Converting renewable electricity into green ammonia has emerged as a new
solution for cross-sector decarbonization and an important source of
flexibility for power systems. Power-to-Ammonia (PtA) plants are an
integration of various electro-chemical-mechanical processes and
multiple storage facilities. Optimal sizing of these modules is an
important but challenging optimization problem at the engineering stage.
It is a two-stage decision-making program, involving both discrete and
continuous variables at two stages, and requiring a large number of
scenarios to accurately capture the stochasticity in renewable energy
generation. We formulate this problem as a general stochastic
mixed-integer programming (SMIP). We use the Benders Dual Decomposition
(BDD) method as an algorithmic framework for this problem, which enables
scenario decomposition and computation parallelization. To reduce the
duality gap of the BDD method, we derive a new type of optimality cut,
the Strengthened Lagrangian Cut (SLC), based on dissimilarity random
scenario bundling. We prove that the SLC is tighter than the Lagrangian
cut, and the resulting master problem provides tighter lower bound and
high-quality feasible solutions for upper bound calculations. We develop
an Improved Benders Dual Decomposition (IBDD) method based on SLCs.
Numerical results show that the traditional BDD method can produce
acceptable results for the PV-powered PtA case but not for the
wind-powered PtA case. In comparison, the IBDD method produces
satisfactory results for both cases and outperforms the BDD method. The
IBDD method is demonstrated to be capable of solving large-scale SMIPs
(containing more than 95,000 mixed-integer variables and 141,000
constraints in the extensive form) by achieving low relative gaps
(< 1%) within manageable computational time (<
1200s) on a standard Laptop.