Towards multi-task transfer optimization of cloud service collaboration in industrial internet platform

Cloud Manufacturing (CMfg) has gained significant attention owing to its capability in reshaping the cooperation paradigm among multiple geographically dispersed enterprises, which is conducive to handle a complex production task flexibly through the industrial internet platform. Cloud Service Assembly (CSA) is concerned with integrating a series of services together for serving a complex manufacturing task, which, as one of bottlenecks for CMfg, plays a critical role in efficient utilization of resources. Evolutionary Algorithms (EAs) have been widely used in resolving CSA in the past. However, they are always executed from scratch for tackling a single task in each run, whereas handling a batch of tasks collectively via leveraging inter-task knowledge transfer has been scarcely studied. Notably, CMfg is often faced with situation of multiple tasks arriving dynamically. In light of this, we propose a Multi-task Transfer EA (MTEA), where several service collaboration tasks are optimized jointly to speed up the search efficiency by exploiting knowledge extraction among tasks. Specifically, data models derived from evolving populations are learned to capture valuable knowledge for transfer so as to boost problem-solving efficacy, a parameter online learning strategy is utilized to tune the intensity of knowledge transfer across tasks. Extensive experiments are conducted on a series of CSA instances, results prove the feasibility and competence of MTEA against state-of-the-art peers.