Recently, there has been growing interest in using machine learning based methods for forecasting renewable energy generation using time-series prediction. Such forecasting is important in order to optimize energy management systems in future micro-grids that will integrate a large amount of solar power generation. However, predicting solar power generation is difficult due to the uncertainty of the solar irradiance and weather phenomena. In this paper, we quantify the impact of uncertainty of machine learning based time-series predictors on the forecast accuracy of renewable energy generation using long-term time series data available from a real micro-grid in Sweden. We use clustering to build different ML forecasting models using LSTM and Facebook Prophet. We evaluate the accuracy impact of using interpolated weather and radiance information on both clustered and non-clustered models. Our evaluations show that clustering decreases the uncertainty by more than 50%. When using actual on-side weather information for the model training and interpolated data for the inference, the improvements in accuracy due to clustering are the highest, which makes our approach an interesting candidate for usage in real micro-grids.

Accurate and robust forecasting is a critical challenge for the reliable operation and optimization of smart microgrids, especially as they increasingly integrate renewable energy sources and rely on edge computing infrastructure. Uncertainty from weather, rapidly changing consumption patterns, and the resource limitations of edge devices hinder the deployment of conventional forecasting models, which typically assume stationary data and require large, clean datasets. This licentiate thesis addresses these gaps by developing a comprehensive and adaptive pipeline for time-series forecasting in microgrids, with a focus on resilience to uncertainty, concept drift, data scarcity, and resource constraints. The research presents a suite of novel and scalable approaches, including clustering-based pre-processing, adaptive deep learning models, probabilistic and uncertainty-quantifying forecasting methods, federated learning, and instance-based transfer learning. These techniques are designed to deliver high predictive accuracy and adaptability under non-stationary conditions, while remaining lightweight enough for deployment on constrained edge devices. Privacy-preserving and collaborative mechanisms, such as federated and transfer learning, enable model generalization and robustness across heterogeneous, distributed microgrid environments. Building on seven peer-reviewed publications, this work establishes a unified pipeline for intelligent forecasting at the grid edge. Key contributions include: (1) dynamic drift-adaptive learning frameworks for real-time adaptation, (2) probabilistic and uncertainty-quantifying forecasting for risk-aware energy management, (3) multi-step prediction strategies for short- and medium-term forecasting, and (4) clustering-guided transfer learning to enhance prediction in data-scarce scenarios. Validation on both real-world and simulated microgrid datasets demonstrates substantial gains: for example, the proposed transfer learning approach improves short-term forecasting accuracy in data-limited buildings by up to 40%, while adaptive and federated models achieve robust performance with reduced computational and communication overhead compared to centralized baselines. Overall, this thesis advances the state-of-the-art in interpretable, scalable, and resilient forecasting for smart microgrids, supporting the transition to adaptive, decentralized, and privacy-aware energy systems capable of thriving in dynamic and uncertain environments.

Load forecasting is a crucial topic in energy management systems (EMS) due to its vital role in optimizing energy scheduling and enabling more flexible and intelligent power grid systems. As a result, these systems allow power utility companies to respond promptly to demands in the electricity market. Deep learning (DL) models have been commonly employed in load forecasting problems supported by adaptation mechanisms to cope with the changing pattern of consumption by customers, known as concept drift. A drift magnitude threshold should be defined to design change detection methods to identify drifts. While the drift magnitude in load forecasting problems can vary significantly over time, existing literature often assumes a fixed drift magnitude threshold, which should be dynamically adjusted rather than fixed during system evolution. To address this gap, in this paper, we propose a dynamic drift-adaptive Long Short-Term Memory (DA-LSTM) framework that can improve the performance of load forecasting models without requiring a drift threshold setting. We integrate several strategies into the framework based on active and passive adaptation approaches. To evaluate DA-LSTM in real-life settings, we thoroughly analyze the proposed framework and deploy it in a real-world problem through a cloud-based environment. Efficiency is evaluated in terms of the prediction performance of each approach and computational cost. The experiments show performance improvements on multiple evaluation metrics achieved by our framework compared to baseline methods from the literature. Finally, we present a trade-off analysis between prediction performance and computational costs.

Accurate forecasting of power production and consumption is essential for the efficient operation of smart energy grids, enabling stable energy exchange and grid reliability. However, the growing integration of photovoltaics (PVs) and electric vehicles introduces significant uncertainty. This paper evaluates multiple Machine Learning (ML) models for single- and multi-step forecasts of PV generation and household consumption, incorporating uncertainty bounds to inform operator decisions. We use data from two Swedish sites and the CityLearn benchmark dataset to compare direct, recursive, and hybrid multi-step strategies. LightGBM with gradient-boosted quantile regression achieves the best single-step performance, with Mean Absolute Error (MAE) as low as 10.19 W in Halmstad and 16.12 W in Uppsala. For multi-step forecasts, the direct method outperforms others, reaching a 48 h consumption MAE of 71.08 W in Uppsala and 52.05 W in Halmstad, and achieving prediction interval coverage probabilities above 0.95. Moreover, personalized models trained on individual households outperform generalized ones, even with smaller datasets, highlighting the value of tailored approaches for improving forecast accuracy under uncertainty.