Energy demand forecasting for smart grids

Energy demand forecasts are a key input for decision making by utility companies and network operators, e.g., bidding load and generation into electricity markets, planning maintenance and day-ahead outages, managing congestions, monitoring asset utilization and health. New challenges are arising from distributed generation (in particular, renewable energy sources) and changes in regulation which put pressure on utilities to provide reliable high-quality power supply under unforeseen network conditions while increasing efficiency. IBM Research Ireland has developed systems and analytics capabilities to deliver next-generation demand forecasting services. A particular focus is on forecasting demand at disaggregated levels (e.g., distribution substations, MV/LV transformers), accounting for distributed generation, utilizing high-resolution weather forecasts, quantifying uncertainty associated with demand forecasts, and developing data models for integrating and curating information from heterogenous sources (telemetry, revenue metering, GIS etc). The solutions have been deployed and validated with leading energy & utility customers in Europe and North America.



Our solution is based on generalized additive models (GAM), a widely popular class of non-linear statistical regression models.

As inputs, our models utilize weather information (temperature, dew point, irradiance, cloud cover etc.), calendar information (time of day, time of year etc) and other socioeconomic variables (bank holidays, vacation periods, demand response events etc).

A particular focus of our work is on data curation, i.e., integrating information from different sources and automatically handling data anomalies in order to obtain consistent training data sets.

Another important area is the development of tools which allow non-power users to create, deploy and maintain demand forecasting models throughout their entire lifecycle. In order to handle uncertainty associated with demand forecasts, we have developed approaches for dealing with the three main constituents of uncertainty: inherent randomness, model bias/variance, uncertainty in the inputs.

We have also developed methods for deriving optimal spatio-temporal weather features from high-resolution weather forecasts with the objective of maximizing the forecasting accuracy.

Energy demand forecasting

Demand modelling:
Joint effect of time-of-day and temperature.

Demand modelling


Forecasting uncertainty in electricity demand
Tri Kurniawan Wijaya, Mathieu Sinn, Bei Chen
AAAI-15 Workshop on Computational Sustainability, Austin, TX, 2015.

Improved electricity load forecasting via kernel spectral clustering of smart meters
Carlos Alzate, Mathieu Sinn
IEEE 13th International Conference on Data Mining (ICDM), pp. 943-948, 2013.

Adaptive learning of smoothing functions: Application to electricity load forecasting
Amadou Ba, Mathieu Sinn, Yannig Goude, Pascal Pompey
Advances in Neural Information Processing Systems (NIPS), pp. 2510-2518, 2012