EV Load Forecasting Method Based on Data Reconstruction of GRU-WGAN and BiLSTM Network

With the popularity of electric vehicles, proper prediction of the charging load of electric vehicles is of great importance in order to ensure the reliable functioning of the electrical grid. However, EV load data will inevitably have missing and abnormal problems during the collection and transmission process, which hinders the improvement of load forecasting accuracy. In this paper, we propose a load forecasting approach based on GRU-WGA-BiLSTM. First, modelling of timing data using gated recurrent unit (GRU) networks to capture the complex dependencies in the charging load data. We employ the GAN to reconstruct the dataset. Then, we utilize the adversarial training mechanism to enhance both the authenticity and quality of the reconstructed data. we apply the Generative Adversarial Network to reconstruct the data and use the adversarial training mechanism to improve the authenticity and quality of the reconstructed data. The reconstructed data is used as input to the BiLSTM network for predicting the charging load. Experimental findings show that the proposed model is superior in handling missing data and predicting load. In comparison to conventional interpolation techniques and forecasting models, the data integrity and prediction accuracy are significantly improved.