Dynamic optimization of chemical engineering problems using affinity propagation based estimation of distribution algorithm

Dynamic optimization has attracted much attention for its wide applications in engineering problems. However, it is still a challenge for high nonlinear, multi-dimensional and multimodal problems. Estimation of Distribution Algorithm was proposed in which probabilistic models extracted relevant features of the complex search space and then generated new individuals during optimization. In order to decrease the dependences among control variables in dynamic optimization, affinity propagation was applied to cluster the individuals in evolutionary iterations. In each cluster, the probabilistic density function of Gaussian mixture model refined the promising spaces with high quality solutions and avoided the random combination of different control variables. To evaluate the performance of the new approach, three dynamic optimization problems of chemical process are used as cases comparing with three state-of-the-art global optimization methods. The results showed that the new approach could achieve the best solution in most cases with less computational effort and higher efficiency.