CIM Issues #7040
Add attributes to time-series to record forecast uncertainty
IEC 61970-301
Description
A known omission in the CIM models is a standard way of describing the confidence levels of a forecast.
Commercial services provide APIs with various methods.
A simple extension to the CIM could follow this example:
https://docs.meteoblue.com/en/weather-apis/forecast-api/forecast-data
Predictability
A single weather forecasts model cannot be optimized for all weather conditions and areas. meteoblue operates a large number of weather models and collects data from multiple national weather services. Some models are more suitable for complex alpine terrain, while other models calculate fog conditions more precise. By combining multiple forecast models with statistics and machine learning algorithms, meteoblue calculates a learning multi-model forecasts (mLM). A byproduct of this approach is the ability to estimate the accuracy of the current forecast for each location.
If the majority of forecast models predicts the same weather conditions for a given location and achieve consensus, a high predictability is indicated. The predictability is given in percent, as well as a predictability_class which is just a simpler representation for the percentage value.
Predictability 0-100 %
Predictability class 0 - 5 1 = very low, 5 = very high
Existing CIM models
Core::BasicIntervalSchedule used for load forecasts, generation forecasts
Environmental::Forecast
Updated by Becky Iverson 3 days ago
11/18/2024: Discussed possible additions to the Forecast class in the Enviromental package. Considering additions for new attributes confidence (percentage) and/or confidenceRange (enum of high, medium and low). Plans to continue discussion in the next Environmental Data Team meeting on 11/25/2024.
Updated by Tom Berry 2 days ago
PLUS
If there are multiple forecasts for the same location & time intervals, then attributes are needed to distinguish which series is which.
Draft IEC 63402 for home/buildings EMS defines a forecast as an array of PowerForecastValue, where PowerForecastValue has multiple values per time slot.
Mandatory
value_expected The expected power value
Optional (1)
value_upper_limit The upper boundary of the range with 100 % certainty the power value is in it
value_lower_limit The lower boundary of the range with 100 % certainty the power value is in it
Optional (2)
value_upper_95PPR The upper boundary of the range with 95 % certainty the power value is in it
value_upper_68PPR The upper boundary of the range with 68 % certainty the power value is in it
value_lower_68PPR The lower boundary of the range with 68 % certainty the power value is in it
value_lower_95PPR The lower boundary of the range with 95 % certainty the power value is in it
If the values were represented in CIM as seven different time series then they could be distinguished by enumerated and/or numeric attributes:
100 = upper limit
95 = upper two sigma boundary
68 = upper one sigma boundary
50 = expected
32 = lower one sigma boundary
5 = lower two sigma boundary
0 = lower limit