DPSIM Modelling: Dynamic Optimization in Large Scale Simulation Models

Although it is well established that dynamically optimal policies should be “closed loop” so that policies take into account changing conditions of a system, it is rare for such optimization to actually be carried out in large-scale simulation models. Computational limitations remain a major barrier to the study of dynamically optimal policies. Since the size of dynamic optimization problems grows approximately geometrically with the state space, this problem will continue to inhibit the identification of dynamically optimal policies for the foreseeable future. In this chapter, we explore in detail the problem of solving dynamic optimization problems for large-scale simulation models and consider methods to work around the computational barriers. We show that a reasonable approach is to solve a small-scale problem to identify an approximate value function that can then be embedded directly in the simulation model to find approximately optimal time-paths. We present and compare two ways to specify the small-scale problem: a traditional “meta-modelling” approach, and a new “direct approach” in which the simulation model is embedded directly in the dynamic optimization algorithm. The methods are employed in a model of the Gulf of Mexico’s red snapper fishery and used to identify the dynamically optimal total allowable catch for the recreational and commercial sectors of the fishery.