PowerNet: Neural Power Demand Forecasting in Smart Grid

TL;DR

PowerNet is a novel neural network architecture designed for accurate power demand forecasting in smart grids, effectively integrating diverse data types and outperforming existing methods in reducing forecasting errors.

Contribution

The paper introduces PowerNet, a new neural network model that incorporates heterogeneous features for improved power demand forecasting, outperforming GBT and SVR methods.

Findings

PowerNet reduces forecasting error by 33.3% compared to GBT.

PowerNet reduces forecasting error by 14.3% compared to SVR.

Empirical analysis of forecast horizon and retraining frequency for PowerNet.

Abstract

Power demand forecasting is a critical task for achieving efficiency and reliability in power grid operation. Accurate forecasting allows grid operators to better maintain the balance of supply and demand as well as to optimize operational cost for generation and transmission. This article proposes a novel neural network architecture PowerNet, which can incorporate multiple heterogeneous features, such as historical energy consumption data, weather data, and calendar information, for the power demand forecasting task. Compared to two recent works based on Gradient Boosting Tree (GBT) and Support Vector Regression (SVR), PowerNet demonstrates a decrease of 33.3% and 14.3% in forecasting error, respectively. We further provide empirical results the two operational considerations that are crucial when using PowerNet in practice, i.e., how far in the future the model can forecast with a decent accuracy and how often we should re-train the forecasting model to retain its modeling capability. Finally, we briefly discuss a multilayer anomaly detection approach based on PowerNet.