Advancements in hybrid and ensemble ML models for energy consumption forecasting: results and challenges of their applications

Abstract

The optimization of energy consumption is crucial for energy management at each level. This research study investigates the different methods that have been used over time to calculate energy consumption using combined machine learning models, specifically hybrid and ensemble models. The first one mentioned combines different machine learning techniques with statistical methodologies that aim to handle non-linear relationships in complex data sets. Some of these representative algorithms are Autoregressive Integrated Moving Average–Long Short-Term Memory (ARIMA-LSTM), hybrid augmented with Generative Adversarial Network (GAN) and Particle Swarm Optimization–Stacking (PSO-Stacking) which have demonstrated remarkable accuracy in various contexts, including residential, commercial and industrial energy systems. On the other hand, ensemble models that include stacking, boosting and bagging methods have been used to reduce calculation errors and handle large-scale data sets that often have heterogeneous behaviors. The results of this literature review indicate that the selection of an appropriate model that combines different machine learning techniques depends on the nature of the data, the objective and the context of the research, since, as mentioned, hybrid models are more effective in terms of complex, temporal and non-linear data, while ensemble models are more versatile in managing high-dimensional data sets and reducing errors. In addition, the main challenges in this type of work are identified, including computational load and data quality. Given this, it was found that solutions such as the implementation of metaheuristic algorithms and feature selection are available to solve these types of problems.As of this writing, systematic reviews specifically comparing combined machine learning models applied to energy consumption calculation are limited. Therefore, this literature review is a novel starting point for future research that wants to focus on this specific field and needs to solve challenges such as data complexity, time de-pendencies, and computational load, which are oriented to the process of organizing and managing energy consumption. On the other hand, this research offers a perspective that analyzes and contrasts the usefulness and effectiveness of these models in different contexts, as well as identifying advantages, disadvantages, limitations, and opportunities in various areas. • Optimization of energy consumption via hybrid and ensemble ML models. • Ensemble ML models to reduce errors and handle large-scale and heterogeneous data sets. • Hybrid models to manage complex, temporal and non-linear data. • Methodology for the selection of an appropriate ML model based on data nature.