Abstract
Offline data-driven optimization does not allow to sample new data during the optimization, making it hard to verify the solution and update the surrogates. One additional challenge is to select appropriate solutions for final implementation, in particular in multi- or many-objective optimization. Nevertheless, this does not necessarily mean that no surrogate management is needed in offline data-driven evolutionary optimization. In this chapter, we at first present a big data driven offline optimization algorithm that adaptively clusters the data to reduce the computation time for trauma systems optimization. Then we describe three model management strategies for offline small data-driven optimization. The first one uses a low-order polynomial that captures the global fitness landscape, which is then used for generating synthetic data for updating a local surrogate. The second management strategy adopts a selective ensemble consisting of a subset of base learners chosen according to the search process. The third strategy builds a randomly sampled subsystem of the original system as the global model, and transfers its knowledge to a local surrogate. In addition, a method for selecting reliable non-dominated solutions for implementation is proposed. Finally, we present an offline data-driven evolutionary algorithm for dynamic optimization, in which a data stream ensemble is adopted and optimized using a multi-tasking evolutionary optimization algorithm for transferring knowledge from previous environments to the present environment.