Forecasting of individual electricity consumption using Optimized Gradient Boosting Regression with Modified Particle Swarm Optimization

The task of forecasting consumers' energy consumption is currently a trend in energy supply companies. An accurate prediction of energy consumption is a powerful tool to check for inconsistencies between what is recorded and the actual amount consumed. In practice, Brazilian energy companies verify inconsistencies in the manual reading of consumption, using a consumption range based on the predicted consumption. This consumption forecast is currently realized by the average of previous consumptions and, therefore, can be improved by the use of machine learning techniques. For this purpose, an Optimized Gradient Boosting Regressor (OGBR) was proposed, which has been optimized by a modified version of the Particle Swarm Optimization (PSO) for fast parameter optimization. The OGBR prediction results on a dataset of over 2 million consumers were compared with its unmodified version and with the Seasonal and Trend decomposition using Loess (STL). In addition, the forecast stability of the OGBR over 12 months was evaluated. Therefore, the energy consumption forecasting performance was improved by using the OGBR and this performance was better than its unmodified version, in all validation metrics, and better than STL, in most classes of consumption.

Automated summary

Forecasting consumers' energy consumption is a trend in energy supply companies, with a focus on improving prediction accuracy. Brazilian energy companies use a consumption range to verify inconsistencies in manual readings, and machine learning techniques can enhance consumption forecasts. The Optimized Gradient Boosting Regressor (OGBR) proposed in this study outperformed its unmodified version and the Seasonal and Trend decomposition using Loess (STL) in most cases of consumption.