Efficient GPU-parallelization of batch plants design using metaheuristics with parameter tuning

https://doi.org/10.1016/j.jpdc.2021.03.012Get rights and content

Highlights

  • Our use case is optimal batch plant design.

  • Our algorithm is based on Ant Colony Optimization (ACO) and Simulated Annealing (SA).

  • We propose and implement on GPU three approaches to parameter tuning of ACO.

  • We confirm the efficiency of our approach by experiments with real-world examples.

Abstract

We address a practice-relevant optimization problem: optimizing multi-product batch plants, with a real-world use case study – optimal design of chemical-engineering systems. Our contribution is a novel approach to parallelizing this optimization problem on GPU (Graphics Processing Units) by combining two metaheuristics – Simulated Annealing (SA) and Ant Colony Optimization (ACO). We improve the implementation performance by tuning particular parameters of the ACO metaheuristic. Our tuning approach improves on the previous methods in two respects: (1) we do not have to rely on additional mechanisms like fuzzy logic or algorithms for online tuning; and (2) we use the high computation performance of GPU to speedup the tuning process. By parallelizing the tuning process on modern GPUs, we allow the user to experiment with large volumes of input data and find the optimal values of the ACO parameters in feasible time. Our experiments on NVIDIA GPU show the efficiency of our approach to parameter tuning for the ACO metaheuristic.

Section snippets

Motivation and related work

Optimal design of multiproduct batch plants is a practice-relevant optimization problem. In this paper, we conduct a case study of designing Chemical-Engineering Systems (CES), which consist of a set of equipment units (tanks, filters, reactors, etc.) manufacturing a variety of products. The problem is to determine, for the given input (production horizon, product demand, etc.), the number and main sizes of technological units to achieve the lowest possible capital and operating cost while

The batch plant optimization: Problem formulation

We provide here a rather intuitive description of the batch plant optimization problem for a particular use case of CES; a detailed formulation is given in [6]. A CES consists of equipment units organized in I processing stages: the ith stage has set Xi of equipment units, while Ji is the amount of units’ variants in Xi. We write Ωe,e=1,E¯ for a CES variant, where E=i=1IJi is the number of system variants.

In Fig. 1 a simplified CES with 4 stages is shown, where each stage has two devices (Ji=2

A parallel metaheuristic approach

Our approach to accelerating the solution of the CES optimization problem is to use metaheuristics and to develop a fast implementation of the metaheuristics using highly-parallel architectures, in particular Graphics Processing Units (GPU).

A heuristic algorithm for an optimization problem considers the most likely, rather then all possible states of the problem. Specific heuristics are usually problem-dependent. A metaheuristic is an algorithmic pattern for finding high-quality solutions for

Parameter tuning for the ACO metaheuristic

The goal in this section is to improve the parallel ACO implementation by using the additional performance potential via tuning particular parameters of ACO. We employ offline parameter tuning according to the classification of [21]. The tuning task is to configure the ACO parameters, such that the CSP is solved as fast as possible. Our idea of improvement by tuning is to find the intervals of the parameter values in which the feasible solutions of the optimization problem (i.e., solutions that

Experimental results

Our experimental setup consists of: (1) a CPU: Intel Xeon 2.3 GHz with 12 hyper-threaded cores, and 192 GB of RAM, (2) a GPU: NVIDIA V100 T with 5120 CUDA cores at 1.53 GHz, and 16 GB of global memory. We employ GNU C++ 6.4.0 and CUDA 10.0.

Our case study is to design a CES of 16 stages with 11 to 20 device variants per stage (i.e., from 11161017 to 20161021 variants). This size is significantly larger than was possible without parameter tuning [5]. We call the ACO algorithm 100 times for each

Conclusion

This paper makes the following contributions to the state of the art in the batch plant design optimization. First, we design and evaluate a parallel combined implementation of metaheuristic-based optimization. Second, we develop three tuning approaches for the parameters of the parallel Ant Colonies Optimization (ACO) metaheuristic. The advantages of our approach compared to related work are as follows: (1) using the high computing power of GPU for tuning, and (2) avoiding any additional

CRediT authorship contribution statement

Andrey Borisenko: Conceptualization, Methodology, Software, Visualization, Investigation, Writing - original draft. Sergei Gorlatch: Supervision, Project administration, Validation, Writing - reviewing and editing, Visualization.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgments

Thanks to the anonymous reviewers for their useful comments, and to NVIDIA for the provided hardware used in the experiments. Andrey Borisenko was partially supported by DAAD (German Academic Exchange Service) and the Russian Ministry of Education and Science within the “Mikhail Lomonosov II”-Program. Sergei Gorlatch was partially supported by the HPC2SE project of BMBF (German Ministry of Education and Research) at WWU Muenster .

Andrey Borisenko Last Name, First Name: Borisenko Andrey Scientific Degrees and Titles: Cand. Sci. (Eng.), Associate Professor University, Department, Position: Tambov State Technical University, Computer-Integrated Systems in Mechanical Engineering, Associate Professor Business Address: Sovetskaya Str. 106, 392000 Tambov, Russian Federation E-mail: [email protected] Homepage: http://www.gaps.tstu.ru/homepages/borisenko/index_en.html ORCID iD: 0000-0001-9315-6167

References (24)

  • ChenC.-C. et al.

    Enhanced ant colony optimization with dynamic mutation and ad hoc initialization for improving the design of TSK-type fuzzy system

    Comput. Intell. Neurosci.

    (2018)
  • DorigoM. et al.

    Ant colony optimization: Overview and recent advances

  • Cited by (0)

    Andrey Borisenko Last Name, First Name: Borisenko Andrey Scientific Degrees and Titles: Cand. Sci. (Eng.), Associate Professor University, Department, Position: Tambov State Technical University, Computer-Integrated Systems in Mechanical Engineering, Associate Professor Business Address: Sovetskaya Str. 106, 392000 Tambov, Russian Federation E-mail: [email protected] Homepage: http://www.gaps.tstu.ru/homepages/borisenko/index_en.html ORCID iD: 0000-0001-9315-6167

    Sergei Gorlatch Last Name, First Name: Gorlatch Sergei Scientific Degrees and Titles: Dr. habil., Professor University, Department, Position: University of Muenster, Parallel and Distributed Systems, Head of Department Business Address: Einsteinstr. 62, D-48149 Muenster, Germany E-mail: [email protected] Homepage: http://www.uni-muenster.de/PVS/mitarbeiter/gorlatch.html ORCID iD: 0000-0003-3857-9380

    View full text