A global MINLP approach to symbolic regression

摘要: Symbolic regression methods generate expression trees that simultaneously define the functional form of a model and parameter values. As result, problem can search many nonlinear forms using only specification simple mathematical operators such as addition, subtraction, multiplication, division, among others. Currently, state-of-the-art symbolic leverage genetic algorithms adaptive programming techniques. Genetic lack optimality certifications are typically stochastic in nature. In contrast, we propose an optimization formulation for rigorous deterministic problem. We present mixed-integer (MINLP) to solve well several alternative models eliminate redundancies symmetries. demonstrate this technique array experiments based upon literature instances. then use set 24 MINLPs from compare performance five local global MINLP solvers. Finally, larger instances portfolio provides effective solution mechanism problems size addressed literature.