A Home Assistant custom integration that provides days 2–7 electricity price and tariff forecasts from AEMO's PD7DAY pre-dispatch data for the National Electricity Market (NEM). Designed to complement Amber Electric's 24-hour Express forecast, this integration covers the window beyond Amber's reach.
AEMO publishes PD7DAY three times per day (07:30, 13:00, 18:00 AEST). This integration fetches those updates on the same schedule and applies a two-stage on-device calibration pipeline — isotonic regression for bias correction, followed by an OLS correction using AEMO STPASA supply/demand features — to produce calibrated estimates with P10/P50/P90 confidence bands.
The integration uses a two-stage forecasting pipeline:
- Stage 1 — PD7DAY isotonic calibration: AEMO's 7-day PD7DAY price forecasts are calibrated using isotonic regression fit on 60 days of rolling history. This corrects systematic bias and shapes the time-of-day profile across all horizons (h0–168).
- Stage 2 — STPASA OLS correction (horizons h22–h120): At medium-range horizons, an OLS model trained on AEMO STPASA supply/demand features (solar UIGF, wind UIGF, surplus capacity, demand 10/50/90) further corrects the isotonic output. Beyond h120, the model falls back to isotonic-only.
- MAE improvement at h24–168: −10.7% vs isotonic alone (14.65 vs 16.06 $/MWh)
- Day-rank Spearman ρ: 0.917 vs 0.850
- Solar UIGF is the dominant STPASA signal (ρ = −0.78)
- Battery dispatch optimisation — use the 7-day calibrated forecast and P10/P90 bands to schedule battery charge/discharge cycles beyond the Amber 24-hour window, minimising import cost and maximising export revenue at peak periods.
- EV charging scheduling — identify the cheapest 2-hour window over the next 7 days (
cheapest_2h_windowattribute) and trigger overnight EV charging at the lowest forecast price. - Hot water pre-heating — use the
min_24h_valueattribute to trigger resistive hot water heating during forecast low-price windows (solar sponge periods). - Grid stress awareness — the
grid_stressbinary sensor and grid notices count sensor provide advance warning of LOR/MSL events, allowing pre-emptive load shifting. - Linear programming dispatch — feed the full
forecastattribute list into an LP optimiser (e.g. EMHASS) as the price signal for multi-day horizon planning. - Tariff arbitrage — use the network-aware tariff sensors (Energex, Ausgrid, etc.) for accurate import/export tariff forecasting that accounts for time-of-use network charges.
- Days 2–7 price forecast — calibrated $/kWh with P10/P50/P90 confidence bands, trimmed to the window beyond Amber Express
- Isotonic calibration — monotone PAV regression bias correction fitted on actual TradingIS vs PD7DAY pairs, with per-bucket compression ratio, iso_mae, and P10/P90 confidence intervals
- Interconnector flows — interconnector MW flow forecasts for the configured region
- Market intervention flag — binary sensor from CASESOLUTION data
- Calibration diagnostic — observation count, active bucket count, fit quality per bucket
- Live charts — two camera entities updated after each refit:
- Price ToD Chart — actual price by time of day (mean, median, P10–P90 spread across all observed intervals)
- Calibration Chart — isotonic calibration goodness dashboard: compression ratio heatmap, iso_mae bars, PAV complexity scatter, and compression ratio drift time-series
- Cloud polling — two independent polling loops:
- PD7DAY fetches at AEMO publish times: 07:30, 13:00, 18:00 AEST (3 requests/day)
- TradingIS fetches actual 5-min dispatch prices every 30 minutes (48 requests/day)
- 5-minute dispatch prices — boundary-aligned
DispatchCoordinatorpolls NEMWEB TradingIS at:01:15,:06:15, ...,:56:15(75 s after each dispatch boundary, after NEMWEB publishes). Used as the livenative_valuefor tariff and spot sensors between 30-minute PD7DAY intervals. - Live sensor state — all forecast sensor states advance automatically every 30 minutes to reflect the current interval, with no fetch required
- No third-party accounts required — actual prices sourced directly from AEMO TradingIS
- Dependencies —
matplotlib,numpyfor chart rendering,astralfor solar elevation (installed automatically by HACS/HA)
- Home Assistant 2024.1 or later
- Network access to
www.nemweb.com.au
- Open HACS in your HA instance
- Go to Integrations → Custom repositories
- Add
https://github.com/purcell-lab/nem_pd7daywith category Integration - Search for NEM PD7DAY and install
- Restart Home Assistant
- Go to Settings → Devices & Services → Add Integration and search for NEM PD7DAY
- Download the latest release zip from the Releases page
- Extract
custom_components/nem_pd7day/into your HA config directory - Restart Home Assistant
- Go to Settings → Devices & Services → Add Integration and search for NEM PD7DAY
- Go to Settings → Devices & Services
- Find NEM PD7DAY and click ⋮ → Delete
- Confirm deletion
This removes the integration and all its entities. Calibration storage files are not automatically deleted. To fully clean up, remove the following files from your HA config .storage/ directory:
rm /config/.storage/nem_pd7day.{region}.observation_log
rm /config/.storage/nem_pd7day.{region}.calibration_coefficients
rm /config/.storage/nem_pd7day.{region}.forecast_history
rm /config/.storage/nem_pd7day.{region}.stpasaReplace {region} with the lowercase region code (e.g. qld1, nsw1).
The integration is configured via a single-step UI flow.
| Field | Default | Description |
|---|---|---|
| NEM Region | QLD1 | The NEM region to monitor: QLD1, NSW1, VIC1, SA1, or TAS1 |
The configured region is used for both price forecasting and calibration. No additional settings are required.
No configuration.yaml entries are required.
Each integration instance monitors one NEM region with full independent calibration. To monitor multiple regions, add a separate integration instance for each via Settings → Integrations → Add Integration → NEM PD7DAY. Each instance maintains its own calibration store, observation log, and forecast sensors.
The forecast and tariff sensors carry large attribute payloads (7 days × 48 intervals each). Home Assistant will log warnings like:
State attributes for sensor.nem_pd7day_*_tariff exceed maximum size of 16384 bytes.
Attributes will not be stored
These warnings are expected and harmless. The sensor state value (the current $/kWh price) is always recorded and available in history — only the large forecast attribute list is dropped by the recorder. No configuration changes are needed.
Do not add
recorder: exclude: entity_globsfor nem_pd7day sensors — doing so prevents the sensor state from being recorded, which breaks the history graph in the HA UI.
The integration runs two independent polling loops.
AEMO publishes 7-day ahead price forecasts three times per day. The integration fetches on the same schedule using async_track_point_in_utc_time:
| NEM time (AEST) | UTC |
|---|---|
| 07:30 | 21:30 (previous day) |
| 13:00 | 03:00 |
| 18:00 | 08:00 |
Source: AEMO NEMWeb PD7DAY
The Stage 2 OLS correction is driven by AEMO's STPASA (Short-Term Projected Assessment of System Adequacy) feed, which provides the supply/demand outlook across all NEM regions in a single ZIP: solar UIGF, wind UIGF, surplus capacity, and demand 10/50/90 percentiles.
Source: AEMO NEMWeb STPASA
Actual NEM dispatch prices are fetched from AEMO's TradingIS reports and used to build calibration observations:
-
URL:
https://www.nemweb.com.au/REPORTS/CURRENT/TradingIS_Reports/ - Schedule: at HH:02 and HH:32 NEM time — 2 minutes after each 30-minute trading interval closes
-
Method: downloads the 6 × 5-minute dispatch files for the closed interval, averages the RRP values (
$/MWh → $ /kWh) -
Observation tagging: each calibration observation records
actual_source: "tradingis"
5-minute dispatch prices for the sensor native_value are fetched from AEMO TradingIS on a boundary-aligned schedule:
- URL:
https://www.nemweb.com.au/Reports/Current/TradingIS_Reports/ - Schedule: 75 seconds after each 5-minute UTC boundary (
:01:15,:06:15, ...,:56:15). AEMO typically publishes 65–90 s after the interval boundary; 75 s sits within that window with margin (_DISPATCH_POLL_DELAY_S = 75). - Alignment: uses
async_track_point_in_utc_timewith self-rescheduling one-shot callbacks — no drift from HA startup time.
Forecast sensor states — price forecast and interconnector flow — advance automatically at each 30-minute interval boundary (:00 and :30 past every hour). The state always reflects the current interval from the most recent fetch, without waiting for the next PD7DAY fetch. This means:
- After the 07:30 fetch, the price forecast sensor state will step through each 30-minute interval for the rest of the day
- After HA restart, the state reflects the current interval immediately on load
- Between fetches, the raw forecast data is unchanged — only the active interval advances
All sensors are grouped under a single HA device named NEM PD7DAY {region} (e.g. NEM PD7DAY QLD1).
sensor.nem_pd7day_{region}_price_forecast — the primary calibrated spot price forecast sensor (days 2–7 only, trimmed to exclude the Amber Express window).
| Attribute | Description |
|---|---|
state |
Calibrated price for the current interval ($/kWh) |
region |
NEM region code |
forecast_generated_at |
ISO-8601 timestamp of the AEMO source file |
forecast |
List of all forecast periods (see below) |
next_value |
Calibrated price for the next interval |
min_24h_value |
Minimum calibrated price in the next 24 hours |
max_24h_value |
Maximum calibrated price in the next 24 hours |
cheapest_2h_window |
Best contiguous 2-hour window over 7 days |
Each entry in forecast contains:
nemtime: "2026-04-15T17:30:00+10:00" # interval END (AEMO convention)
time: "2026-04-15T17:00:00+10:00" # interval START
raw_value: 0.084 # raw AEMO forecast ($/kWh)
calibrated: 0.142 # isotonic-calibrated value
p10: 0.091 # 10th percentile (optimistic)
p50: 0.138 # 50th percentile (median)
p90: 0.231 # 90th percentile (conservative)
ols_mae: 0.038 # mean absolute error of calibration fit
calibrated_source: isotonic+stpasa # "isotonic+stpasa" (h22–h120), "isotonic", "passthrough_high", "passthrough_sanity", or "passthrough"
n_obs: 147 # observations used for this bucket
horizon_hours: 36.5 # hours ahead
value: 0.142 # alias for calibrated (template compat)Timestamp convention:
nemtimeis the interval END timestamp as published by AEMO.timeis the interval START (nemtime − 30 minutes). This matches the AEMO dispatch interval convention.
sensor.pd7day_{region}_ic_{interconnector} — one sensor per interconnector touching the configured region.
| Attribute | Description |
|---|---|
state |
Current period MW flow (positive = export) |
interconnector_id |
Interconnector identifier |
forecast |
List of forecast periods with nemtime, time, mw |
Default interconnectors per region:
| Region | Interconnectors |
|---|---|
| QLD1 | NSW1-QLD1, N-Q-MNSP1 |
| NSW1 | NSW1-QLD1, VIC1-NSW1, N-Q-MNSP1 |
| VIC1 | VIC1-NSW1, SA1-VIC1, V-S-MNSP1, T-V-MNSP1 |
| SA1 | V-SA, V-S-MNSP1 |
| TAS1 | T-V-MNSP1 |
binary_sensor.nem_pd7day_{region}_intervention — ON when AEMO has flagged a market intervention in the CASESOLUTION data. Under normal market conditions this is OFF.
sensor.nem_pd7day_{region}_calibration — calibration system diagnostic. This sensor is in the Diagnostic category and is hidden from the default HA dashboard (visible under the device's Diagnostic section).
| Attribute | Description |
|---|---|
state |
Total observations logged |
active_buckets |
Buckets with ≥ 20 observations (max 24) |
total_buckets |
24 (6 horizons × 4 time-of-day labels) |
fitted_at |
ISO-8601 timestamp of last model refit |
summary |
Per-bucket isotonic diagnostics: n, iso_n_steps, compression_ratio, iso_mae, x_min, x_max, q10_a, q90_a |
| Attribute | Description |
|---|---|
forecast_history_entries |
Total forecast entries stored across all tracked intervals |
forecast_history_intervals |
Number of unique interval keys in storage |
forecast_history_oldest |
ISO-8601 timestamp of the oldest tracked interval |
forecast_history_newest |
ISO-8601 timestamp of the most recent tracked interval |
forecast_history_runs_avg |
Average number of forecast runs per interval |
Use these to verify h48_96/h96plus buckets are accumulating — forecast_history_entries should grow by ~3 per interval per day (one entry per AEMO fetch).
sensor.nem_pd7day_{region}_price_tod_stats (Diagnostic) — current 30-minute slot's mean actual price ($/kWh), with full 48-slot statistics as attributes.
| Attribute | Description |
|---|---|
state |
Mean actual $/kWh for the current time-of-day slot |
unique_intervals |
Total unique intervals with actuals |
date_from / date_to |
Date range of recorded actuals |
slots |
List of 48 dicts — one per 30-min slot, each with mean_kwh, median_kwh, p10_kwh, p25_kwh, p75_kwh, p90_kwh, n |
sensor.nem_pd7day_{region}_source_file_datetime— timestamp of the latest AEMO PD7DAY source filesensor.nem_pd7day_{region}_data_updated— timestamp of the last coordinator data refresh
Both are diagnostic sensors (EntityCategory.DIAGNOSTIC) and do not appear on the default dashboard.
Two diagnostic sensors expose the full underlying forecast payloads as unrecorded HA attributes. Both are in the Diagnostic category and their large attribute lists are not saved to the HA recorder/database.
| Sensor | State | Attribute | Description |
|---|---|---|---|
sensor.nem_pd7day_{region}_pd7day_data |
Forecast generation time | forecast |
Full calibrated 7-day forecast list (330 intervals). Not saved to HA recorder. |
sensor.nem_pd7day_{region}_stpasa_data |
STPASA run time | intervals |
Full STPASA supply/demand interval list (288 intervals). Not saved to HA recorder. |
One sensor per (distributor, tariff_code) for the configured region. Tariff sensors cover the same days 2–7 window as the spot price sensor — the near-term Amber Express window is trimmed from the forecast attribute. The native_value (current interval tariff) is unfiltered and always returns the current price.
Supported distributors per region:
| Region | Distributors |
|---|---|
| QLD1 | Energex, Ergon |
| NSW1 | Ausgrid, Endeavour, Essential, EvoEnergy (serves the ACT) |
| VIC1 | Jemena, Powercor, United, AusNet, Victoria |
| SA1 | SAPN |
| TAS1 | TasNetworks |
EvoEnergy (ACT) — available tariff codes: 015, 016, 017, 018, 026, 090. The default enabled tariff is 026 (Battery Feed-in Trial), which is the export tariff.
The native_value is driven by the boundary-aligned DispatchCoordinator (5-minute live dispatch price → spot_to_tariff()), falling back to the current PD7DAY forecast interval if dispatch data is unavailable.
The spot field in each forecast entry reflects the calibrated spot price (same isotonic correction applied by the spot price sensor), not the raw PD7DAY value. This ensures the tariff forecast is consistent with the spot price forecast at spike intervals where spike_credible: false.
sensor.nem_pd7day_{region}_{network}_{tariff_code}_tariff
| Attribute | Description |
|---|---|
state |
Current interval tariff price ($/kWh), incl. additional usage fee + 10% GST |
forecast |
Days 2–7 tariff forecast list (see entry structure below) |
tariff_code |
Tariff code (e.g. 6900) |
distributor |
Distribution network name |
tariff_periods |
Time-of-use period structure with rates |
daily_supply_charge_$ |
Daily supply charge ($/day) |
Each entry in the forecast list contains:
| Field | Description |
|---|---|
time |
Interval START timestamp (nemtime − 30 minutes) |
nemtime |
Interval END timestamp (AEMO convention) |
spot_raw |
Uncalibrated spot price $/kWh (before bias correction) |
spot |
Calibrated spot price $/kWh |
value |
Final tariff |
period |
Tariff period name for that interval (e.g. peak, shoulder, off-peak) |
network_rate |
Network component $/kWh for that interval |
{
"time": "2026-06-03T10:00",
"nemtime": "2026-06-03T10:30",
"spot_raw": 0.082341,
"spot": 0.079812,
"value": 0.142500,
"period": "peak",
"network_rate": 0.08210
}sensor.nem_pd7day_{region}_{network}_{import_code}_export_tariff
Export tariff sensors are registered for battery-eligible network tariffs where an export program exists. The state is the feed-in tariff ($/kWh) for the current interval, computed from the live dispatch price via spot_to_feed_in_tariff(). The forecast attribute gives the days 2–7 export tariff using the calibrated spot price.
Export tariff formula: result_c_kwh / 100 (no additional usage fee, no GST — export tariffs do not include these charges).
| Network | Import tariff | Export tariff | ToD structure |
|---|---|---|---|
| Ausgrid (NSW1) | EA025 | EA029 | +3.85 c/kWh peak, −1.23 c/kWh solar sponge |
| Endeavour (NSW1) | N71 | N61 | +12.43 c/kWh peak, −1.97 c/kWh solar sponge |
| Essential (NSW1) | BLNT3AL | BLNREX2 | +11.57 c/kWh peak, −0.82 c/kWh solar sponge |
| EvoEnergy (NSW1) | 026 | 026 | Battery Feed-in Trial |
| SAPN (SA1) | RESELE | RESELE | +12.25 c/kWh peak, −1.00 c/kWh solar sponge |
number.nem_pd7day_{region}_additional_usage_fee — a configurable number entity (default 0.0293 $/kWh) added to import tariff calculations. Does not apply to export tariffs.
Three camera entities are registered on the device and can be added to any HA dashboard using a Picture or Camera card.
| Entity | Description |
|---|---|
camera.nem_pd7day_{region}_forecast_chart |
7-Day Pre-Dispatch Spot Price Forecast — raw vs calibrated with P10/P90 confidence band, per-day min/max labels, spike callouts, LOR/MSL notice bands |
camera.nem_pd7day_{region}_price_tod_chart |
Actual price by time of day — mean, median, and P10–P90 spread across all observed intervals |
camera.nem_pd7day_{region}_calibration_chart |
Isotonic calibration goodness dashboard: compression ratio heatmap, iso_mae bars, PAV complexity scatter, and compression ratio drift time-series |
All charts are re-rendered after each calibration refit (07:30, 13:00, 18:00 NEM). The calibration chart reads live isotonic diagnostics so the heatmap values, n counts and confidence indicators update as calibration matures.
The forecast chart shows the full 7-day ahead price window for the configured NEM region (note: the chart displays the full window for visual context, while the sensor attributes are trimmed to days 2–7):
- Calibrated line (blue solid) — isotonic-calibrated price forecast with P10/P90 confidence band
- PD7DAY Raw (grey dashed) — AEMO's raw pre-dispatch forecast before calibration
- Daily max/min labels — peak and trough $/kWh values annotated per day
-
AEMO Spike Forecast (red triangle callouts) — intervals where
raw_value ≥ $3/kWhandspike_credible: True; one callout per contiguous cluster pointing at the cluster peak, with the peak price labelled - Clip line (red dotted) — dynamic display ceiling at p99 + 15% headroom
- LOR/MSL notice bands — shaded vertical regions for active NEMWEB reserve (LOR1/2/3) and minimum load (MSL1/2/3) notices, with staggered labels and a dynamic legend showing only notice types present in the window
-
Dual y-axis —
$/kWh (left) and $ /MWh (right)
The value key in each forecast period is an alias for calibrated (or raw_value in passthrough), maintained for backward compatibility.
template:
- sensor:
- name: "PD7DAY Min Price 24h"
unit_of_measurement: "$/kWh"
state: >
{{ state_attr('sensor.nem_pd7day_qld1_price_forecast', 'min_24h_value') | round(4) }}template:
- sensor:
- name: "PD7DAY Cheapest Window Start"
state: >
{{ state_attr('sensor.nem_pd7day_qld1_price_forecast', 'cheapest_2h_window')['time'] }}template:
- sensor:
- name: "QLD1 PD7DAY Buy Cost"
unit_of_measurement: "$/kWh"
state: >
{% set forecast = state_attr('sensor.nem_pd7day_qld1_price_forecast', 'forecast') %}
{% if forecast %}
{{ forecast[0]['calibrated'] | round(4) }}
{% else %}
unavailable
{% endif %}All timestamps in this integration use AEST (UTC+10:00) with no daylight saving adjustment, matching AEMO's published data. Timestamps are always ISO-8601 with explicit +10:00 suffix, e.g. 2026-04-14T07:30:00+10:00.
The nemtime field in forecast periods is the interval end timestamp (AEMO convention). The time field is the interval start (nemtime − 30 minutes).
- Days 2–7 only — the near-term window (next 24 hours) is intentionally excluded to avoid duplication with Amber Electric's Express forecast. If you need the full 7-day window including day 1, the raw
pd7day_datadiagnostic sensor exposes all 330 intervals. - Calibration warm-up — the two-stage pipeline requires historical data to activate. Isotonic calibration starts at ~5–7 days; STPASA OLS correction (h22–h120) activates after the first nightly refit once STPASA feature data has been joined to observations. Expect isotonic-only output for the first 24 hours after install.
- STPASA OLS limited to h22–h120 — beyond h120, STPASA supply/demand signals degrade and the model falls back to isotonic calibration only.
- Spike forecasts are capped — raw PD7DAY values ≥ $3/kWh are treated as bid-based pre-dispatch placeholders and are capped at the isotonic model maximum. Genuine scarcity spikes are flagged via
spike_crediblebut are not passed through at face value. - Python 3.12+ required — the integration uses
asyncio.run()and other Python 3.12 patterns. - matplotlib build dependency — on some HA OS versions (notably Python 3.14 environments), matplotlib may fail to build from source. If this occurs, the chart camera entities will be unavailable but all price forecast and tariff sensors continue to function normally.
Check the HA log for errors from custom_components.nem_pd7day. The most common cause is a network issue reaching nemweb.com.au.
The first fetch runs at integration load. Check Settings → System → Logs filtered to nem_pd7day for fetch errors.
Normal for the first 5–7 days. Calibration requires at least 20 observations per bucket. Check the Calibration sensor state for current observation count.
These buckets require observations from intervals 2–4 days ahead. They begin accumulating approximately 48 hours after first install (or after upgrading from a version prior to v1.9.1 which fixed forecast history persistence).
These are expected and harmless — see the Recorder warnings section above. No configuration changes are needed. Do not add recorder: exclude: entity_globs as this breaks sensor history.
This occurs when HA's entity registry has a stale entity_id from a previous version registration. SA1 installs are unaffected as they were registered after the fix. To resolve:
- Go to Settings → Devices & Services → NEM PD7DAY [region] device
- Delete or disable/re-enable each affected entity
- Restart HA — entities re-register with correct IDs
Note: historical data for the affected entity will restart from the re-registration date. This is a one-time migration step.
| Version | Changes |
|---|---|
| 3.0.0 | Two-stage forecasting pipeline: Stage 1 isotonic calibration + Stage 2 STPASA OLS correction (h22–h120). New STPASA client fetches all-region STPASA ZIP once per cycle and distributes to per-region stores. fit_ols_stage2() trains a 9-feature OLS model per bucket using STPASA supply/demand features (solar UIGF, wind UIGF, surplus capacity, demand 10/50/90). calibrated_source attribute is isotonic+stpasa at medium-range horizons, falls back to isotonic beyond h120. New diagnostic sensors: <region>_pd7day_data and <region>_stpasa_data expose full forecast and STPASA payloads as unrecorded HA attributes. MAE improvement −10.7% at h24–168 vs isotonic alone. |
| 2.3.57 | Intermediate release (superseded by v3.0.0) |
| 2.3.56 | Refactor: centralise STPASA download — single ZIP fetch distributes to all region stores; eliminates duplicate per-region downloads |
| 2.3.55 | Initial STPASA OLS stage2 implementation |
| 2.3.47 | EvoEnergy tariffs for NSW1/ACT (codes 015, 016, 017, 018, 026, 090; default export tariff 026 Battery Feed-in Trial); forecast entries now include spot_raw, period, network_rate; freshness gate accepts settlement >= boundary; dispatch poll delay set to 75 s (AEMO publishes 65–90 s after boundary); clarified 16kB recorder warnings are harmless |
| 2.3.40 | Fix: DispatchCoordinator now boundary-aligned — polls NEMWEB TradingIS at 35 s past each 5-minute UTC boundary (NEMWEB publishes ~30 s after boundary). Replaces rolling update_interval that drifted up to ~4 min from startup offset, causing tariff sensors to show stale dispatch prices each interval |
| 2.3.39 | Fix: tariff sensors (import + export) were passing raw PD7DAY RRP into spot_to_tariff() / spot_to_feed_in_tariff() — store now wired into all tariff sensor constructors; _calibrated_value() applies isotonic calibration identical to spot sensor; spot attribute in forecast lists now reflects calibrated $/kWh |
| 2.3.38 | Fix: iso_chart.py missing matplotlib graceful fallback — adds same try/except ImportError + _placeholder_png() pattern as forecast_chart.py
|
| 2.3.37 | Fix: export tariff formula was including additional usage fee + 10% GST (overcounting ~4–5 c/kWh); export tariff now returns result_c_kwh / 100 only |
| 2.3.36 | Export tariff sensors: Ausgrid EA025→EA029, Endeavour N71→N61, Essential BLNT3AL→BLNREX2, SAPN RESELE→RESELE; new SAPN RESELE import program for SA1; Day 2–7 sensors moved to EntityCategory.DIAGNOSTIC; _suppress_stdout() wraps all aemo_to_tariff calls to silence sapower.py debug prints |
| 2.3.14 | Fix notice pipeline: NEMWEB directory deduplication; first-run bootstrap now backfills last 7 days instead of skipping all files; upgrade-path auto-reset in coordinator resets stuck cursor when last_seen > 0 but total_notices == 0 |
| 2.3.13 | Fix AEMO market notice period regex: date follows each time token (From HHMM hrs DD/MM/YYYY to HHMM hrs DD/MM/YYYY); multi-period notice support; all periods merged to widest window |
| 2.3.12 | Chart improvements: per-cluster spike callouts, anti-crossover arrow placement, chart title updated to 7-Day Pre-Dispatch Spot Price Forecast, x-axis ticks at 06:00/18:00 only, MSL/LOR notice staggered labels and legend patches |
| 2.3.x | Multiple fixes: matplotlib thread-safety (OO API), Grid Notices sensor device attachment, p10/p90 None crash, blocking event loop calls, sanity check log spam |
| 2.3.0 | MSL/LOR grid stress notices: NEMWEB poller, HA .storage persistence, 7-day chart annotation bands (LOR1/2/3, MSL1/2/3), Grid Stress binary sensor, Grid Notices count sensor |
| 2.2.11 | Sanity guard tuning: SANITY_RATIO_RAW_FLOOR=0.05 |
| 2.2.10 | Isotonic calibration: replaced weighted OLS with PAV monotone isotonic regression; per-bucket compression_ratio, iso_mae, iso_n_steps diagnostics; iso_chart camera; unconditional startup refit |
| 2.2.4 | Critical fix: spike actual prices (actual_rrp >= $3.00/kWh) were not being excluded from OLS training buckets — only the forecast passthrough path applied SPIKE_THRESHOLD. The 05 May spike event ($8,000–$15,000/MWh actuals) collapsed all medium/long-horizon peak and shoulder slopes to near zero. Fix is retrospective — call nem_pd7day.force_refit after updating to immediately restore correct coefficients. |
| 2.2.3 | Fix: adding a second NEM region failed with "This region is already configured" — config entry unique_id is now region-scoped (nem_pd7day_{region}) instead of hardcoded |
| 2.2.2 | New 7-day forecast chart camera entity per region — shows raw vs calibrated forecast with p10/p90 confidence band, per-day min/max labels (__init__.py (now_nem/to_nem_iso removed from calibration_engine imports); fix CI workflow — astral, numpy, matplotlib now installed in test runner; fix spike forecast callout label in |
| 2.2.1 | Fix total_buckets sensor attribute (was hardcoded 18, now derived dynamically as len(TOD_LABELS) × len(HORIZON_BANDS) = 24); add nem_pd7day.force_refit service — triggers immediate calibration refit and coordinator refresh for one or all regions without waiting for the next scheduled fetch; fix GitHub Actions CI workflow — install astral, numpy, matplotlib in test runner |
| 2.2.0 | Adaptive Calibration — weighted OLS with exponential time decay (λ=0.033, half-life≈21d); solar elevation angle ToD classification via astral (peak/solar/morning_ramp/shoulder); 4 labels × 6 horizon bands = 24 active buckets; astral>=2.2 dependency |
| 2.1.3 | Version bump |
| 2.1.2 | Multi-region audit: remove hardcoded entity_id from Gas sensor; drop Gas Generation Forecast sensor (not relevant to regional price forecasting) |
| 2.1.1 | Multi-region fix: add name= to camera and ToD Stats sensor device_info so HA generates correctly scoped entity IDs per region |
| 2.1.0 | Multi-region support: region-scoped unique_id and device_info for all entities; real ToD curves in bias chart (actual / AEMO raw / calibrated); calibration_result passed into tod_stats.compute()
|
| 2.0.9 | Camera entities: Price ToD Chart (actual price by time of day) and Bias Chart (live duck curve + OLS heatmap); Price ToD Stats sensor; matplotlib + numpy dependencies |
| 2.0.8 | Fix 30-min sensor state advance (async lambda bug); spike passthrough at $3/kWh; 90-day rolling window; negative passthrough threshold -$0.10/kWh (solar trough correction) |
| 2.0.7 | Calibration: passthrough when raw <= 0 (negative price regime); clamp quantile slopes to 0 |
| 2.0.6 | Fix missing await on async_added_to_hass — sensors now register correctly on HA restart |
| 2.0.5 | Live sensor state advance (30-min intervals); gas_tj fix; calibration sensor → diagnostic; forecast history merged into calibration sensor |
| 2.0.4 | Region-scoped storage keys (multi-instance safe, auto-migration); gas_tj populated in forecast history |
| 2.0.3 | Single-region enforcement — one region per integration instance, full calibration coverage |
| 2.0.2 | Clean sensor names, remove Amber dependency, add Forecast History sensor |
| 2.0.1 | Version bump post-integration |
| 2.0.0 | Multi-region sensor support, TradingIS actual prices (no Amber required), forecast persistence |
| 1.9.1 | Forecast history persistence fix — h48_96/h96plus calibration buckets now survive HA restarts |
| 1.9.0 | Comprehensive pre-deployment test suite (107 tests) |
| 1.5.0 | AEMO interval convention: nemtime (end) + time (start) on all forecast periods |
| 1.4.0 | Timezone overhaul: all timestamps ISO-8601 +10:00, UTC-safe scheduling |
| 1.3.0 | Replaced polling with async_track_point_in_utc_time at AEMO publish times |
| 1.2.0 | OLS + IRLS quantile calibration engine, calibration diagnostic sensor |
| 1.1.0 | CASESOLUTION (binary sensor), MARKET_SUMMARY (gas), INTERCONNECTORSOLUTION sensors |
| 1.0.0 | Initial release: PRICESOLUTION forecast sensor |
MIT License — see LICENSE for details.