Analytics and Predictive Modeling for Load Forecating

What is Load Forecasting?

Load forecasting is the use of statistical forecasting techniques to predict electrical power generation requirements. There are three categories of load forecasting: short term (within ~1 day), medium term (weeks), and long term (years). The terms correspond to different planning periods for generator activation, fuel and electrical power markets, and infrastructure development/capital investment.

Why Use Predictive Modeling for Load Forecasting?

For short term planning, the supply and the load on an electrical grid must always be balanced to maintain power quality and avoid blackouts. However, keeping additional generation capacity online or actively producing more power than needed is expensive. Likewise, reliance on short term emergency generation or load shedding (cutting power to large consumers) in the event of not having sufficient reserves results in exorbitant costs. These types of loads are heavily dependent on the weather and local events but possess an inherently stochastic nature making statistical predictive modeling a particularly effective choice for forecasting.

Typically, electricity providers predict medium-term and long-term loads with methods like micro- or macro-economic models for a given grid region. Much like physics-based models, these load models rely on knowing phenomena rather than on replicating observed behavior with statistical models. Given the number of empirical parameters and high degree of uncertainty in these models, Uncertainty Quantification and predictive modeling techniques can be useful for these models like physics-based simulations.

Challenges of Short-Term Load Forecasting

In load forecasting, accuracy directly translates into better planning and lower costs. Thus, providers always push to predict loads more accurately across a wide range of scenarios and with a longer lead time. Additional cost reductions can come from reducing the engineering effort of creating short-term forecasting models by making it easier to combine information and by making predictive model fitting quicker or less computationally expensive.

SmartUQ Solutions for Predictive Modeling in Short-Term Load Forecasting

SmartUQ addresses these challenges using a variety of analytics tools:

• Cutting-edge predictive modeling tools make generating highly accurate models quick and easy.
• Dimension reduction, data subsampling, and filtering tools can reduce the amount of data that needs to be processed to fit a forecasting model to a new scenario.
• Sensitivity analysis allow important inputs to be identified and analyzed.
• Statistical calibration and inverse techniques can be used to identify bias signal in noisy data, determine underlying input uncertainties, and tune operating parameters for medium- and long-term models.

Summary

Advanced load forecasting techniques remain an area of active research for many companies due to the potential benefits of making more accurate and faster predictions. What can your engineers do with easier to use tools and more accurate models?

To learn more about analytics for load forecasting applications, email us at [email protected].