As the global energy system evolves towards a low-carbon model, Renewable energy, exemplified by solar and wind power, serves as a critical pillar supporting the advancement of modern energy infrastructure. However, such renewable energy sources have significant temporal and spatial uncertainties and dynamic change characteristics. With the continuous increase of its proportion in the power system, the random disturbance factors on both the source and load sides have grown exponentially, which makes the traditional load forecasting model face the severe test of decreased accuracy when dealing with complex systems with multi-dimensional coupling. Therefore, High-accuracy load predicting is essential for maintaining the stability and cost-effectiveness of modern power infrastructure through proactive resource allocation. An innovative hybrid model is developed in this research to enhance forecasting precision through the integration of CEEMDAN with adaptive noise, VMD decomposition, a modified swarm intelligence optimization technique, and a deep neural network architecture that combines temporal convolutional modules with bidirectional recurrent components. This model first employs a two-stage decomposition approach to break down the raw load data into multiple modal components. Following this, the integrated optimization and deep learning architecture is utilized on specific signal segments to execute autonomous estimation processes. Ultimately, the predicting results derived from decomposed signal components are integrated to reconstruct the comprehensive power load prediction. The effectiveness of the developed short-term load prediction model was assessed through experimental analysis on historical power grid operational data. The proposed model’s performance characteristics were empirically validated by benchmarking against established forecasting techniques across multiple practicality and reliability metrics.
Li et al. (Fri,) studied this question.