Time-based Scheme for Demand Side Management of ML-based Forecasted Load Data using an Optimization Algorithm

Abstract

This study focuses on developing a methodology to enhance the efficiency and sustainability of electrical grid operations through advanced load forecasting and demand-side management (DSM) strategies. This research aims to provide a robust solution to manage electricity consumption effectively, ensuring a balance between supply and demand. The study begins with an in-depth review of existing load forecasting methodologies, including traditional statistical approaches and modern machine learning techniques. Traditional methods, such as regression analysis and time series models (ARIMA), are discussed alongside more advanced techniques like neural networks and hybrid models, emphasizing their limitations and potential improvements. The research introduces a novel forecasting model integrating XGBoost with Particle Swarm Optimization (PSO) and an improved Long Short-Term Memory (LSTM) network. These models leverage comprehensive data from the PJM Hourly Energy Consumption dataset, providing a rich temporal coverage for accurate predictions. The dataset, spanning from 2002 to 2018, is meticulously pre-processed and split for training and testing to simulate real-world forecasting scenarios. The thesis also delves into various DSM strategies, including peak clipping, valley filling, load shifting, strategic conservation, and flexible load shaping. These techniques are essential for optimizing power consumption patterns, reducing operational costs, and enhancing environmental sustainability by minimizing carbon emissions. The implementation of these DSM strategies is analyzed in terms of their sociocultural, environmental, and ethical impacts. The study also highlights the benefits of incentivizing off-peak electricity usage, rate adjustments in achieving cost efficiency and reliable energy supply.