Coupling stochastic optimization with agent-based simulation: A framework for efficient power expansion planning under uncertainty

Policymakers today face many, interrelated uncertainties. In addition, they have to strike a balance between efficiency, cost-effectiveness, and overarching social objectives. Addressing these problems requires a coupling of several approaches. Thus, we model the power generation expansion planning (PGEP) problem as a combined simulation-optimization problem. Since agent-based simulations (ABM) are able to effectively represent markets, we formulate the PGEP as a multi-stage multi-scale mixed-integer linear optimization problem, where the results of the ABM are integrated into a stochastic optimization model using affine cuts. First, we propose a double decomposition framework combining Benders decomposition and stochastic dual dynamic programming (SDDP) algorithms to solve the PGEP problem. Second, we couple the stochastic optimization model with an agent-based electricity market simulation (AMIRIS) to evaluate power portfolio decisions from a market perspective. We discuss the process of extracting dual values from agent-based simulations with the goal of calculating optimality cuts for the Benders decomposition, to incorporate the simulation results into the optimization model. In particular, we investigate three coupling strategies connecting the optimization and AMIRIS models. Our results show that integrated simulation-optimization approaches yield superior portfolio decisions using both centralized and decentralized operations. Furthermore, they combine recourse and wait-and-see solutions, enhancing resilience against uncertainties.