Leandro Von Krannichfeldt

Residential electricity consumption represents a substantial share of total demand in many countries, especially in cold-climate regions with high electrification rates. These systems show strong intra- and inter-seasonal variability driven by behavioral, climatic, and technological factors. As renewable energy integration and electrification accelerate, understanding residential consumption heterogeneity becomes increasingly important for grid operation and demand response design. Yet, despite the growing availability of smart meter data, large-scale residential customer segmentation remains limited, constraining the generalizability of existing studies. This study presents a scalable, data-driven segmentation framework using Switzerland as a representative case study with relevance to similar contexts worldwide. It analyzes smart meter data from more than 53,000 households. Representative daily and seasonal weekday/weekend profiles are constructed by averaging 15-minute measurements and normalizing them to emphasize load shape rather than absolute magnitude. These profiles are then clustered using k-means, with Davies-Bouldin index validation, to identify three distinct consumer groups with characteristic daily and seasonal electricity-use patterns.The identified clusters correspond to three robust and interpretable residential archetypes: (i) a large group with moderate dual-peaked daily profiles typical of conventional family and workday routines, (ii) a smaller but high-consuming group with persistently elevated baseload and weak diurnal variation, suggestive of all-day occupancy or electric heating, and (iii) a group with pronounced nighttime peaks and strong seasonal variability, likely influenced by heat pumps or night-setback water heating technologies. Seasonal and weekday-weekend comparisons show that winter load increases are especially pronounced in the high-baseload and nighttime-peaking clusters, while the conventional routine cluster remains comparatively stable. Weekend profiles exhibit higher load factors and flatter peaks, highlighting the combined influence of technology, occupancy, and behavior on residential demand. These differentiated load profiles provide actionable insights for utilities and policymakers seeking to design targeted time-of-use tariffs and more effective demand response strategies.

In practical engineering, electricity theft detection is usually performed on highly imbalanced datasets (i.e., the number of fraudulent samples is much smaller than the benign ones), which limits the accuracy of the classifier. To alleviate the data imbalance problem, this paper proposes simple data augmentation tricks (SDAT) to boost performance on electricity theft detection tasks. SDAT includes five simple but powerful operations: adding noises to electricity consumption readings, drifting values of electricity consumption readings, quantizing electricity consumption readings to a level set, adding a fixed value to electricity consumption readings, and adding changeable values to electricity consumption readings. In addition, eight potential tricks are also mentioned. Numerical simulations are conducted on a real-world dataset. The simulation results show that SDAT can significantly boost the performance of different classifiers, especially for small datasets. Besides, specific suggestions on how to select parameters of SDAT are provided for its migration use to other datasets.

Higher accurate load forecasts help the power system operator make better resource allocation and reduce operational costs. Ensemble learning has been widely used to improve the accuracy of final forecasts by combining multiple individual forecasts. In the digital economy era, the system operator can buy high-quality load forecasts from the data market and then combine them in an ensemble model to further enhance the quality of final forecasts. Consequently, the operator should share its operational profit (or reduced cost) from forecasting improvement with forecast providers (agents). However, forecasts from different agents jointly affect the performance of the ensemble model, making it hard to quantify the contribution of each individual forecast. Even though several works have been done on the smart grid data market, there are very few works regarding energy forecast trading and valuation. To fill this gap, this paper builds up a novel framework for day-ahead load forecast trading and valuation in an ensemble model, which includes historical credit evaluation, data transaction, and payoff allocation. Specifically, three categories of payoff-allocating schemes with distinct characteristics are proposed and compared in terms of applicable scope, computational complexity, and synergy consideration. Case studies on a real-world dataset illustrate how individual forecasts can be evaluated in an ensemble model.

This paper presents a new privacy-preserving framework for the short-term (multi-horizon) probabilistic forecasting of nodal voltages in local energy communities. This task is indeed becoming increasingly important for cost-effectively managing network constraints in the context of the massive integration of distributed energy resources. However, traditional forecasting tasks are carried out centrally, by gathering raw data of end-users in a single database that exposes their private information. To avoid such privacy issues, this work relies on a distributed learning scheme, known as federated learning wherein individuals’ data are kept decentralized. The learning procedure is then augmented with differential privacy, which offers formal guarantees that the trained model cannot be reversed-engineered to infer sensitive local information. Moreover, the problem is framed using cross-series learning, which allows to smoothly integrate any new client joining the community (i.e., cold-start forecasting) without being plagued by data scarcity. Outcomes show that the proposed approach achieves improved performance compared to non-collaborative (locally trained) models, and is able to reach a trade-off between privacy and performance for different architectures of deep learning networks.

The wide popularity of smart meters enables the collection of massive amounts of fine-grained electricity consumption data. Extracting typical electricity consumption patterns from these data supports the retailers in their understanding of consumer behaviors. In this way, diversified services such as personalized price design and demand response targeting can be provided. Various clustering algorithms have been studied for electricity consumption pattern extraction. These methods have to be implemented in a centralized way, assuming that all smart meter data can be accessed. However, smart meter data may belong to different retailers or even consumers themselves who are not willing to share their data. In order to better protect the privacy of the smart meter data owners, this paper proposes two federated learning approaches for electricity consumption pattern extraction, where the k-means clustering algorithm can be trained in a distributed way based on two frequently used strategies, namely model-averaging and gradient-sharing. Numerical experiments on two real-world smart meter datasets are conducted to verify the effectiveness of the proposed method.