Generalizability of Machine Learning Algorithms for Modelling and Control of Thermostatically Controlled Loads

作     者：Attila Balint 

作者单位：Universitat Politecnica de Catalunya (UPC) 

学位级别：硕士

导师姓名：Joaquim Rigola

授予年度：2018年

主      题：hot water systems residential buildings modelling energy flexibility oc-cupant behaviour heat pumps control algorithm 

摘      要：With the proliferation of variable energy sources, flexible energy loads will become more and more important to help stabilize the energy grid. Increasing electrification of heating systems means that the thermal inertia of buildings and hot water vessels can provide a widespread, low cost alternative to electrical storage for providing this energy flexibility. In this thesis, I demonstrate the modelling capabilities and generalizability of state of the art machine learning techniques on residential hot water systems, using data from multiple large scale real world case studies. I exhibit that an improved control algorithm with these models is capable to reduce the energy consumption of these hot water systems by up to 30%. I also use these models to quantify the effect of major factors influencing available energy flexibility of residential hot water systems. All the houses considered in the analysis feature the same type of hot water system which eliminates any differences in energy flexibility caused by device characteristics. A number of metrics are used from existing literature to quantify flexibility, and find that ambient conditions, control algo- rithm and occupant behaviour all play significant but very different roles. There are also some key differences in the way these factors influence energy demand and flexibility. The available capacity and recovery periods can differ by as much as two to four times for the same storage vessel, meaning that these differences have to be taken into account during operational planning with flexible loads. We conclude with a discussion on the implications of hot water system modelling and its generalizability, the variations in flexibility they can provide and the controllers that can be adopted to influence it in practice.