OEDI SI/Scenarios/Point On Wave Data Imputation

Point On Wave Data Imputation​ Summary

• Date Created: 2024/09/28
• Organization: ORNL
• Objective: Accurate PoW data is essential for identifying faults in electrical systems. Missing or corrupted data can lead to incorrect diagnostics, potentially resulting in unaddressed issues that could escalate into larger failures. To address the issue of missing points in Point On Wave (PoW) data due to data corruption, the goal is to employ data imputation as a necessary preprocessing step to accurately fill in these missing data points in the PoW data stream.
• Use Case: DER Aggregation Algorithm
• Methodology
• Inputs
• Outputs
• Configuration
• Webinars
OEDI SI online POW Data Imputation.pdf

Docker Container

Download the docker container at:

Run Locally

http://localhost:8080/edit_scenario?Point On Wave Data Imputation

    Run Locally

Input Data

Download input data at:

Output Data

Download output data at:

References

Back to DER Aggregation Algorithm

Property "Objective" (as page type) with input value "Accurate PoW data is essential for identifying faults in electrical systems. Missing or corrupted data can lead to incorrect diagnostics, potentially resulting in unaddressed issues that could escalate into larger failures. To address the issue of missing points in Point On Wave (PoW) data due to data corruption, the goal is to employ data imputation as a necessary preprocessing step to accurately fill in these missing data points in the PoW data stream." contains invalid characters or is incomplete and therefore can cause unexpected results during a query or annotation process.

Long Short-Term Memory (LSTM) algorithm: LSTM is a type of recurrent neural network (RNN) specifically designed to learn from time-series data. It addresses the vanishing gradient problem common in traditional RNNs, making it particularly effective for sequences with long-range dependencies.

Sequence Imputation: The LSTM model learns to identify patterns in PoW data by leveraging temporal relationships between data points, filling in missing values by accurately predicting the next point in the sequence based on previous observations.

Noise Reduction: The architecture can learn to distinguish between actual signals and noise within the data, which enhances the fidelity of the imputed values and the overall dataset integrity.

The model is trained on real-world PoW datasets from ORNL’s Grid Signature Library. Specific fault conditions (e.g., single-phase to ground, two-phase, and three-phase to ground faults) and ID-filtered data help calibrate the model to handle the nuances of missing data under fault conditions.

Data Source: Real-world PoW data from the Grid Signature Library (https://gesl.ornl.gov/).

Data Characteristics: Data with a 1% loss ratio, specifically filtered for: Fault Types: Single-phase to ground, Two-phase, Three-phase to ground. (Provider 1))

ID Numbers: The following IDs are used for filtering: ['242', '243', '245', '246', '247', '252', '253', '254', '255', '266'].

Imputed point on wave data

https://hub.docker.com/repository/docker/openenergydatainitiative/ornl-data-imputation

https://github.com/bol22/POW-data

Real-world PoW data from the Grid Signature Library (https://gesl.ornl.gov/).

https://github.com/bol22/POW-data

The evaluation metric used in this scenario is the Root Mean Square Error (RMSE) range ratio, which measures the imputation accuracy by comparing RMSE to the range of observed values. The average error values across all cases are 0.20396% for Phase A, 0.20312% for Phase B, and 0.20266% for Phase C