The normalized version of the interval API is now enhanced to fill gaps in interval data with estimates using machine learning.

What is the feature?

This feature is the most significant upgrade yet to the normalized API of Plug Intervals. The API response now fills gaps in utility published interval data for a meter with estimated values using a machine learning model to ensure you get a continuous stream of interval data. This means you can rely on our data for more accurate analysis without having to worry about gaps in utility published data.

What problem does it solve?

Utility published data often has gaps. These gaps could be 15 minutes, 1 hour or even multiple days long. If not accounted for, such gaps distort calculations and lead to inaccurate analysis or solutions. By using a machine learning model to enhance the interval data we deliver with estimated data, we help you spend less time cleaning data to find and account for gaps and spend more time on your actual use-case. With a high estimation accuracy, our imputation engine delivers reliable data for you to build more accurate solutions.

Some key features of Interval Imputation

1. Ensures interval data stream for a meter is as complete as possible.
2. Includes inference logic to derive net data values when only import & export are published by the utility.
3. Plug dashboard is upgraded to power the intervals graph with normalized data instead of the raw utility published data.
4. CSV download for interval data now has additional columns to indicate whether a given interval record is Actual (as published by utility) or Estimated (imputed by Arcadia using a ML model)
5. Normalized API response contains additional fields under readings/imputedReadingsPositions to indicate which positions of the response data are imputed
6. Interval imputation is only performed on normalizedAPI response. No updates to the raw interval API. It will always deliver utility published data only.
7. Interval imputation will not be performed if a meter has less than 30 days of history data or gap is larger than 2 weeks to avoid inaccurate estimates.