New power systems are integrating a wide range of large-scale harmonic sources. These sources are diverse

dispersed

and exhibit time-varying characteristics. The harmonic pollution of the power system is aggravated and presents features of strong coupling and strong randomness. In order to ensure high quality power supply

it is urgent to accurately determine the harmonic sources and clarify the dominant influence under the interaction of large-scale harmonic disturbances. First

the process of multi-frequency interaction after the grid connection of nonlinear devices is analyzed

and the idea of harmonic contribution evaluation considering the influence of harmonic coupling is proposed. Then

the degree of harmonic impacts for different orders is analyzed by using the harmonic coupling model

and the decoupled harmonic contribution evaluation index is proposed. On this basis

considering the harmonic time-variant and inter-frequency multicollinearity

kernel principal component analysis and improved Gath-Geva fuzzy clustering method are used to realize the multivariate time-sequence adaptive segmentation

which reduces the dimensionality of data analysis and improves the accuracy of modeling and evaluation. Based on the simulation and measured data

the necessity of considering multi-frequency coupling is proved

and the validity of the proposed index is demonstrated. The proposal combines the dynamic time-varying characteristics of harmonic to identify the dominant disturbance frequency

which will guide targeted governance.