Dynamic Clustering Method for Distributed Photovoltaic Systems Based on Improved Fuzzy C-means Algorithm

Detailed modeling of large-scale distributed photovoltaic (PV) power system encounters challenges such as significant computing burden, making it critical to establish a clustering equivalent model. However, existing clustering algorithms suffer from drawbacks such as the sensitivity to noise and the tendency to converge on a local optima. To tackle these challenges, this paper proposes a dynamic clustering method based on the improved fuzzy C-means algorithm. Firstly, a comprehensive clustering indicator that effectively characterizes the combined effect of the equivalent electrical distance, the voltage support function and the inverter capacity is introduced. Secondly, the particle swarm optimization algorithm (PSO) is employed to identify the global optimal solution, which serves as the initial clustering center. Following such initialization, the fuzzy C-means algorithm (FCM) is applied to partition the PV units into several clusters. The optimal number of clusters is then determined by maximizing the Calinski-Harabasz Index (CHI). The improved clustering method proposed in this paper reduces the great dependence of clustering performance and initial clustering centers when using the fuzzy C-means algorithm, thus significantly enhancing the clustering accuracy. Simulation results from MATLAB/ Simulink demonstrate the effectiveness of the proposed clustering method and resultant equivalent model for decreasing the relative error in the reactive power emulation from 15.8% with FCM to 2.11% with the improved algorithm, which accurately reflects the dynamic response characteristics of distributed PV system and their diverse voltage support behaviors.