Sliding Window GBDT for Electricity Demand Forecasting

Abstract

Electricity consumption prediction plays a crucial role in energy management and urban planning. Accurate predictions enable utilities to optimize their power generation and distribution, while city planners can make informed decisions regarding infrastructure development and resource allocation. V arious methods have been proposed for electricity consumption prediction, including time series models, machine learning models, and hybrid models. Gradient Boosting Decision Tree (GBDT) models have shown promising results in several fields. However, traditional GBDT models do not consider the temporal dependencies of the input data, which can limit their predictive performance. To address this issue, sliding window approaches have been proposed to incorporate temporal dependencies into GBDT models. In this study, we propose a method for predicting city-level electricity consumption based on GBDT models with sliding windows. Our approach aims to capture both the temporal and spatial characteristics of electricity consumption patterns by training a GBDT model for each city using input features such as weather data, time-of-day, and day-of-week indicators. We evaluate our method using various metrics and show that it outperforms baseline models and other state-of-the-art methods in terms of accuracy and robustness.