FORECASTING PEAK LOAD ON THE POWER GRID

Abstract

management of power grids. The paramount importance of this task necessitates a comprehensive examination of various forecasting methodologies, leveraging hourly electricity consumption data and a diverse array of predictive models.

This article is dedicated to a thorough analysis of distinct peak load forecasting methods, elucidating the research methodology encompassing data preprocessing, model selection, and parameter optimization. The models under scrutiny encompass a spectrum of techniques, including ARIMA, SARIMA, LSTM, GRU, and Random Forest. To gauge their performance, a suite of evaluation metrics such as Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), Mean Absolute Percentage Error (MAPE), R-squared, and Receiver Operating Characteristic Area Under the Curve (ROC AUC) were employed.

The findings of this investigation underscore the nuanced strengths and limitations inherent to each forecasting model when tasked with predicting peak electricity consumption. Notably, certain approaches exhibit superior accuracy in short-term forecasting scenarios, while others excel in long-term predictions. The selection of the optimal forecasting method becomes contingent upon the specific conditions, constraints, and objectives of the study at hand.

The LSTM and GRU models, representing deep learning neural networks, manifest their prowess in addressing the intricate dynamics of electricity consumption data. Their capacity to discern intricate patterns, nonlinearities, and long-term dependencies positions them as formidable contenders in the domain of long-term peak consumption forecasting.

The Random Forest model emerges as a versatile choice, adept at accommodating the multifaceted characteristics of electricity consumption data. Its ability to autonomously identify complex dependencies, nonlinear relationships, and seasonal patterns while considering external factors amplifies its utility across a broad spectrum of forecasting scenarios.

This comprehensive work is of great importance for the practical study of various methods of forecasting peak electricity consumption. The results obtained from this analysis have significant implications for improving power grid management strategies, ultimately contributing to microgrid stability and resilience.