Doctoral Thesis Machine Learning Algorithms for Smart Electricity Markets
The shift towards sustainable electricity systems is one of the grand challenges of the twenty-first century. Decentralized production from renewable sources, electric mobility, and related advances are at odds with traditional power systems where central large-scale generation of electricity follows inelastic consumer demand. Smart Markets and intelligent Information Systems (IS) could alleviate these issues by providing new forms of coordination that leverage real-time consumption information and prices to incentivize behaviors that remain within the grid's operational bounds. However, the best design for these artifacts, and the societal implications of different design choices is largely unclear. This dissertation makes three contributions to the debate. First, we propose and study a design for Brokers, a novel type of IS for autonomous intermediation in retail electricity markets. Second, we propose a probabilistic model for representing customer preferences within intelligent IS, and we study its performance in electricity tariff and other choice tasks. And third, we propose and study Competitive Benchmarking, a novel research method for effective IS artifact design in complex environments like Smart Grids where the social cost of failure is prohibitive. Our results provide guidance on IS design choices for sustainable electricity systems, and they highlight their potential societal positives and negatives.
Autonomous Agents, Competitive Benchmarking, Gaussian Processes, Machine Learning, Preferences, Reinforcement Learning, Retail Electricity Markets
Time frame2013 - 2014
M. Peters, Machine Learning Algorithms for Smart Electricity Markets, Promotor:prof.dr. Wolfgang Ketter, http://hdl.handle.net/1765/77413