A service composition method using improved hybrid teaching learning optimization algorithm in cloud manufacturing

Jun Zeng, Juan Yao, Min Gao, Junhao Wen

Abstract

In Cloud Manufacturing (CMfg), the Service Composition and Optimization Selection (SCOS) problem is critical for managing complex tasks across distributed resources. This paper presents Improved-TC, a hybrid optimization algorithm designed to navigate massive search spaces more effectively than traditional methods. By integrating Teaching-Learning-Based Optimization (TLBO) with Crisscross Optimization (CSO) and a novel one-dimensional search strategy, the proposed method achieves faster convergence and higher stability. Experimental evaluations on a synthetic dataset of 20,000 services demonstrate that Improved-TC consistently finds superior solutions while avoiding the local optima that plague standard industry algorithms.

Introduction: The SCOS Problem

Cloud Manufacturing treats resources and capabilities as on-demand services. When a complex manufacturing task is received, the system must decompose it into subtasks and select the optimal combination of individual services—such as machining, 3D printing, and logistics—to form a composite service. This process, known as Service Composition and Optimization Selection (SCOS), is NP-hard due to the exponential explosion of possible combinations as the number of subtasks and candidate services increases.

Fig. 1: Service composition flowchart illustrating the mapping of subtasks to service sets.

Traditional optimization algorithms often struggle in large-scale CMfg environments, suffering from slow convergence and a tendency to get trapped in local optima—solutions that appear optimal in a limited search area but are inferior to the global best.

Methodology: The Improved-TC Algorithm

The authors propose Improved-TC, which enhances the Hybrid-TC framework through three primary strategies:

  1. Teaching-Learning-Based Optimization (TLBO): This parameter-free approach mimics a classroom environment where a "teacher" (the best current solution) improves "students" (other solutions), who also learn from one another.
  2. Crisscross Optimization (CSO) Crossover: During the teaching phase, horizontal crossover from CSO is incorporated. This allows for fine-grained, dimension-level updates rather than updating entire solutions at once, facilitating a more thorough exploration of the search space.
  3. One-Dimensional Search: To prevent solutions from being dragged down by a single dimension stuck in a local optimum, the algorithm employs a one-dimensional search in the learning phase. This allows specific dimensions to "jump out" of local optima independently.
  4. Skyline Query Initialization: Before optimization, a Skyline query filters out services dominated by others across all Quality of Service (QoS) categories, significantly narrowing the initial search space.
Fig. 4: The framework of the Improved-TC algorithm, detailing the initialization, teaching, learning, and output phases.

Experimental Results and Discussion

Since no standardized dataset exists for CMfg service composition, the authors developed a synthetic dataset featuring 20,000 services, each characterized by six QoS metrics: time, price, reliability, success rate, cost, and reputation. Improved-TC was compared against Particle Swarm Optimization (PSO), Improved Grey Wolf Optimizer (IGWO), standard TLBO, and the original Hybrid-TC.

Key Findings

  • Convergence Speed: Improved-TC reached convergence at approximately the 220th iteration, significantly faster than Hybrid-TC and other variants, which required between 580 and 700 iterations.
  • Solution Quality: Across nine different problem sizes (varying subtasks from 10 to 30 and candidate services from 150 to 450), Improved-TC consistently identified solutions with higher QoS fitness values.
  • Stability: Statistical analysis via box plots showed that Improved-TC had the lowest variance and fewest outliers, indicating highly reliable performance across multiple runs compared to the unstable results of PSO and GWO.

Conclusion and Future Work

Improved-TC represents a significant advancement in solving the SCOS problem for large-scale cloud manufacturing. By utilizing dimension-level granularity and hybrid optimization strategies, it provides a robust and efficient method for service selection. While the current model is static, future research will focus on integrating Reinforcement Learning to develop a dynamic adaptive model capable of handling real-time fluctuations in service attributes and network conditions.

Summary updated

Abstract

In the cloud manufacturing process, service composition can combine a single service into a complex service to meet the task requirements. An efficient service composition strategy is crucial, as it affects the efficiency of resource and capacity sharing in the cloud manufacturing system. However, in the face of a large-scale environment, the existing methods have the problems of slow convergence and instability. To solve the problems above, we propose an improved optimization method, named improved-TC. Specifically, we are inspired by the horizontal crossover of CSO in the hybrid-TC teaching phase, the Hybrid-TC is proposed in our previous work, which is a hybrid of the teaching-learning-based optimization algorithm (TLBO) and the crisscross optimization algorithm (CSO). Improved-TC is an improvement on the learning phase of hybrid-TC algorithm, we change the search method of hybrid-TC in the learning phase to a one-dimensional search, thereby some dimensions in the population that are trapped in the local optimum have the chance to jump out of the iteration. Experiments show that our proposed method has a faster convergence speed and more stability in the face of service composition in large-scale environments.
BibTeX
@article{zeng2022service,
  author = {Jun Zeng and Juan Yao and Min Gao and Junhao Wen},
  title = {{A service composition method using improved hybrid teaching learning optimization algorithm in cloud manufacturing}},
  journal = {J. Cloud Comput.},
  year = {2022},
  doi = {10.1186/s13677-022-00343-0},
}
Figure