Short-term charging load prediction study of electric vehicle charging stations based on K-Medoids clustering and multi-factor optimization decomposition

The strong randomness and high volatility of electric vehicle charging behaviour make the accuracy of short-term charging load prediction at charging stations low. Effective electric vehicle charging station charging load prediction is the key to fully and reasonably increase the utilisation rate of charging piles and improve the charging experience. In order to improve the short-term charging load prediction accuracy of electric vehicle charging stations, a combined model based on K-Medoids clustering and multifactor optimization decomposition prediction, Crested Porcupine Optimizer-Variational Mode Decomposition-Bidirectional Gate Recurrent Unit for short-term charging load prediction of electric vehicle charging stations. The K-Medoids algorithm is used to cluster them to improve the quality of the dataset to be predicted. Adaptive optimisation of variational mode decomposition core parameters is set using crested porcupine optimizer and historical charging load data is decomposed to weaken its non-stationarity. Finally, the decomposed feature matrix is inputted into the bidirectional gate recurrent unit model to achieve the short-term charging load prediction objective. A charging station in the US ANN-DATA public dataset was subjected to real-world arithmetic simulation, and the root-mean-square error and average relative error were reduced by 56.95% and 41.60% on average when comparing with the standalone model, unoptimised model, and optimised combination model. The validity and practicality of the proposed method are verified.