Centralizētās siltumapgādes pieprasījuma prognozēšana, izmantojot mašīnmācīšanos, lai optimizētu katlumājas procesus
Author
Ciekurs, Mārtiņš
Co-author
Latvijas Universitāte. Datorikas fakultāte
Advisor
Cinks, Ronalds
Date
2023Metadata
Show full item recordAbstract
Energoefektivitātes un vides ilgtspējas pieaugošā pieprasījuma dēļ arvien lielāka nozīme tiek pievērsta siltuma pieprasījuma prognozēšanai siltumapgādes sektorā. Precīza prognozēšana var palīdzēt optimizēt darbību, samazināt siltuma zudumus, uzlabot budžeta un uzturēšanas plānošanu, kā arī samazināt siltumnīcefekta gāzu emisijas. Šī bakalaura darba mērķis ir izstrādāt mašīnmācīšanās modeli precīza siltuma pieprasījuma prognozēšanai, īpaši pievēršoties Latvijas kontekstam. Bakalaura darbs ir rakstīts angļu valodā. Šī darba risināmā problēma ir siltuma piegādes neefektīva pārvaldība, ko izraisa neprecīzas siltuma pieprasījuma prognozes, kas noved pie paaugstinātām izmaksām un liekiem izmešiem. Pētījuma mērķis ir izpētīt mašīnmācīšanās tehniku potenciālu šīs problēmas risināšanā un izstrādāt prognozēšanas modeli, kas viegli integrējams esošajos lēmumu pieņemšanas rīkos, piemēram, Power BI. Pētījumā tika atlasīti un apkopoti dažādi dati, tostarp laika apstākļu dati, dati par laiku, dienām un brīvdienām un vēsturiskie ražošanas dati. Dati tika apstrādāti un sagatavoti piemērotā formātā datoram. Dažādi mašīnmācīšanās modeļi tika novērtēti un optimizēti, lai paredzētu nepieciešamo siltuma jaudu. Pētījuma rezultāti parāda, ka apkures sezonā šāds risinājums ir precīzs, tomēr ne apkures sezonā nav piemērots. Atslēgvārdi: siltuma pieprasījuma prognozēšana, mašīnmācīšanās, siltuma pieprasījums, enerģijas efektivitāte The increasing demand for efficient planning, energy efficiency and environmental sustainability has highlighted the importance of heat load forecasting in the district heating. Accurate forecasting can help optimize operations in district heating. This bachelor thesis aims to develop a machine learning model for accurate heat demand forecasting, focusing on the Latvian context. The problem addressed is the inefficient management of heat supply due to a lack of accurate heat load forecasts, resulting in increased costs and environmental pollution. The study aims to explore the potential of machine learning techniques in addressing this issue and develop a forecasting model that can be easily integrated into existing decision-making tools, such as Power BI. The research method used for this thesis involves selecting and gathering relevant data, including weather, temporal, and historical production data. The data was preprocessed: cleaned, outliers processed, features set, and data normalized. Various machine learning models were evaluated and optimized. The study results show that the XGBoost and Random Forest Regression models provided good forecasts for the heating season (October to April), while the results for the non-heating season were not very good. The model was deployed to a web server, with the results visualized in an easy-to-understand chart. The system can be easily integrated into Power BI through an API or accessed directly on a website. The benefits of the developed forecasting model include reduced heating losses, lower greenhouse gas emissions, improved financial planning capability, and enhanced maintenance planning capability. These benefits can help companies in the heating industry increase their operational efficiency, reduce costs, and contribute to a more sustainable future. However, the study has limitations, such as the need for a larger and more qualitative dataset to improve the model's performance. Future research should focus on expanding the dataset and exploring additional machine learning models to enhance the accuracy and applicability of heat demand forecasting and generalizing the model to more boilerhouses. In conclusion, this bachelor thesis demonstrates the potential of machine learning techniques in heat demand forecasting and provides valuable insights into improving the efficiency and sustainability of district heating systems. The success of the XGBoost and Random Forest Regression models highlights the benefits of data-driven approaches in addressing the challenges faced by the heating industry.