Weather Forecast to Predict Site Outage

The restoration process right after the wake of extreme weather has been a main concern for utilities. The unpredictability of power outages leaves utilities significantly less time to improve power storage along with a substantial amount of operating cost. This ultimately leads to customer dissatisfaction and is a major obstacle.

Predicting Outage uses available meteorological data and other forecasting expertise to develop machine learning predictive models and thus proactively prepares companies beforehand regarding weather anomalies. Thus this enhances electric system reliability and benefits utility customers. With the best arrangements, we are improving the functionality and reliability of the system by generating the most accurate predictions of power outages and better characterizing the confidence of weather forecasts.

Model Accuracy 92%

Weather Forecast Based Proactive Resource Mobilization to prevent Site Outage Due to Long Term Power Outage. Next Day Weather Forecast Accuracy: 75-80%

Overall Accuracy 70%

Heavy Rainfall and Storm triggers Grid Power Outage and requires telecom sites to run on back up power. For prolonged outage Generator and addition power back up resources needs to be mobilized. Our solution identifies sites prone to high duration outage and informs on ground maintenance teams so that they can get prepared to handle long hour Grid Failure.

 

What Internal Teams Tried Before

  1. Outage Data Visualization over time and by region/site. Sites go down more during monsoon specially during cyclones.
  2. Correlate Flood/water Level around the region with Site Outage. No Usable Outcome
  3. Correlate Rainfall with Site Outage. No Usable Outcome

How Are We Different

  1. Correlating Multiple (6+) Weather Parameters used instead of 1 or 3 – without Machine Learning it is nearly impossible to do such model development.
  2. Every site is different – so there is NO SINGLE MODEL CAN FIT ALL – we deployed out Micro-Modelling Technique – that generate different model for different sites.
  3. Automation of Data Collection
  4. Use the Prediction to Operations – put it in use.

Our Solution