Accurately predicting electricity loads is vital for optimizing energy usage plans of ports. However, terminal electricity load data is characterized by high noise, significant fluctuations, and complexity, posing challenges for precise load forecasting. Therefore, this study proposes a data-driven approach to thoroughly analyze terminal electricity load data and designs an anomaly replacement method based on Generative Adversarial Networks (GANs) to address anomalies in port electricity load data, thereby enhancing energy utilization efficiency. Firstly, using spectral residual analysis to detect anomalies in port electricity load data and proposing the Positional Fourier Transform Bi-directional and Conditional Generative Adversarial Net (PF-BiCGAN) method based on GANs for anomaly replacement. Secondly, to generate more accurate synthetic data, PF-BiCGAN introduces a positional encoder and frequency domain enhancement module. The positional encoder helps random noise acquire sequential information to generate synthetic data that mimics the distribution of original data, and the frequency domain enhancement module improves the discriminator’s ability to distinguish between real and synthetic data. Extensive experimental analyses on real port electricity load data in a certain domestic port validate the effectiveness of the proposed methods. Particularly in anomaly handling, compared with other data generation methods, our approach demonstrates significant advantages. Additionally, comparative experiments in electricity load forecasting further confirm the model’s superiority in predicting energy consumption.