OEDI SI/Scenarios/Event Classification for Field Measured Data

From Open Energy Information

Event Classification for Field Measured Data​ Summary

  • Date Created: 2024/09/24
  • Organization: PNNL
  • Objective: The scenario presents a convolution neural network (CNN) developed to identify voltage events at a photovoltaic (PV) inverter. Our CNN is trained on synthetic event data generated for a modified IEEE13 feeder. We simulate two common voltage events: faults and voltage sags. The CNN is built to evaluate both voltage and current waveforms from three-phase PVs and has excellent identification on training data. Field measured data is used to test the CNN performance on real events. The CNN has high performance in identification of faults and voltage sags.
  • Use Case: Event Detection and Identification
  • Methodology
    • Inputs
      • Outputs
        • Configuration
          • Webinars


          Docker Container

          Download the docker container at:


          Run Locally

          http://localhost:8080/edit_scenario?Event Classification for Field Measured Data
              Run Locally


          Input Data

          Download input data at:


            Output Data

            Download output data at:

              References


                Back to Event Detection and Identification



                Property "Objective" (as page type) with input value "The scenario presents a convolution neural network (CNN)</br>developed to identify voltage events at a photovoltaic (PV)</br>inverter. Our CNN is trained on synthetic event data generated</br>for a modified IEEE13 feeder. We simulate two common voltage</br>events: faults and voltage sags. The CNN is built to evaluate</br>both voltage and current waveforms from three-phase PVs and</br>has excellent identification on training data. Field measured data is used to test the CNN performance on real events.</br>The CNN has high performance in identification of faults and voltage sags." contains invalid characters or is incomplete and therefore can cause unexpected results during a query or annotation process.


              1. A learning-based approach using convolutional neural networks (CNNs) to identify voltage events at PV location is presented.
              2. The CNN-based model acts as a data-driven relay which learns hidden structures and patterns in the voltage waveforms to identify the events like sag and faults.
              3. The Python scripts to generate transient data for using the ATP simulator are provided in the transient data generation use-case.
              4. The method is demonstrated on IEEE-13 bus model and its ATP model file is available https://github.com/pnnl/oedisi_transients
              5. The point-on-wave (POW) voltage waveforms at the PV location obtained using ATP simulations are vertically stacked as an image and used as input to the model.
              6. Field measured voltage waveforms in p.u are used to validate the trained CNN model.
              7. Each waveform is classified into 3 labels. (a) no fault (b) fault (c) voltage sag
              8. The codes in docker container are generalized to be used for any network’s dataset. The CNN model is trained for voltage waveform in p.u. Hence it can be used for network model of different voltage levels and can be tested on field data of a different location and voltage levels.
              9. https://hub.docker.com/r/openenergydatainitiative/pnnl-event-detection-field-cnn
              10. Sample simulation and field measured data is available at : https://github.com/pnnl/oedisi_transients
              11. After the CNN model is trained it saves training accuracy and loss. It also plots the confusion matrix and saves the trained model to the local workstation.
                12. Retrieved from "https://openei.org/w/index.php?title=OEDI_SI/Scenarios/Event_Classification_for_Field_Measured_Data&oldid=1084771"