altogether different grid-connected operating limit. Each limit belongs to the appropriate native equipment, not a third-party automation.
| Operating mode | Correct question | Common data trap | Proper control owner |
|---|---|---|---|
| Normal grid operation | What does grid import and per-phase current show? | Assuming solar production equals site spare capacity | Meter and commissioned supply management |
| Battery charging | How much import/export margin remains with charging active? | Counting battery charge as household appliance use | Inverter/BMS native rules |
| Battery discharge | Will discharge change when SOC/reserve is reached? | Assuming discharge stays constant all evening | Inverter/BMS |
| Backup/EPS mode | What is the islanded inverter’s continuous and surge rating? | Applying grid-connected thresholds to island mode | Certified transfer and inverter control |
| Generator operation | Can the generator accept instantaneous step loads? | Assuming total average kWh predicts stability | Generator/controller and designed priority loads |
| Solar export limiting | What equipment enforces contractual export limits? | Using an HA export sensor as enforcement | Approved inverter/export-limiting installation |
If a home has a grid-following PV inverter and no approved islanding system, an ordinary grid outage does not automatically provide usable solar power. Do not imply that Home Assistant can create backup capability by reconfiguring relays. Similarly, a battery’s state of charge says little about the safe instantaneous output during a transient; use the manufacturer’s specified operating mode and protection systems.
Owners considering advanced battery rules can refer to our home battery EPS and backup output guide and the battery ownership-cost analysis. Both topics are related to load management but have their own equipment and safety boundaries.
Choosing Which Load to Reduce: A Simple Decision Model
When several loads are simultaneously eligible, do not simply sort by nameplate watts. Consider reducible power now, how long the reduction can persist, probability that the command works, service deadline, restart consequences and operator preference. A device not currently running offers zero immediate demand reduction even if its rating plate says 3 kW. Likewise, a load that must restart within five minutes is a poor candidate for a one-hour high-demand window.
You can create a decision worksheet in a spreadsheet before writing any automation. Score load candidates on control reliability and acceptable delay, but keep absolute exclusions non-negotiable. Never let a numerical priority score override a safety exclusion, manufacturer instruction or user-declared critical requirement.
| Candidate | Reducible demand now (illustrative) | Delay tolerance | Native interface | Recommended first action |
|---|---|---|---|---|
| EV charging | 1–5 kW depending on current setting | Until charging deadline | Often available on compatible charger | Request supported lower current or pause |
| Planned dishwasher | 0 kW if not yet started | Before required completion | Sometimes scheduled start | Defer start; do not cut running cycle |
| Immersion hot-water demand | Up to rated active element | Conditional on stored heat and hygiene | Only with suitable approved controller | Defer discretionary programme within constraints |
| Heat pump | Highly variable | Tight in cold weather | Manufacturer dependent | Use only documented demand-response interface |
| Small noncritical charger | Tens to hundreds of watts | Usually high | Possibly rated smart plug | Optional low-impact candidate after assessment |
| Refrigerator | Uncertain compressor duty | Not an acceptable default candidate | Not appropriate | Never include in casual shedding |
The largest practical benefit often comes from a single controlled EV charging point, because it offers a substantial, variable load with a clear completion deadline. For homes without an EV, it may be better to schedule appliances than build an automated shedding installation. A rule that sends a message saying “wait fifteen minutes before starting the dishwasher” may outperform a complicated relay scheme in cost, safety and reliability.
How Much Money Can Load Shedding Actually Save?
The economics depend on the electricity contract. On a flat domestic energy tariff, shifting 1 kWh from 19:00 to 01:00 does not reduce its purchase price if the unit rate is unchanged. If a tariff offers lower overnight prices, the price difference may be material. If a home has solar PV, using solar internally has an opportunity cost equal to the export revenue forgone. And if a controlled load loses energy through cycling or storage, you need to include efficiency effects.
An illustrative EV charging shift of 12 kWh from a 32p/kWh period to a 12p/kWh period gives a gross tariff saving of £2.40, assuming exactly the same metered energy and no other changes. If the vehicle charges later with different charger/vehicle efficiency, use actual wallbox input kWh rather than a battery-side estimate. If charging from daytime solar instead, compare the avoided import against any foregone export rate. For a battery, include charging/discharging round-trip losses and cycle wear before treating arbitrage as profit.
| Calculation | Formula | Illustrative outcome | Main caveat |
|---|---|---|---|
| Tariff shift gross saving | Shifted kWh × rate difference | 12 × £0.20 = £2.40 | Only if energy and eligibility are unchanged |
| Shifted immersion heat | Element kW × delayed hours | 3 × 0.5 = 1.5 kWh | Heat may still be required later |
| Heat-pump preheat impact | Heat delivered / effective COP | 6 kWh heat / COP 3 = 2 kWh input | COP changes with outdoor and flow temperatures |
| Battery shifting cost | Imported kWh + conversion losses + degradation | Site-specific | Not all battery output replaces grid import |
| Standby cost | Always-on W × hours / 1000 × tariff | 5 W × 8760 / 1000 = 43.8 kWh/year | Add controller/network power correctly |
| Payback | Net installed cost / demonstrated annual benefit | Calculated after pilot | Don’t invent a guaranteed savings percentage |
A home may value comfort, supply-capacity management and resilience more than small bill reductions. Those benefits are real if measured or demonstrably delivered, but should be presented separately from guaranteed savings. If your tariff has no demand-based charge, do not attach a fictional financial value to every avoided peak. See our EV charger total cost guide for installation and subscription considerations.
Commissioning Checklist: From Read-Only to Carefully Governed Control
Use an evidence-driven rollout. The phases below describe how to assess a supervisory system, not how to modify household mains wiring. The physical design and any fixed switching must be handled by the appropriate professional.
- Inventory every relevant load. Record exact model, power, current type, restart characteristics, manufacturer guidance, and whether a supported pause or demand-response function exists. Include loads that are not connected to Home Assistant.
- Write the exclusion register. Identify critical medical, life-safety, refrigeration, drainage, security and heating functions that the automation may never interrupt. Verify physical circuit boundaries with the electrician if anything is uncertain.
- Identify the controlling constraint. Record whether it is contractual import, approved charger current limiting, inverter capacity, backup capability or an energy-cost target. Avoid applying the same threshold to different constraints.
- Commission measurement independently. Confirm physical CT orientation, phase assignment, calibration, unit and sign convention. Measure a known load step and compare cumulative energy over a representative period.
- Collect a baseline. At least two weeks of data is a reasonable start for ordinary demand-pattern observation; a longer period is needed to capture seasonal heating or solar variation. Record peak times and appliance overlap.
- Run the policy in monitor-only mode. Use the YAML demonstration or UI helpers to produce recommendations without sending a single actuator command. Compare each suggested action with what was actually happening.
- Test offline and stale-data behaviour. Simulate unknown sensors, invalid values, integration disconnects, HA restart and cloud disconnection. Ensure no message claims that protection was provided by the simulation.
- Document per-load conditions. Include maximum delay, permitted command interface, manual override, minimum off/run time, supported resumption and how to verify actual demand reduction.
- Use a qualified installer for fixed electrical control. Confirm equipment ratings, the suitability of control inputs, isolation, protection, inspection and testing. Keep the load’s native safety controls authoritative.
- Introduce one eligible load at a time. Observe acknowledgements, achieved demand changes, false triggers and occupants’ experience. Retain clear stop/rollback instructions.
- Verify restoration, not just curtailment. Test deadlines, staggered recovery, failed action notification, morning schedules and the interaction with vendor apps or charger timers.
- Audit after change. Firmware updates, new appliances, a changed EV, household occupancy or revised tariff can all invalidate assumptions. Maintain simple records of the safety exclusions and effective thresholds.
A useful pilot report separates technical reliability, operational usefulness and monetary benefit. Did the meter have fresh measurements 99.9% of the time? Did the suggested stage correspond to a real load that could be reduced? Did the owner have to intervene? Were service deadlines met? Did the grid import actually change? A small number of honest answers provides better guidance than a pretty dashboard showing a theoretical 3 kW “saved” whenever an appliance’s virtual switch was set to off.
Troubleshooting Load Management in Home Assistant
| Symptom | Plausible cause | First investigation | Unsafe shortcut to avoid |
|---|---|---|---|
| Advisory never triggers | Wrong units, missing update, trigger already true at startup | Inspect Developer Tools states, template and automation trace | Lowering a protective electrical rating in software |
| Advice changes repeatedly | No hysteresis, power fluctuations, sensor jitter | Plot import power and dwell duration | Triggering faster and faster |
| Reading stays at zero | Meter/integration offline or incorrect sign | Confirm actual meter and device availability | Assuming zero means spare capacity |
| Device does not reduce demand | Native controller rejected request, wrong target, load not active | Check response/acknowledgement and live watts | Repeated mains switching |
| Everything starts on reboot | Entity restore policies or vendor schedules | Review every actuator’s startup behaviour | Assuming HA restore state is a safety guarantee |
| Three-phase warning is wrong | Wrong CT phase/order, aggregate kW assumption | Compare phase currents with electrician’s documentation | Dividing total watts by one phase voltage |
| Heat-pump performance worsens | Curtailment conflicts with minimum runtime or weather compensation | Review heat-pump alarms and manufacturer manual | Bypassing thermal or compressor protections |
| Appliance programme fails | Unsuitable remote power interruption | Remove from candidate list; use manufacturer scheduling | Retrying power cuts to “unstick” cycle |
| After cloud outage, commands conflict | Multiple schedule owners | Identify one authoritative controller per load | Two automations fighting over one setting |
| Calculated bill savings too high | Shifted energy counted as avoided energy; export ignored | Compare actual import kWh and rates | Reporting theoretical watts as annual savings |
For tracing automation behaviour, use Settings → Automations & scenes → open automation → Traces, and compare the timestamp with the underlying meter series. A trigger can be perfectly valid while the wrong system model still produces an undesirable recommendation. Treat unexpected behaviour as a reason to disable discretionary control and investigate, not an invitation to bypass safeguards.
Frequently Asked Questions
1. Can Home Assistant stop my main fuse from blowing?
No. Home Assistant can monitor a grid meter and suggest or request reductions in suitable discretionary loads, but it is not an approved protective device and cannot guarantee a bounded response during faults or rapid demand changes. Appropriate installation protection and independently commissioned current-management equipment must provide the safety function. A dashboard alarm is supplementary information only.
2. Is load shedding the same as dynamic load balancing?
Not necessarily. Load shedding usually means removing or deferring selected demand; dynamic load balancing can continuously vary a compatible device’s current to accommodate other loads. An EV charger with an approved local CT/meter arrangement may provide native load balancing independently of Home Assistant. Prefer that function over building an improvised supply-limiting loop in a general automation platform.
3. Which household loads should always be excluded?
Start with medical and life-safety equipment, sump and sewage pumps, required alarms, accessibility support, essential refrigeration and heating functions whose interruption risks harm. The household may have further essential devices—such as an aquarium system or network infrastructure supporting alarms. An exclusion should be enforced by design, not merely by removing a card from the user interface.
4. Can I switch my immersion heater using a 16 A smart plug?
Do not assume so. Fixed immersion circuits and high-current loads need controls and protective equipment that are suitable for the specific installation and appliance. An advertised plug current rating alone does not establish safe continuous or inductive duty, nor does it authorise modifying fixed electrical wiring. Prefer an installer-designed, certified controller or an approved native immersion-diversion system that preserves thermal protection and hygiene programmes.
5. Can I cut power to a wall-mounted EV charger during peak demand?
Do not use repeated upstream mains disconnection as a substitute for the charger’s specified charging-current control. Select a model with an approved load-management system and compatible CT/meter, or use a manufacturer-supported supervisory pause/current setting where available. The charger must still enforce its independently commissioned circuit and supply limits when Home Assistant is offline.
6. What about washing machines and dishwashers?
A manufacturer-supported scheduled start or pause/resume feature can be useful. Abruptly removing power mid-cycle may leave water, interrupt drainage or fail to resume cleanly. Unless that exact appliance model documents safe remote interruption, defer the start rather than cutting the running machine’s supply. A smart plug’s ability to switch an appliance is not proof the resulting process is safe.
7. What is the best control interface for a heat pump?
Use the manufacturer’s documented demand-response or energy-management interface, if supported by the exact model and installed controller. Such interfaces may impose compressor, pump, flow and minimum-run constraints. Avoid using a generic relay to kill compressor power or imitating a safety thermostat. Heat-pump zoning and backup heating warrant an installer-reviewed design.
8. How much hysteresis is enough?
There is no universal threshold gap. It should exceed ordinary sensor noise and the expected recovery dynamics while leaving sufficient headroom for the actual non-safety scheduling goal. Begin by measuring several days of import fluctuations and desired load response. An example 6 kW enter, 4.5 kW exit and 90 s/5 min dwell illustrates the concept, not a recommended setting for a particular supply.
9. Will my Home Assistant automation run if the internet is disconnected?
Some automations and locally integrated devices continue over the LAN, depending on the hardware and data source. Others depend on cloud APIs, online schedules or authentication. The internet, LAN, broker, Home Assistant host and device are separate dependencies. Even a fully local design should not be entrusted with a protective electrical function that must remain available during a host or network failure.
10. What happens to for: triggers when Home Assistant restarts?
Home Assistant documentation states that pending trigger-duration waits do not survive restart or automation reload. A long-running action sequence can also be interrupted. Use startup reconciliation, well-understood helpers and direct device-native constraints for reliable supervisory scheduling. Do not claim that a timer or Home Assistant restart sequence replaces hardwired protection.
11. How do I identify stale meter readings?
Prefer native availability and explicit heartbeat or expiry features. For MQTT, Home Assistant documents expire_after and availability topics; retained payloads can recreate a plausible old reading. A number remaining on screen is not evidence of a live update. Compare the expected sampling cadence with actual receipt timestamps and design the supervisory system to stop issuing new load requests when data is suspect.
12. Can I use total three-phase kW to manage a per-phase current limit?
Not safely as a universal method. Three phases can be unevenly loaded, and reactive power or voltage variation complicates deriving amperes from a total power value. Use a proper phase-aware meter and the relevant manufacturer’s approved management function. An electrician should determine phase mapping and any applicable supply constraints. Total kW remains useful for energy-cost monitoring.
13. Does shedding a 3 kW load for one hour save 3 kWh?
It removes up to 3 kWh from that hour if the appliance would otherwise have run continuously at 3 kW. If the same work happens later, the energy was shifted, not eliminated. Efficiency, standby power and appliance behaviour can change the net result. Quantify savings with measured import kWh and the actual tariff, not by adding up nameplate watts of commands sent.
14. Is a relay’s reported OFF state proof that the load is disconnected?
No. An entity state usually represents what a controller believes or requested; it is not independent electrical verification. Mechanical contactors may weld, feedback may be absent, a different supply path may exist, or the integration may be stale. Where feedback matters, an installer should specify appropriate auxiliary contacts or measurements. Do not treat software status as a certified isolation indication.
15. Should every controllable load default to OFF on reboot?
No. That can itself be harmful for essential pumps, refrigeration, frost protection or equipment requiring a controlled shutdown. Startup behaviour must be chosen load by load within the manufacturer’s rules. A general supervisory automation can default to monitor-only and avoid issuing new commands until valid readings and manual permission are available; the physical plant’s safe state is a separate engineering question.
16. How should manual override work?
A user override should clearly suspend optional Home Assistant scheduling for the relevant device, show why a request was cancelled and never alter certified protective limits. The system should not immediately fight the user’s command on the next meter update. Include an expiry or review mechanism where appropriate, and test what happens after a restart. Anyone at home should be able to understand the override without reading YAML.
17. How do I know whether the automation is worthwhile?
First measure the actual number and duration of high-demand intervals. Then count legitimate curtailment opportunities, successful device responses and any missed departure or comfort deadlines. Finally compare metered kWh by tariff period and export revenue before and after the pilot. For many domestic households, the strongest benefit comes from native EV charger load management and simple schedules, not an elaborate multi-relay system.
18. What should I buy first?
Usually a correctly installed, locally readable whole-home/per-phase meter—if you do not already have one—and a reliable host or dashboard for collecting a baseline. If an EV charger is the main flexible load, prioritise its approved load-management accessory. Do not buy a basket of smart relays before documenting the actual problem, exclusion list and supported control interfaces. Obtain professional design and installation quotations for any fixed electrical changes.
Final Buying and Implementation Recommendation
Start small and keep authority clear. A house with an EV charger may require one suitable meter and a manufacturer-approved charger limiter; Home Assistant can then plan charging around energy costs and departure deadlines. A household without EV charging may achieve most of its useful shifting by coordinating scheduled hot-water heating and when to start appliances, without remotely interrupting any active mains circuit. For a battery-backed home, keep backup capacity and power limiting inside the certified inverter/transfer equipment and use HA to observe and request optional schedules only.
Spend money in this order: accurate metering → load audit and safety exclusions → manufacturer-supported controls → pilot measurement → additional hardware only where the evidence justifies it. If a system needs complicated exception logic simply to remain safe when Wi-Fi disappears, you have probably assigned the wrong responsibility to Home Assistant.
Official References and Further Reading
The following vendor and public documents are useful for checking current software behaviour and installation responsibilities. They supplement but do not replace the manual for your exact hardware revision or your local electrical installation requirements.
- Home Assistant: Automation triggers — state, template and
forbehaviour, including restarts. - Home Assistant: Automation modes — concurrent runs and interrupted actions.
- Home Assistant: Template integration — binary sensors and availability templates.
- Home Assistant: Input Boolean helper — startup defaults and restoration.
- Home Assistant: MQTT sensor —
expire_afterand retained values. - Home Assistant: MQTT integration — availability and birth/will messages.
- Home Assistant: Shelly — local integration behaviour and communication prerequisites.
- Home Assistant: Integrating your electricity grid — meters and electrical-cabinet safety caveats.
- GOV.UK: Approved Document P — electrical safety requirements for dwellings in England.
- myenergi — consult the current charger installation manual for grid CT and load management, matching the exact model and firmware.
Editorial note: examples are educational supervisory controls. They do not provide a certified protective function, authorise electrical installation work or guarantee savings.