Stochastic hydrothermal unit commitment models via stabilized benders decomposition

The high penetration of wind generation has prompted the development of stochastic hydrothermal unit commitment (SHTUC) models, which are more difficult to be solved than their thermal-based counterparts due to hydro generation constraints and inflow uncertainties. For handling the uncertainty, the problem is usually formulated as a two-stage stochastic model (2S-SHTUC), although multistage (MS-SHTUC) formulations have gained increasing attention due to their more realistic assumptions about on-off decisions over the planning horizon. Benders decomposition (BD) is one of the most common methodologies used for solving 2S-SHTUC and MS-SHTUC problems. To overcome the well-known slow convergence of the classical BD when applied to large problems, most authors use accelerating techniques. In this paper, we implement state-of-the-art stabilization methods tailored for speeding up the convergence of the classical BD: local branching and the level regularization. Our experiments are conducted in a real-life SHTUC problem with 11 thermal units, 16 hydro plants, 46 buses and 95 lines. The results show that 2S and MS-SHTUC can benefit from stabilization. The savings in computing times range from 69 to 95% for the 2S-SHTUC model and from 77 to 89% for the MS-SHTUC.

Automated summary

This study implements advanced stabilization methods to speed up the convergence of the classical Benders decomposition (BD) method in solving two-stage stochastic unit commitment (2S-SHTUC) and multi-stage stochastic unit commitment (MS-SHTUC) problems. The experimental results show that stabilization methods can significantly reduce computing times, with savings ranging from 69% to 95% for the 2S-SHTUC model and from 77% to 89% for the MS-SHTUC model.