A model-free distributed control scheme that implements active voltage control in low voltage distribution network (LVDN) is proposed. By solving an 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. A detailed mathematical framework is given, accompanied by a discussion on the different entries of uncertainties. The proposed control scheme uses a broadcast signal to indicate the probability distribution in mean field theory, and to streamline the demand on Fokker-Planck-Kolmogorov PDE or Mc-Kean Vlasov SDE, which relieves the computational burden. A realistic semiurban distribution network is modified as the study case, with a benchmark of centralized ACOPF to study the performance of the proposed approach. Moreover, two special cases including communication latency and packet loss are given as well, in order to test the robustness of the proposed approach. The results prove that the proposed approach is able to deliver good approximation to the optimal control with uncertainty in a model-free and distributed manner. Index Terms-Active distribution networks, Distributed optimization, Mean field theory, Hamilton-Jacobi-Bellman equation, Model-free, Voltage control.
I. INTRODUCTIONW ITH the development of distributed energy resources (DERs), renewable penetration increases not only in medium voltage distribution network, but also low voltage distribution network (LVDN). Along with this trend, conventional passive distribution networks are being transformed into active distribution network (ADN) [1]. Meanwhile, shapeable/deferrable loads spring up in LVDNs, which enlarges the applicability of ADN. Nevertheless, the implementation of ADN is the synonym of a dramatic increase in the amount of smart meters, sensors, and control devices, via bidirectional communication networks. On one hand, high burdens on both communication and computation makes the infrastructure and operation expensive; on the other hand, the rising demand in data collection is being more and more discordant with the awakening public concerns on the privacy protection of data. In this paper, the term users denotes all the DERs connected to LVDNs, which includes but is not limited to photovoltaics (PV), residential wind turbines, smart households and so on. Therefore, the main challenge to implement active controls B. Wei and G. Deconinck are with the