Two-Period Two-Layer Electrical Model-Free Voltage Calculation for Active Distribution Network via Improved Broad Learning System

Due to the lack of topology and line parameters, traditional voltage calculation methods based on power flow analysis in low-voltage active distribution networks are difficult to apply. Further, existing electrical model-free voltage calculation methods often exhibit unstable and inaccurate results during regulation, particularly when faced with imbalanced datasets. To address those issues, an electrical model-free voltage calculation method for a low-voltage active distribution network is proposed in this paper, which integrates a two-period two-layer scheme with an improved broad learning system. First, by integrating physical models with data analysis, the calculation process is divided into a) a photovoltaic generation period and b) a non-photovoltaic generation period. Then, they are further decomposed into a base operation layer and a regulation increment layer. Further, by extracting dynamic latent variables from the raw dataset, high-quality data with dynamic features is provided for voltage calculation. Finally, considering attention mechanisms for dynamic temporal feature weights and parameter optimization techniques, an improved broad learning system is used to achieve electrical model-free voltage calculation. The case study demonstrates that our method effectively utilizes historical data from an unregulated low-voltage active distribution network to accurately calculate voltages under regulation, exhibiting high calculation accuracy, short training time, suitability for a small dataset, and high scalability.