A wide range of factors affect the electricity market in different ways, including fuel prices, weather patterns, government policies, and consumer preferences. Traditional techniques, on the other hand, frequently struggle to extract underlying patterns effectively due to feature overlap, the fragility of descriptive identifiers, and the lack of a full holistic perspective. This study tackles these restrictions by using canonical correlation analysis to generate linear combinations of feature sets, allowing for the identification of strong determinants of power spot prices. This method enhances forecasting accuracy by incorporating multi-view feature extraction and multivariate analysis, providing an advantage in evaluating small changes in volatile markets. This method improves predictive accuracy and lowers forecasting mistakes by including multi-view feature extraction and multivariate analysis in the forecasting model. Thus, a significant advantage in price forecasting is gained by evaluating small changes in volatile markets as part of a whole. The proposed approach reduces underfitting issues in existing models that suffer from high dimensionality by using smooth and discriminative representations. Also, incurring long computational time attributed to matrix multiplication operations is effectively reduced. The average root mean square error of the proposed model is reduced by 14–20%. Multiple views in the embedding space also contribute to the prediction of asynchronous price movements under global and local information.
Cite this article as: R. G. Birdal, “Multi-perspective prediction of day-ahead electricity spot prices via canonical cross-covariance analysis,” Electrica, 25, 0211, 2025. doi: 10.5152/electrica.2025.24211