Data-Driven Sizing Planning of Renewable Distributed Generation in Distribution Networks with Optimality Guarantee

In this paper, we study the optimal sizing planning of renewable distributed generation (RDG) in distribution networks to minimize the long-term cost, including the investment cost, maintenance cost, and operating cost. In particular, the operating cost itself is optimized by solving an optimal power flow (OPF) problem at each time t based on uncertain time-varying RDG output and load demand. As a result, the sizing planning problem is a bilevel stochastic programming problem, which is hard to solve. Instead of resorting to conventional meta-heuristic algorithms, this paper first proposes a novel data-driven approach based on the philosophy of online convex optimization to solve the problem with drastically lower complexity. As a key step to facilitate the algorithm, we derive a closed-form expression to iteratively update the sizing solution upon drawing each data sample. With sufficient data samples, the proposed algorithm guarantees to converge to the global optimal solution regardless of the underlying probabilistic distribution of RDG output and load demand. Numerical results on the IEEE 13-bus test feeder, the IEEE 33-bus test feeder, and the Southern California Edison (SCE) 56-bus feeder show that our data-driven method drastically outperforms the other methods in terms of both the solution optimality and computational complexity.