基于轻量化图注意力机制的露天矿卡车调度优化算法  

Truck Scheduling Optimization Algorithm for Surface Mine Based on Lightweight Graph Attention Mechanism

作　　者：黄石 陈钊宇[2] 曾蕾 HUANG Shi;CHEN Zhaoyu;ZENG Lei(College of Automotive Technology,Sichuan Vocational and Technical College,Suining 629000,China;College of Architecture&Environment,Sichuan University,Chengdu 610065 China;Department of Mechanical and Electrical Engineering,Sichuan Transportation Vocational School,Chengdu 611000,China)

机构地区：[1]四川职业技术学院汽车技术学院,四川遂宁629000 [2]四川大学建筑与环境学院,四川成都610065 [3]四川交通运输职业学校机电工程系,四川成都611100

出　　处：《金属矿山》2024年第4期202-208,共7页Metal Mine

基　　金：四川省教育厅科研项目(编号:17ZB0395)。

摘　　要：有效地管理和调度露天矿卡车,有助于大幅提升运输效率、降低矿山运营成本。现有研究聚焦于利用深度强化学习(Deep Reinforcement Learning,DRL)构建学习模型求解路径优化问题,然而,该模型针对Transformer架构的参数训练时,会产生大量参数冗余。为此,提出了一种轻量化图注意力机制的露天矿卡车调度优化算法。将微分方程数值解法——阿当姆斯(Adams)法用于Transformer模型的权重学习中,通过Adams的残差训练方法,可提高网络后期的优化精度,进一步压缩模型的规模,高效求解露天矿卡车调度优化问题。研究表明:该方法在降低最优间隙的同时将源模型的参数量压缩1/2,减少了对GPU设备的训练依赖。采用随机生成的露天矿卡数据集算例对该算法性能进行了验证,反映出采用Adams-Transformer模型有助于提升露天矿卡车调度效率。Effectively managing and scheduling open-pit mine trucks can significantly improve transportation efficiency and reduce mining operation costs.Existing research focuses on using Deep Reinforcement Learning(DRL)to construct learn-ing models for solving path optimization problems.However,when training models with Transformer architecture parameters,a large number of redundant parameters are generated.To address this issue,this paper proposes a lightweight graph attention mechanism for optimizing open-pit mine truck scheduling.Specifically,the Adams method,a numerical solution for differential equations,is employed in the weight learning of the Transformer model.A residual training method based on Adams is proposed to improve the optimization accuracy of the network in the later stages and further compress the model size,efficiently solving the open-pit mine truck scheduling optimization problem.The research shows that this method can reduce the optimal gap while compressing the parameter size of the source model to half,reducing the training dependency on GPU devices.Performance ver-ification of the algorithm is conducted using randomly generated open-pit mine truck datasets,demonstrating that the Adams-Transformer model helps improve the efficiency of open-pit mine truck scheduling.

关 键 词：露天矿 卡车调度优化 阿当姆斯法 图注意力机制 深度强化学习 

分 类 号：TD57[矿业工程—矿山机电] TD67