M. Berlinger, B. Cerruti, R. Derech, R. Vargas
Consolidated Edison Company of New York, Inc.
Con Edison has been using a decision support tool, known as the Con Edison Storm Impact Model (CESIM), since 2007. Its purpose has been to help our utility prepare and acquire resources in advance of an impactful storm. Linear regression is used to fit the CESIM from a database of weather and customer outage tickets. Forecasts are generated by manually entering weather variables into a spreadsheet containing the CESIM. . It currently focuses on Con Edison’s overhead electric system in Westchester County, NY which serves 1 million people with a spatial coverage of 310 mi2.
The first iteration of the tool, CESIM1.0 developed in 2007 relates wind speed, foliage, and two week rainfall to system impact. It operates with a database of 50 impact events. Con Edison updated the model (CESIM2.0) in 2014 to 217 impact cases to incorporate events such as Hurricane Irene, Superstorm Sandy, a powerful March 2010 nor’easter, and two wet snow events on 25 February 2010 and 31 October 2011. CESIM2.0 includes new variables for wet snow, wind speed and direction dependence, and a major storm identifier. Verification of the CESIM2.0 displays a strong bias to over predict events, especially less impactful events. This is because similar to CESIM1.0; CESIM2.0 is developed with a population of select cases. To address the bias in CESIM2.0 and port the tool to other territories of the Con Edison electric system, the utility is updating the model (CESIM3.0). The update is focused on creating an objective historical storm database from daily weather variables to ensure that all cases, ranging from major storms to blue sky days, are identified properly in CESIM3.0.CESIM1.0 and 2.0 use linear regressions to subjectively fit an algorithm used to generate impact forecasts. CESIM3.0 will incorporate multiple linear regressions to ensure the CESIM is fit objectively.
Daily temperature, wind gusts, and precipitation are extracted from 7 ASOS weather stations to build the database using the National Climate Data Center (NCDC) from 2000 to present. The wind data in particular is quality controlled through a process of eliminating erroneous wind gusts and correcting them with METAR reports. River discharge data is extracted from 16 United States Geological Survey (USGS) streamflow sensors from 2000 to present. Severe weather reports by county are obtained from the Storm Prediction Center (SPC) from 2000 to present. Daily snowfall is extracted from 13 counties, where available, between 2000 to present from NCDC COOP data. Outages are assessed via the collection of job tickets internally. A rollover technique is used to ensure that all outage tickets caused by a given storm are aligned with the initial storm event.
In CESIM2.0, soil moisture was replaced with a major storm identifier and foliage is subjectively calculated based on date. In CESIM3.0, an objective approach was implemented to assess soil moisture and foliage. Soil moisture is objectively measured using a normalization technique applied to USGS Daily average river discharge data. The normalization is computed by first calculating the averaged 60-day median discharge and averaged 60-day standard deviation discharge for each Julian day. Then the soil saturation level on a given day is calculated as a ratio of observed daily discharge to the sum of the averaged 60-day median and standard deviation. Foliage in CESIM3.0 is objectively measured through cumulative Growing Degree Days (GDD) and Dying Degree Days (DDD) to calculate the spring growth and autumn senescence, respectively. The CESIM3.0 foliage approach is calibrated with data from the PhenoCam network for two sites nearby Con Edison’s service territory.
In order to assess the performance of the latest enhancements, CESIM 2.0 and 3.0 forecasts are generated using both models, with each prediction converted into a corresponding staffing response level. Results show the latest enhancements performed 77% better than the previous model at predicting the correct staffing storm response level using a combined Brier Score.
References
Andrew D. Richardson, 2016: Phenocam Network, 1 August 2016. [Available online at https://phenocam.sr.unh.edu/webcam.]
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner