MIDS Capstone Project Spring 2023
Real-Time Electric Load Forecasting
Overview
Reliable electricity is one aspect of our daily lives that we often take for granted — when was the last time you thought about what is powering your household lights and appliances?
But behind the scenes, electricity generation is quite complex. Because it is not freely available in nature, it must be produced. In order to reliably run an electrical grid and generate the necessary electricity to power our cities, system operators must first know the amount of electrical load that must be served. Failure to properly predict and supply load can result in disruptive blackouts.
One of the most recent disruption was the 2021 Great Texas Freeze. Texas suffered a major power crisis when three severe winter storms hit the United States, triggering the worst energy infrastructure failure in Texas state history. The Electric Reliability Council of Texas (ERCOT), which supplies power to 90% of the state, greatly underestimated electric demand. This resulted in at least 57 deaths and over $195 billion in property damage.
Therefore, electric load forecasting, which is predicting the amount of electrical power required to meet customer demand, is incredibly important. By better predicting load, we can
- Reduce Liability: If we have enough electricity, we reduce potential major blackouts, saving billions in property damages.
- Decrease Price of Electricity: Electricity bills can get pricy. By preventing waste, we can optimize for more cost saving measures for consumers.
- Increase Reliability: Better forecasts means we can better plan capacity and ensure consumers are supplied with the required energy to keep their homes running.
Project Details
While there are already several load forecasting programs on the market, these programs primarily focus on forecasting hourly loads days in advance. Our objective is to provide system operators with load in real time, allowing them to account for more sensitive real time weather events like sudden drops in wind or a large system of clouds rolling in.
Our team explored four different classes of models and leveraged historic time series load data, real-time weather data, and real-time solar data to predict [up to 90 minutes in advance] the forecasted electric load. We focused on New York City and plan on partnering with New York Independent Systems Operators (NYISO) to implement this model in their control center.
Course
Data Science 210. Capstone, Spring 2023
