Multi-objective Agent Model for XGBoost Controlled VFTO Simulator Based on Bayesian Optimization Algorithm

Abstract

VFTO generated from GIS may cause interference and damage to terminal equipment on the primary and secondary sides, but experimental measures to evaluate the affected equipment and electromagnetic effects of the equipment through actual measurements are quite limited, so it is desired to build a VFTO simulator in the laboratory for experimental research and laboratory testing. Most of the current researches are conducted by building a true-type experimental platform for VFTO simulation, but this method has the disadvantages of large footprint, high cost and the scope of application can only simulate VFTO in a certain situation. Therefore, this paper establishes a broadband circuit model considering the pulse generator and the frequency domain equivalent transmission line model, and introduces the machine learning algorithm XGBoost to fit the agent model of the functional relationship between the design target and the structural parameters by establishing the sample data of the input circuit parameters and the output frequency waveform characteristics of the simulator, which is capable of dynamically adjusting the input parameters based on the waveform characteristics according to the changes of the target waveform, and ensures that a controllable waveform is obtained accurately in the laboratory the accuracy of obtaining controllable waveforms, thus realizing the output of controllable analog VFTO waveforms by changing the parameters of circuit components. Finally, the Bayesian optimization process is used for the hyper-parameter optimization of the XGBoost model, which has a fast convergence speed, and the characteristics of the output model are compared and analyzed, which verifies the feasibility and superiority of the proposed agent model as well as the multi-objective optimization method.