Electricity demand forecast model based on meteorological and historical demand data using artificial neural networks

Abstract

Accurate forecasting of electricity demand is crucial for improving transmission system operation through optimized use of resources, operation planning, and minimized outages. The dynamic of electricity demand depends on exogenous factors (e.g., meteorological conditions), but the relationships between demand and factors are complex and nonlinear, posing a challenge for accurate prediction. With the aim of predicting electricity demand, this work explores the relationship with meteorological conditions for the province of Entre Ríos (Argentina). We propose a recurrent neural network model based on long short-term memories, which receives the raw input data without feature extraction. We evaluate its performance and compare it with a state-of-the-art method. The exploratory analysis of the data shows that temperature extremes present a strong influence on consumption patterns. The proposed models achieve a performance of 0.77 in determination coefficient when comparing predicted electricity demand with observations. This indicates the potential as a powerful tool for optimizing the system operation in Entre Ríos.