分布式并行算法在长周期原油混输调度中的应用

Application of distributed parallel algorithm for long term crude oil scheduling

为了有效求解长周期原油混输调度问题,提出了基于事件树的分布式并行算法。该方法把原油混输调度问题分解为码头调度子问题和厂区调度子问题,采用基于事件树的建模方法,并根据两个子问题的求解顺序提出了原油混输调度问题的分布式并行算法。本方法采用主从式并行结构,主节点把求解码头调度子问题所需的原油质量要求信息发送到各从节点,然后各从节点把与质量要求信息对应的码头调度最优解返回给主节点,通过综合比较两个子问题的解,从而得出最优的调度方案。实例计算表明,该并行算法可以有效减少问题的求解时间,特别是对不同常减压对原油质量要求不同时的长周期调度(如4周)问题,采用串行算法在48h内都无法得到可行解,而采用此算法用3台计算机可以在25h内得到最优解。

A novel distributed parallel algorithm for crude oil scheduling is presented in order to shorten the long computation time in the scheduling of a long scheduling horizon problem. The event-tree based modeling and optimization algorithms are adopted and the whole problem is divided into two sub problems （the portside sub problem and the refinery sub problem）. Thus, the algorithm is proposed by analyzing the solving sequence of the two sub problems. The algorithm is based on master-slave distributed model. The master node solves the refinery sub problem firstly and sends the quality changeover sequence that is needed to start the portside scheduling to slave nodes, then the slave nodes send the results back to the master node. By comprehensive comparison of the results of the two sub problems, an optimized scheduling scheme will be obtained. The numerical example demonstrates that the algorithm is efficient in reducing the time overhead, especially when the crude oil distillation units （CDUs） have different feed quality requirements and the scheduling horizon is very long. For example when the horizon is 4 weeks, the sequential algorithm can not get a feasible solution in 48 h while the proposed algorithm can get optimal solution within 25 h.