Abstract:

To address the problem of the traditional single-driven modeling methods being difficult to accurately characterize the complex dynamic characteristics of photovoltaic (PV) hydrogen production stations under different operating states, an equivalent modeling method for large-scale photovoltaic hydrogen production stations driven by physics-data collaboration is proposed. Firstly, key factors that can characterize the station's dynamic characteristics are extracted based on physical mechanism analysis. Secondly, a sparse autoencoder is used for dimensionality reduction of the key factor data; under single-phase fault, the data dimension of a single PV unit can be reduced to 3 dimensions with an information retention rate as high as 92.2%. Then, the spectral clustering algorithm is applied to cluster the dimensionality-reduced data features to achieve the equivalent modeling of the entire station. Finally, the effectiveness and accuracy of the proposed equivalent modeling method are verified through single-phase and three-phase fault simulations on the PSCAD/EMTDC platform. Quantitative comparison results indicate that under single and three-phase fault, the fitting accuracy of the power response curve of the proposed equivalent model is superior to the traditional physical-based grouping method and the pure data-driven method.

Key words: physics-data collaborative driven, photovoltaic hydrogen production station, equivalent modeling, sparse autoencoder, spectral clustering