A Digital Twin Approach for Distributed Photovoltaic Group Regulation and Group Control

The combination of distributed PV and the digital twin concept is an effective method to cope with the challenges of large-scale grid-connected PV group control and an important element in promoting the construction of new power systems. This paper proposes a digital twin method for distributed PV group regulation and group control. It intends to support the construction of the twin system from three aspects: PV consistency characterization, twin model reconstruction, and power deduction and prediction. A PV consistency characterization method is proposed based on the K-means algorithm, using Pearson's coefficient weighting to construct a meteorological factor with the voltage sensitivity of the PV grid-connected nodes as an indicator to cluster PVs with similar voltage impacts on the distribution network. A high-quality camera pose estimation method based on the SURF algorithm. A low error sparse point cloud reconstruction method is proposed based on the principle of SFM, and then a point cloud densification scheme is proposed in combination with the principle of MVS to form a high-fidelity PV model reconstruction algorithm based on SFM-MVS. A PV short-term power prediction method is proposed based on LSTM network, which utilizes Pearson's coefficient weighting to process the input light and temperature data, and proposes a residual compensation mechanism and physical significance constraints in two parts to improve the prediction result accuracy. The article concludes with experiments to verify the rationality of cluster division, the superiority of PV model reconstruction, and the power prediction accuracy.