Influence of Hyperparameters in Neural State-Space Models for Data-Driven Black-Box Modelling of Synchronous Motors: A Case Study

This paper presents an empirical study on the influence of network and training settings in data-driven black-box modelling of synchronous motors using neural state-space models. The objective is to provide a foundational insight into neural state-space modelling and reference points for researchers involved in this field. The study examines correlations between hyperparameters in terms of design and training parameters and the resulting model performance, which are difficult to express analytically. Additionally, an in-depth analysis of training times is conducted to identify configurations that achieve comparable model accuracy with reduced computational time. These findings aim to facilitate efficient model development by offering a systematic approach to parameter selection and optimization in black-box modelling for electrical motor applications.