Parallel Immune Algorithm Based On GPU And Its Application For Production Scheduling Of Tandem Cold Mill

The artificial immune system is one of the high-powered AI systems, which is proposed on the basis of the immune mechanism and characteristic. Nowadays, it plays an important reference role for solving many complex projects. However, the solving efficiency of this type of algorithms still cannot satisfy the real-time requirements in engineering application. Parallel computing can effectively shorten the solving time by a large margin, and has received attentions from more and more researchers. But, since the communication between the processes in parallel computing is rather cost-consuming, the actual parallel or distributed algorithm comes with substantial expenditures of hard ware, management, and maintenance.Graphic processing unit (GPU) has rapidly developed in recent years, and its abilities of high floating-pointing operations, parallel processing and programmable function provide a good platform for the general calculation besides graphic processing. At present, the parallel computing based on GPU has become a research hotspot in the field of high performance computing. In this study, a parallel immune algorithm (PIA) based on GPU acceleration which is a case study on combination optimization problem, is proposed, in which the combination with tabu search greatly improves the search ability and the convergence speed. In engineering application, we map the implementation of PIA into the processing of single instruction multiple threads on GPU structure. For verifying the effectiveness of the proposed algorithm, we take various traveling salesman problems (TSP) as the study instances. The experiments indicate that the PIA not only obtains the sound results, but greatly improves the computing efficiency for the TSP.Finally, we take the production scheduling problem of cold rolling line as the study background and employ an existing modeling for the cold-rolled steel coils planning. The proposed algorithm is then applied to optimize the scheduling model. The parallel immune algorithm is compiled to a dynamic link library (DLL) and is embedded in the backstage executable code of the system which is programmed by C# language. Using the DLL solves the linking problem between the system and parallel computing kernel. And, the system takes the advantage of user-friendly interface, ease of servicing and maintenance and parallel computing with CUDA. Besides, if there is a need to change the cold rolling process constraint, then the algorithm can be realized only via re-compiling the algorithm DLL without modifying the other program parts. In such way, the advantage of system openness and the flexibility of the algorithm can be exerted.