Firefly Algorithm based Semi-Supervised Learning with Transformer Method for Shore Power Load Forecasting

Load forecasting of shore power (SP) plays an important role in the decision-making electrical grid due to docked ships are necessary plug into grid. However, obtaining a large amount labeled data on is time-consuming, presenting challenge for Shore Power Forecasting. Additionally, multiple raw information entries could lead feature redundancy. To address these issues, we proposed novel three-stage load method which includes attributive selection, semi-supervised learning (SSL) mean distribution prediction, and transformer-based model variance prediction. Firstly, Firefly Algorithm (FA) adopted extract representative attribute features deal with Next, selected set label divided two parts: few data. And propose Π-model-based SSL predict distribution. Finally, Our takes account all historical each ship context learning. Further, consider that would also affect so latent Π-model served as initial condition concatenates We evaluated our using 328 from various berth at Zhenjiang Port power, totaling approximately 21,521 hours. The experiments prove accuracy efficiency model, producing promising results.