Load Forecasting Model for Battery Swapping Stations Based on Fuzzy Clustering-Markov Chain

Abstract

Battery-swap stations significantly reduce the replenishment time for electric vehicles and offer substantial potential for grid regulation. An accurate load prediction model is essential for their effective participation in grid auxiliary services. To address the stochastic nature of user power exchange demand, this study proposes a load prediction model for swap stations based on fuzzy clustering and a Markov chain. First, the Poisson distribution is employed to predict the number of EVs requiring power exchange at each time step, establishing demand constraints. Second, an adaptive fuzzy C-means clustering algorithm is utilized to dynamically partition battery clusters in the swap stations based on their state of charge, eliminating the subjectivity of manual partitioning. Finally, a Markov chain is used to model the battery clusters across multiple states, including charging, discharging, and waiting. The proposed demand prediction method and load prediction model are simulated, validated, and compared with the Monte Carlo simulation method. Results demonstrate that the Poisson distribution effectively predicts EV demand, while the load prediction model accurately captures the power states of the swap station during charging and discharging, reducing load prediction volatility.