626 / 2022-03-30 20:53:02

Digital Twinning of Helical Type Windings Based on FRA Data and GA Optimisation

frequency response analysis,digital twinning,Genetic Algorithm (GA),transformer winding geometry

Purpose/Aim

Accurate geometrical parameterization of a transformer, such as conductor and winding sizes, is critical for accurate modelling of transformers.However, those information are not usually available due to the manufacturer’s confidentiality. The grey-box model provides an approach to estimate those parameters directly from the frequency response analysis (FRA) measurement of the transformer. Meanwhile, considering that obtaining the frequency response analytically would repeat a large number of matrix calculations at each frequency point, which is a time-consuming process in each iteration, it is also not conducive to ruling out fluke results through multiple simulation. This paper proposes a grey box modelling method based on FRA resonance/antiresonance point data, which significantly shortens the computational time while maintaining high accuracy.



Experimental/Modeling methods

In this study, a single winding with helical structure is tested because of its small series capacitance and clearly discernible resonance point, while the effect of iron core and tank are considered. The FRA fingerprint is produced by RLC electrical lumped element model, the elements of which are determined analytically from an artificially assembled transformer winding geometrical dimension. An intelligent optimisation algorithm, Genetic Algorithm (GA), is used as a tool to search and select the best fitting solution according to the fitness assessment between the simulated and reference FRA resonance/antiresonance point data. By taking the advantage of fast simulation speed, transformer geometry parameterasition is achieved and the reliability of the corresponding parameter estimation can be evaluated.



Results/discussion

The FRA resonance/antiresonance points data can be used as a fitness calculation criterion for grey box modelling optimisation. The FRA produced by the final estimated geometry data is in line with the reference. Figure 1 shows the results of the FRA spectrum comparison between one of the simulations and the reference, and the corresponding estimated value are listed in Table 1, with relative errors are at acceptable levels.



Conclusions

A new improved optimisation method is applied to identify the geometrical parameters of a transformer grey box model. The method only used the FRA resonance points data as terminal input instead of the full bandwidth of FRA. Genetic algorithm was used to solve the optimization and solution finding problem. The following achievements were obtained. 1. The fitness calculation between simulation result and reference in GA was only performed at each FRA resonance point, hence the computational time was massively reduced. 2. The sensitivity and reliability of each estimated parameter are discussed by analysing the results of multiple runs.