New parallel computing framework for chemical process simulation

A novel parallel computing framework for chemical process simulation has been proposed by researchers from the East China University of Science and Technology and the University of Sheffield. This framework, known as the process-simulation parallel computing framework (PSPCF), aims to address the computational challenges faced by conventional sequential-modular (SM) based process-simulation techniques, particularly in large-scale design and optimization tasks.

The PSPCF introduces a groundbreaking approach to process simulation by formulating simulation problems as task graphs and utilizing Taskflow, an advanced task graph computing system, for hierarchical parallel scheduling and execution of unit operation tasks. This framework also integrates an advanced work-stealing scheme to automatically balance thread resources with the demanding workload of unit operation tasks.

The SM approach has long been the preferred method for process simulation due to its intuitive alignment with real chemical processes, ease of programming and maintenance, and robust convergence stability. However, it faces significant challenges in terms of computational time and resource consumption, especially for large-scale problems. The PSPCF addresses these challenges by parallelizing SM-based chemical process simulations at the unit operation level, offering more coarse-grained parallelism than most published equation-oriented (EO)-based parallel computing strategies.

The framework’s ability to handle processes with recycles is particularly noteworthy. It achieves this through a hierarchical task graph generation system that efficiently parallelizes processes with recycles using conditional and composable tasking models. This allows for the iterative execution of recycle processes, which is a common scenario in chemical process simulations.

The performance of the PSPCF was evaluated using two case studies: a simpler parallel column process and a more complex cracked gas separation process. The results showed significant time savings, with over 60% reduction in processing time for the simpler process and a 35%–40% speed-up for the more complex separation process. These improvements highlight the potential of the PSPCF to enhance the efficiency of chemical process simulations.

The PSPCF’s architecture includes a main graph setting system (MGSS) and a recycle subgraph generation system (RSGS). The MGSS constructs a hierarchical executable task graph from a simulated process flowsheet, while the RSGS handles the iterative recycle procedure. This hierarchical structure enables layered parallelism in process-simulation calculation, allowing for efficient execution of both out-of-block and in-block tasks.

The researchers also highlighted the importance of load balancing in achieving optimal performance. The work-stealing mechanism integrated into the PSPCF dynamically adjusts the number of worker threads based on the task load, ensuring efficient utilization of computing resources. This adaptive scheduling scheme is crucial for handling the complex dependency structures often found in chemical process simulations.

The PSPCF represents a significant advancement in the field of chemical process simulation. By leveraging parallel computing and advanced task graph techniques, it offers a more efficient and scalable solution for large-scale process simulations. Future work will focus on integrating artificial intelligence models to optimize thread allocation and developing advanced scheduling algorithms that can scale to both graphics processing units (GPUs) and central processing units (CPUs), further enhancing the framework’s capabilities.

The paper “A Hierarchical Task Graph Parallel Computing Framework for Chemical Process Simulation,” is authored by Shifeng Qu, Shaoyi Yang, Wenli Du, Zhaoyang Duan, Feng Qian, Meihong Wang. Full text of the open access paper: https://doi.org/10.1016/j.eng.2024.06.019. For more information about Engineering, visit the website at https://www.sciencedirect.com/journal/engineering.