Power systems are currently experiencing a significant transformation as a result of the energy transition, which is driven by climate change. The electrification of transportation and heat together with renewables is reshaping the electricity energy industry towards a more sustainable and environmentally friendly. There is a pressing need to restructure the power system, allowing for greater flexibility. With digitalization, energy companies also have to handle big data, especially from smart meters. Therefore, to properly integrate flexibility into the grid, scalable, and efficient machine learning techniques that can utilize big data from smart meters are critically needed. With the integration of new smart-grid technologies, such as demand-response and distributed energy resources, load forecasting using machine learning is becoming increasingly important at various levels of distribution networks. Here, accurate forecasts are often needed as input for many applications including grid management, storage optimization, peer-to-peer trading, demand-response, and related tasks.
In this thesis, we present three key contributions, that can be utilized in the scope of demand response programmes: (i) First, we propose a global modeling framework for load forecasting in distribution networks that, unlike its local competitors, relies on a single global model to generate forecasts for a large number of loads. The global nature of the framework, significantly reduces the computational burden typically required when training multiple local forecasting models, efficiently exploits the cross-series information shared among different loads, and facilitates forecasts even when historical data for a load is missing or the behavior of a load evolves over time. To further improve the performance of the proposed framework, an unsupervised localization mechanism is proposed. (ii) Our second contribution explores whether the previously proposed global modeling approach can improve the performance over commonly used baseline load estimation models that are all based on local modeling. We propose a global baseline load estimation model that utilizes bi-directional sequence modeling. The biggest advantage of the proposed model is its ability to optimally learn the importance of past and future sequences around demand response events to predict the baseline. (iii) In the third contribution we focus on forecasting the maximal demand flexibility of heat pumps, which is highly related to the Non-Intrusive Load Monitoring task, where the task is to estimate the output signal based on the input signal. However, existing approaches have proven impractical as they depend on high-frequency data and solely target the combined load of individual consumers. As a solution, we introduce a comprehensive framework for forecasting demand flexibility, that can seamlessly integrate with existing smart meter infrastructure.
|
