基于图神经网络和深度强化学习的二维矩形排样优化方法研究

Research on optimization method of two-dimensional rectangular nesting based on graph neural network and deep reinforcement learning

本文结合生产实际中的零件母板带约束二维矩形下料优化问题,设计并提出了二维矩形排样问题的异构图和深度强化学习的算法架构。通过图神经网络和强化学习算法对排样问题中零件和母板的特征进行高度的集成和学习,并对零件的排布顺序和排布位置进行决策,在更短的时间内得到相比于传统优化算法更优秀的计算结果。实验证明,本文的深度强化学习算法训练的模型可以在较短时间得到良好的排样结果,且基于小规模问题训练的模型解决较大规模的问题实例也可以获得较好的效果,证明了算法具有较好的泛化能力。

In this paper,a heterogeneous graph and deep reinforcement learning algorithmic architecture for the 2D rectangular nesting problem is designed and proposed in combination with the 2D rectangular nesting optimization problem with constraints on the mother plate of the part in production practice.Through the graph neural network and reinforcement learning algorithm,the features of the parts and motherboards in the nesting problem are highly integrated and learned,and the decision of the order and location of the parts is made,so that better computational results are obtained in a shorter time compared with the traditional optimization algorithms.Experiments have proved that the model trained by the deep reinforcement learning algorithm in this paper can get good nesting results in a shorter period of time,and the model trained based on a small-scale problem to solve larger-scale problem instances can also get better results,proving that the algorithm has a better generalization ability.