Most electric energy consumption in the building sector is provided by fossil fuels, leading to high greenhouse gas emissions. However, the increasing need for sustainable infrastructure has triggered a significant trend toward smart buildings, which enable optimal and efficient resource usage. In this context, accurate mid-term energy load forecasting is crucial for energy management. This study proposes a hybrid forecasting model obtained through the combination of machine learning (ML) and deep learning (DL) approaches designed to enhance forecasting accuracy at an hourly granularity. The hybrid two-layer architecture first investigates the model’s performance individually, such as decision tree (DT), random forest (RF), support vector regression (SVR), Extreme Gradient Boosting (XGBoost), FireNet, and long short-term memory (LSTM), and then combines them to leverage their complementary strengths in a two-layer hybrid design. The performance of these models is assessed on smart building energy datasets with weather data, and their accuracy is measured through performance metrics such as mean squared error (MSE), root mean squared error (RMSE), and R-squared ($R^2$). The collected results show that the XGBoost outperformed other ML models. However, the hybrid model obtained by combining FireNet and XGBoost models delivers the highest overall accuracy for the performance parameters. These findings highlight the effectiveness of hybrid models in terms of prediction accuracy. This research contributes to reliable energy forecasting and supports environmentally sustainable practices.

In smart buildings, time series forecasting of electrical load is essential for energy optimization, demand response, and overall building performance. However, the mid-term load forecasting (MTLF) can be particularly challenging due to several uncertainties, such as sensor malfunctions, communication failures, and external environmental factors. These problems can lead to missing data patterns that may impact the accuracy and reliability of forecasting models. The purpose of this study is to explore the impact of random missing data patterns on the MTLF predictions’ accuracy. Therefore, several data imputation techniques are evaluated using a complete dataset (i.e., with no missing values) acquired on a smart commercial building, and their influence on load forecasting performance is assessed when different percentages of randomly distributed missing data patterns are assumed. Moreover, the deep learning (DL) approach based on a recurrent neural network, namely, long short-term memory (LSTM), is employed to predict the smart building electrical energy consumption. The obtained outcomes demonstrate that the pattern of random missing data significantly impacts the forecasting accuracy, with machine learning (ML) imputation techniques having better results than statistical and hybrid imputation techniques. Based on these findings, it is evident that robust data preprocessing and the handling of missing values are important in order to improve the accuracy and reliability of mid-term electrical load forecasts.

The operation and management of Smart Distribution Networks (SDNs) heavily relies on the Short-Term Load Forecasting (STLF) function in the context of Demand-Side Management (DSM), Distributed Energy Resource (DER) integration, and grid reliability. With the wide use of smart meters and Advanced Metering Infrastructure (AMI), it is now possible to obtain high-resolution electricity consumption data, which has opened up new opportunities for data-driven forecasting at the transformer, feeder, substation, and consumer levels. Forecasting in SDNs, however, still faces difficulties because of the variability of the loads, the heterogeneity of the data, the distributed nature of the energy resources, and the stochastic behavior of consumers. This paper provides a systematic literature review of clustering-based and data-driven STLF methodologies for Smart Distribution Systems (SDNs). According to PRISMA guidelines, publications from 2018 to March 2026 were retrieved from Google Scholar, Scopus, and Web of Science and analyzed in terms of clustering techniques, forecasting methods, data sources, and fields of application. The focus is on K-Means, K-Medoids, Hierarchical Clustering, Dynamic Time Warping (DTW), and DBSCAN, along with statistical, Machine Learning (ML), Deep Learning (DL), Federated Learning (FL), and Edge-computing forecasting frameworks. The review indicates that the clustering-based preprocessing is highly useful in the field of characterizing and forecasting load profiles. In the reviewed studies, DL models, especially LSTM, GRU and CNN models, were repeatedly found to perform better than statistical and ML baseline architectures in their respective experimental scenarios, while FL and edge-computing methods show promise for real-time and privacy-preserving forecasting. Lastly, the review presents existing problems and future research directions on how to create scalable, intelligent, and privacy-preserving forecasting systems in SDNs.