Electric vehicle charging current scenario generation based on generative adversarial network combined with clustering algorithm

The generation of charging current scenario is an important step in the operation and planning of power systems with high electric vehicle (EV) penetrations. With the development of the modeling method, a number of methods based on probabilistic models are applied to generate scenarios. Model-based methods are often difficult to scale or sample. Data-driven technologies use a large number of data to mine the mapping relationships, instead of explicitly specifying a model. In this paper, we proposed a data-driven approach to generate scenarios using generative adversarial networks (GANs), which can learn the distribution of the charging current of EVs and obtain more abundant scenarios. The proposed method is applied to time-series data from the charging current dataset of EVs. Firstly, the K-Means clustering algorithm is used to preprocess the data to divide the distribution of charging current into four areas. Then, aiming to improve the training speed, GANs with gradient penalty (GP) is used for the generation of EV scenarios, which can use the GP term to optimize the Lipschitz limit. Finally, statistical methods are applied to estimate the quality of the generated data. Results show that the proposed method can effectively extend the historical data for the operation and planning of EVs in the future compared with the traditional GANs.

Automated summary

The paper introduces a data-driven approach using generative adversarial networks (GANs) to generate scenarios of electric vehicle (EV) charging, which can learn the distribution of EV charging current and obtain richer scenarios. The method divides the charging current distribution into four areas using K-Means clustering algorithm, and leverages GANs with gradient penalty (GP) for faster training and optimization of the Lipschitz limit. Statistical methods are then used to estimate the quality of the generated data, demonstrating the effectiveness of the proposed method for extending historical data for future EV operation and planning compared to traditional GANs.