基于资源流网络和图卷积策略网络的多元物资存储空间保障仿真中的资源重调度方法研究

Research on Resource Rescheduling Method in Multi-Resource Storage Space Support Simulation Based on Resource Flow Network and Graph Convolutional Policy Network

摘要:

对于大型存储基地,因其存储的物资的种类多样、数量庞大,为了有效管理,需要按照设备的类别进行分类存放,如重设备存储空间、轻设备存储空间、电子设备存储空间等,并对上述空间按照要求进行不同形式的保障,如清扫、检修等。因为空间的容量、类别及其内部存储物资的不同,保障需求在时间周期、保障事件等方面差异较大。该保障过程以保障任务序列为基础,在保证序列中每项任务的保障资源得到满足的前提下,力求资源配置调整最小化。首先对保障资源调度问题进行详细描述,并建立相应的模型。其次,针对现有资源流网络模型在复杂多元物资存储空间保障任务关系表示方面存在的不足,对该模型进行改进,并将保障资源重调度问题转换为完备资源流网络生成问题。再次,利用并扩展图表示学习、强化学习和生成对抗网络,提出基于图卷积策略网络(GCPN)的完备资源流网络生成方法。最后,开展仿真实验,对保障资源调度模型进行仿真验证。

Abstract:

In large-scale storage bases, due to the diversity and vast quantity of stored materials, effective management requires classifying and storing materials according to their categories, such as heavy equipment storage space, light equipment storage space, electronic equipment storage space, etc. Different types of spaces require various forms of support, such as cleaning, maintenance, etc. Due to differences in space capacity, categories, and the materials stored within them, support demands vary significantly in terms of time cycles and support events. This support process is based on a sequence of support tasks, ensuring that the resources required for each task in the sequence are met while striving to minimize resource reallocation adjustments.Firstly,conduct a detailed description of the support resource scheduling problem and establish the corresponding model.. Secondly, to address the shortcomings of existing resource flow network models in representing complex multi-resource storage space support task relationships, the model is improved, and the resource rescheduling problem is transformed into a complete resource flow network generation problem. Thirdly, leveraging and extending graph representation learning, reinforcement learning, and generative adversarial networks, a complete resource flow network generation method based on a Graph Convolutional Policy Network (GCPN) is proposed. Finally, simulation experiments are conducted to validate the proposed resource support scheduling model.