A hybrid method for structural-parametric identification of dynamic systems

In various fields of research and production activities the problem of structural-parametric identification of dynamic systems is very important. This task consists in constructing a mathematical model of the system based on the experimental data of observation of its behavior. The obtained model can be used both for investigation of system properties and for system control. Evolutionary algorithms, especially genetic programming, have found great popularity in building such models. The result of identification by genetic programming is represented in a symbolic form, which facilitates subsequent analysis and control. However, evolutionary algorithm are inherently stochastic algorithms, which means that they rely on random processes, which often leads to a suboptimal solution. Improving the efficiency of identification of dynamic systems by the genetic programming method is necessary. This paper presents GP – SINDy (genetic programming with sparse identification), a new hybrid method for identification of dynamic systems using genetic programming and sparse identification. In the proposed method, the identification process is divided into two stages: first, genetic programming is applied to determine the model structure, and then sparse identification is used to determine the corresponding optimal parameters. A model in the form of differential equations is constructed on the basis of observations. The effectiveness of the proposed method is demonstrated on the identification of three dynamic systems. The results show that the GP – SINDy allows finding models witha high accuracy and a low complexity (higher interpretability), which makes it promising for structural-parametric identification of dynamical systems.