Kaukolämpötoiminnan epävarmuuksia hallitseva optimointimalli pitkä- ja lyhytai-kaismuistineuroverkon avulla
Year: 2024
Type: dissertation
Abstract: Although recognized as a key factor in reducing emissions of the heating sector, many district heating systems globally are still dependent on fossil fuels. The future of district heating relies on their decarbonization to maintain it a competitive technology in alignment with climate regulations. The use of renewable energy sources, such as biomass, along with electrification, will inevitably continue to grow in the sector. Their intermittent nature makes future district heating systems highly reliable on flexibility strategies to stabilize operational costs. These include energy storages, demand side management from load forecasting, and optimization. The listed technologies were studied in the field of district heating. Previous studies have highlighted the advantages of artificial intelligence, particularly its subset neural networks, in district heating load forecasting. A forecasting model with a long-term short-term architecture was chosen and trained for this study. To account for uncertainties, a chance-constrained programming model was selected for the optimization. As a result, the model gave hourly district heating load forecast values and an optimized power plant schedule for a forecasting horizon of seven days. The neural network model showcased good performance in capturing complex, non-linear relationships between district heating load and influencing parameters with a mean absolute percentage error of approximately 6 %. The integration of the model with chance-constrained programming allowed for an optimized dispatch of energy sources, minimizing both operational costs and reliance on fossil fuels while accommodating the inherent uncertainties in load forecasting. Utilizing renewable electricity significantly decreased operational costs, which urges for further electrification of district heating systems, and smart scheduling given by the optimization model encourages exploitation of advanced technologies. The error in the load forecast was dependent on weather forecasting errors, hence focus on improving the latter is essential. Next steps include experimenting with ensemble techniques, building an electricity price forecasting model, and exploring network-wide impacts of incorporating flexibility strategies in district heating.
Source: Aaltodoc (Aalto University)
Author: Toivanen, Linnea
Language: English
Subfield: Electrical and Electronic Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7048308834