Orateur
Description
The Stochastic Dual Dynamic Programming (SDDP) algorithm is widely used to solve multi-stage stochastic problems, such as hydrothermal dispatch in power systems. Due to its iterative nature and the need to handle large volumes of data and multiple future scenarios, SDDP is a computationally intensive method. With the increasing complexity of modern systems and the need to respond to energy market fluctuations and climate variability, computational efficiency in SDDP is a topic of great importance. This study aims to explore different parallelization approaches for the SDDP algorithm to improve its performance in terms of execution time and efficient use of computational resources.
The primary motivation for this research lies in the fact that, despite advancements in processing power, the volume of data and the number of scenarios required for accurate analysis have also grown, creating bottlenecks in the practical application of the algorithm. Thus, investigating parallelization techniques enables the application of SDDP in more complex and dynamic contexts, promoting faster and more reliable responses for energy system operation and planning. The parallelization approaches analyzed include process queue-based parallelization, scenario-based parallelization, and stage-based parallelization. Each approach has specific characteristics that may be advantageous depending on the problem structure and available hardware architecture. All approaches will be compared against serial runs, which use only a single computer core.
For performance analysis, prototypes of the approaches were implemented in a controlled environment using a representative set of hydrothermal dispatch problems. The results on the stability, speed, and reliability of each parallelization scheme will be reported in this study.
The choice of the most suitable parallelization technique depends on several factors, including the number of scenarios, the length of the time horizon, the available hardware architecture, and the memory requirements of the problem. Based on the results, a hybrid approach is recommended for large-scale and complex problems, where scenario-based parallelization and subproblem-based parallelization are combined to maximize efficient use of computational resources. This strategy can provide a robust and efficient solution for hydrothermal dispatch and other applications requiring sequential optimization under uncertainty.
In conclusion, studying different parallelization methods for SDDP is essential to enable its application in large-scale energy systems, where agility and decision accuracy are crucial. This research contributes to advancing performance optimization methods that can be applied not only in the energy sector but also in other areas.
