ROOFTOP_PV_ACTUAL_RUN
Contains prediction runs for rooftop PV area estimated actuals, with a 5-minute and 30-minute resolution for different estimate types. This is the parent table to the child table
ROOFTOP_PV_ACTUAL_PRED, which contains the corresponding actual predictions.
Column Schema
AREAID
Area identifier, aligning with the load forecasting areas.
SQL Type: VARCHAR2(10)
Polars Type: pl.Categorical
This column is mandatory, and a primary key.
ESTIMATE_TYPE
Type of Rooftop PV estimate, for example, MEASURED, SATELLITE and so on.
SQL Type: VARCHAR2(20)
Polars Type: pl.Categorical
This column is mandatory, and a primary key.
INTERVAL_DURATION
Duration of each interval (in minutes) for this prediction, for example, 5 or 30.
SQL Type: NUMBER(3,0)
Polars Type: pl.UInt16
This column is mandatory, and a primary key.
OFFERDATETIME
Datetime when this prediction submission was loaded.
SQL Type: DATE
Polars Type: pl.Datetime()
This column is mandatory, and a primary key.
Example: 2026/01/20 00:05:00
PREDICTION_PRIORITY
Priority of prediction run, higher number is used in preference to lower number for the same provider.
SQL Type: NUMBER(10,0)
Polars Type: pl.UInt64
This column is mandatory, and a primary key.
PREDICTION_RUN_DATETIME
Datetime (interval ending) from which this prediction run is valid.
SQL Type: DATE
Polars Type: pl.Datetime()
This column is mandatory, and a primary key.
Example: 2026/01/20 00:05:00
PROVIDERID
Provider identifier, for example, AEMO, PROVIDER_A and so on.
SQL Type: VARCHAR2(20)
Polars Type: pl.Categorical
This column is mandatory, and a primary key.
INSTALLED_CAPACITY
Installed rooftop PV capacity used for the prediction run, in MW.
SQL Type: NUMBER(18,8)
Polars Type: pl.Float64
This column is not mandatory.
LASTCHANGED
Datetime when the prediction run was written into AEMO database.
For more information, see the dedicated documentation page.
SQL Type: DATE
Polars Type: pl.Datetime()
This column is not mandatory.
Example: 2026/01/20 00:05:00
MODEL_USED
Metadata describing the model used to produce the prediction run.
SQL Type: VARCHAR2(30)
Polars Type: str
This column is not mandatory.
PROVIDER_TIMESTAMP
Datetime when the provider created the forecast.
SQL Type: DATE
Polars Type: pl.Datetime()
This column is not mandatory.
Example: 2026/01/20 00:05:00
REMARKS
Comments relating to the prediction run.
SQL Type: VARCHAR2(300)
Polars Type: str
This column is not mandatory.
SUPPRESSED_PROVIDER
Flag indicating if the prediction run was suppressed by the provider when submitted. Suppressed predictions are not used by downstream forecasting systems. Suppressed = 1, Unsuppressed = 0.
SQL Type: NUMBER(1,0)
Polars Type: pl.Boolean
This column is not mandatory.
New Data Timing
Every 5 or 30 minutes when a new rooftop PV estimated actual is available, covering the most recent 5 or 30-minute interval that is available
Code Examples
Polars
Casting is enums, categoricals etc is not essential, but may speed up parsing.
Swapping int for pl.UInt16, pl.Int32 would also speed up parsing. However some of the datatypes are inferred, so they may be wrong.
This code deduplicates for you. Deduplication is normally only required across multiple months.
This code assumes that all columns are present, which may not be true for older files.
import polars as pl
import polars.selectors as cs
# e.g. "2029/03/15 00:05:00"
dt_s_fmt = "%Y/%m/%d %H:%M:%S"
csv_paths = ["SOMEFILE.CSV", "OTHERFILE.CSV"]
lf = (
pl.concat(
[
pl.scan_csv(
csv_path, # must be unzipped beforehand
skip_rows=1, # skip the header
infer_schema=False, # much faster
)
.filter(cs.by_index(0) == "D") # drop the footer
.select(~cs.by_index(range(4))) # drop metadata columns
for csv_path in csv_paths
],
how="diagonal_relaxed",
)
.with_columns(
pl.col(
"OFFERDATETIME",
"PREDICTION_RUN_DATETIME",
"LASTCHANGED",
"PROVIDER_TIMESTAMP",
).str.strptime(pl.Datetime, dt_s_fmt),
)
# Polars can't cast string straight to bool. Need to go via something else.
.cast({"SUPPRESSED_PROVIDER": pl.UInt8})
.cast(
{
"AREAID": pl.Categorical,
"ESTIMATE_TYPE": pl.Categorical,
"INTERVAL_DURATION": int, # or pl.UInt16
"PREDICTION_PRIORITY": int, # or pl.UInt64
"PROVIDERID": pl.Categorical,
"INSTALLED_CAPACITY": pl.Float64,
"MODEL_USED": str,
"REMARKS": str,
"SUPPRESSED_PROVIDER": pl.Boolean,
}
)
# Deduplicate
# We want one row per ['AREAID', 'ESTIMATE_TYPE', 'INTERVAL_DURATION', 'OFFERDATETIME', 'PREDICTION_PRIORITY', 'PREDICTION_RUN_DATETIME', 'PROVIDERID']
# Tie-breaker: greatest value of LASTCHANGED
.sort(LASTCHANGED, descending=False)
.unique(
subset=[
"AREAID",
"ESTIMATE_TYPE",
"INTERVAL_DURATION",
"OFFERDATETIME",
"PREDICTION_PRIORITY",
"PREDICTION_RUN_DATETIME",
"PROVIDERID",
],
keep="last",
)
)
Nemosis
This table is not yet available in Nemosis.
Availability
AEMO's documentation states that this table's visibility is "Public". However, it is not available in the public monthly MMSDM archive. AEMO probably call this public because it is available to all market participants over private FTP.
There is a chance that is is still public to non-participants. It may only be available for some months; or (rarely) it may be available in a non-standard location.
Deduplication and PDR Loader Configuration
The primary keys are:
The transaction type is "insert", which means that you will usually not see duplicate data. If you do, you can deduplicate based on the primary keys, without sorting first. Rows from different files with the same primary keys should have the same values.
| Report Name | Report Subtype | Version |
|---|---|---|
| DEMAND | ROOFTOP_PV_ACTUAL_RUN | 1 |
This means that the CSV files start with metadata columns:
Links
This page as JSON (schema documented here, but may change)
Download a JSON Schema specification for this table.
Official AEMO schema documentation
Other Information
Package: DEMAND_FORECASTS