**6. Conclusions**

In this paper, identification of wind turbine mechanical dynamics is studied under non-excitation condition. Identification performance under different wind scenarios is tested using three types of methods. For VSVP wind turbine, the drive-train subsystem is structurally identifiable under closed-loop condition and direct identification is feasible for the two-mass model. Through MI calculation, nonlinear correlations among identified variables can not only validate whether the linkage of these variables are consistent with the control loop but also reveal the relationship between identified data and identification performance. It can be found that higher correlations among identified variables can yield better identification performance. In contrast, state-space model from optimal identification can reflect the physical meaning of parameters and natural stability of the identified system which is important for advanced control algorithms. In summary, grey-box optimal identification shows its feasibility to identify complex wind turbine dynamics and its grea<sup>t</sup> potential in advanced control design. Additionally, the limitation of the simplified mechanism model to represent complex and practical dynamics should be paid attention. In future, dynamic compensation to the identified simple mechanism model based on machine-learning will be studied. It may balance modeling complexity and difficulty and would be attractive to the application of digital-twin modeling of wind turbines.

**Author Contributions:** The individual contributions of the authors are provided as follows: conceptualization, J.C., L.Y. and Y.H.; methodology, Y.H., C.P. and L.P.; validation, Y.H.; writing—review and editing, Y.H.; visualization, Y.H. and C.P.; supervision, L.P.; funding acquisition, J.C. and L.Y.

**Funding:** This research was funded by 'the research on Intelligent Control Technology of Wind Turbine (Guodian United Power Technology Company Limited), gran<sup>t</sup> number 17001', 'Hebei Provincial Key Research and Development Program, gran<sup>t</sup> number 18214316D'.

**Conflicts of Interest:** The authors declare no conflict of interest.
