Identifikation und Optimierung von Lastverschiebungspotenzialen durch Smart Meter-Daten

Abstract

The rapidly growing share of renewable energy sources and the increasing electrification of various sectors are placing ever greater demands on the power grid in terms of load management and grid stability. In recent years, not only large-scale consumers have been responsible for maintaining grid stability, individual households can also make a significant contribution through targeted load shifting, helping to relieve the grid and reduce electricity costs. The aim of this thesis is to identify the potential of temporally shiftable loads based on smart meter data and to derive economic benefits from them. The analyses are based on high-resolution electricity consumption data from private households, recorded in 30-minute intervals over an extended period. After targeted preprocessing of the data, clustering is performed to group typical consumption behaviors. Based on the identified groups, deep neural networks are trained, specifically using Gated Recurrent Units (GRUs) to forecast future load developments using a multi-step prediction approach. The model quality is evaluated using standard metrics such as MAE, RMSE, and the coefficient of determination (𝑅 2 ). Next steps involve the development of load shifting scenarios based on the original input data. In particular, charging sessions of electric vehicles are automatically identified from the smart meter data using rule-based algorithms and then shifted to more cost-effective time slots based on dynamic electricity tariffs (aWATTar). In addition to a power threshold, the minimum charging duration is also considered to reliably detect charging events. The simulation includes various scenarios, distinguishing between weekdays and weekends, charging durations, and daytime versus nighttime charging behavior. The results show that intelligent temporal scheduling of charging sessions can lead to electricity cost savings of up to 45% without requiring any additional input from users. At the same time, grid loads can be reduced in a targeted manner, contributing to the stabilization of the overall system. This thesis thus provides a scalable, datadriven methodology for the forecasting and optimization of major household loads and demonstrates how smart meter data can serve as a foundation for flexible, decentralized load control in sustainable energy systems.

Date of Award	2025
Original language	German (Austria)
Supervisor	Andreas Müller (Supervisor)