Data reconstruction-enabled robust hybrid models for intraday electricity price forecasting against data contamination attacks

Accurate electricity price forecasting is crucial for optimizing bidding strategies and managing financial risks in deregulated energy markets. However, forecasting is challenging due to the high volatility, abrupt price spikes, and complex seasonal patterns inherent in electricity prices. Moreover, cyber-attacks, such as false data injection (FDI) and fast gradient sign method (FGSM) attacks, can further degrade forecasting accuracy, posing significant risks to market operations. To address these challenges, this study proposes a hybrid model, which decomposes input data using variational mode decomposition (VMD), extracts relevant features through convolutional neural networks (CNN), and captures temporal dependencies using gated recurrent units (GRU). A comparative analysis demonstrates that the proposed model outperforms existing methods in both univariate and multivariate short-term price forecasting. For the Ontario intraday electricity market, the model achieves mean absolute error (MAE) and root mean square error (RMSE) of 1.1618 and 1.6170 for multivariate forecasting, and 0.2764 and 0.4581 for univariate forecasting, respectively. To enhance resilience against cyber-attacks, an autoencoder (AE) is integrated with the VMD-CNN-GRU model to reconstruct compromised data, ensuring reliable forecasts under adversarial conditions. The results indicate that the proposed AE-VMD-CNN-GRU framework not only improves forecasting accuracy but also strengthens system robustness, offering electricity market operators a practical tool for informed decision-making and effective energy management.