Promotor / Supervisor
Supervisor: Geert Deconinck
Co-supervisor: Zhifeng Qiu
Samenvatting van het onderzoek / Summary of Research
Introduction / Objective
With the development of distributed energy resources, the low voltage distribution networks have to manage more and more multifarious users. This development trend has created quite a few opportunities to implement new smart grid application to improve energy efficiency; meanwhile, it calls for innovative approaches to implement active control to the ``last mile'' of the power system.
To help the modern low voltage distribution network face the new challenges, this dissertation proposed three distributed algorithms to optimize the users' active power consumption in a given low voltage distribution network while ensuring the voltage constraints and additional objectives. The first algorithm leverages statistic methods and game theory to make individual users in LVDN achieve local optimum autonomously. A variant of the log-linear trial and error learning process is applied in a novel "suggest-convince" mechanism, which guarantees the convergence to Nash Equilibrium (NE) and uses player compatible relations to form a specific equilibrium. The second algorithm employs Bernoulli trials to imitate the searching process in the classical gradient descent approach. The player-compatible relationship is employed to play the role of gradients to indicate the direction of the search. Working in a model-free manner without relying on iterations, this algorithm offers an approximate optimization to minimize the accumulated compensation of reshaping/deferring the shapeable energy resources in a given LVDN while respecting the system constraints. The third algorithm uses the mean-filed theory. By solving individual Hamilton-Jacobi-Bellman-Flemming function with public information, users can compute a good approximation to their optimal control trajectory and take uncertainties into account in a distributed manner. Besides the distributed optimization algorithms, to address the dilemma where a user is not able/willing to follow the control signal, a real-time peer to peer flexibility trading scheme is proposed.
Results & Conclusions
A 103 nodes test network based on a realistic Belgian semiurban distribution network is used for validation for all the proposed algorithms. Different profiles and special cases of various failures are used for test purpose depends on the specific algorithm characteristics. Moreover, a classical AC optimal power flow algorithm is implemented for further validated and benchmark. The results of the case studies prove the effectiveness and robustness of the proposed algorithms. A conceptual comparison of the three proposed algorithms is given in the conclusion section of this dissertation. Their corresponding advantages and disadvantages are discussed in control accuracy, communication demand, computational burden, robustness, and scalability. For each algorithm, a typical application scenario is given and discussed respectively. Besides, the compatibility with other optimization approaches of the proposed flexibility trading scheme is discussed case by case.