TL;DR
A novel DRL-based approach that integrates bidirectional scheduling with graph-theoretic features to effectively solve JSSP is introduced, which achieves superior performance compared to traditional heuristics and classic DRL methods, while maintaining competitive results against recent state-of-the-art approaches.
This study proposes a structurally simple, low-node-density disjunctive graph model, then performs feature extraction using Graph Neural Networks (GNNs), and finally optimize convergence using the Migrating Birds Optimization—an intelligent optimization algorithm proven effective for scheduling decision problems.
The Job Shop Scheduling Problem (JSSP) is commonly represented as a disjunctive graph in which nodes denote operations, while edges encode technological precedence constraints and machine‐sharing conflicts. Most existing deep reinforcement learning (DRL) approaches model this graph as homogeneous by merging precedence and contention edges into a single relation type, thereby overlooking their distinct semantics. To address this limitation, we propose the Heterogeneous Graph Transformer (HGT)‐Scheduler, a reinforcement learning framework that explicitly models the JSSP as a heterogeneous graph. Unlike existing learning methods, the proposed framework employs edge‐type‐specific attention mechanisms to distinguish precedence and machine‐contention relations, enabling richer scheduling representations. The scheduling policy is optimized using Proximal Policy Optimization (PPO). The proposed framework is evaluated on the Fisher–Thompson benchmark instances. On FT06, the HGT‐Scheduler achieves an optimality gap of 8.4%, significantly outperforming both an identical architecture without edge‐type awareness () and a Graph Isomorphism Network (GIN) baseline. On the larger FT10 instance, the proposed approach demonstrates favorable scalability, although heterogeneous and homogeneous representations exhibit comparable performance under a 50,000‐step training budget. Ablation studies identify a three‐layer attention architecture as the most effective configuration. Overall, the results demonstrate that explicitly modeling edge semantics improves reinforcement learning for intelligent job shop scheduling.
Flexible job shop scheduling with multiple automated guided vehicles (FJSP-AGV) is a challenging production scheduling problem in intelligent manufacturing, where operation sequencing, machine assignment, AGV allocation, and transportation decisions are tightly coupled. Existing exact and meta-heuristic methods can obtain high-quality solutions, but they usually require considerable computational time for large-scale instances. Meanwhile, conventional dispatching rules can make fast decisions but often fail to capture the complex interactions among operations, machines, and AGVs. To address these challenges, this paper proposes an end-to-end deep reinforcement learning framework based on heterogeneous graph neural networks for solving FJSP-AGV. Specifically, a heterogeneous graph is constructed to represent the scheduling state, where operations, machines, and AGVs are modeled as different types of nodes, and their relationships are described by operation–machine and operation–AGV arcs. Based on this representation, a heterogeneous graph neural network is developed to extract scheduling information from different production resources. In particular, a meta-path aggregation mechanism is introduced to capture the complex interaction patterns among operations, machines, and AGVs. The proximal policy optimization algorithm is then employed to train the scheduling policy in an end-to-end manner. Experimental results on public benchmark instances and real-world cases demonstrate that the proposed method outperforms composite heuristic rules and achieves a favorable balance between solution quality and computational efficiency compared with existing state-of-the-art methods. These results indicate that the proposed HGNN-DRL framework is effective for fast and intelligent scheduling decision-making in FJSP-AGV environments.
PGMPO is proposed, a novel learning framework consisting of a simple but effective multi-policy modeling approach that allows a single network to represent multiple decision-makers, and a preference-driven model optimization method that effectively guides policies to learn diverse and specialized problem-solving strategies without the need for explicit reward functions.
Rising energy costs and the increasing share of renewable generation create incentives to align production schedules with dynamic electricity prices and on-site solar generation. We address the resulting Multi-Objective Flexible Job Shop Scheduling Problem by proposing a deep reinforcement learning framework that jointly minimizes makespan and energy cost. A single preference-conditioned policy approximates the Pareto front at inference time, eliminating the need to train separate models for different objective weightings. The agent acts as a hyper-heuristic, selecting among heuristic actions at each decision point, including strategies that intentionally delay operations to exploit periods of lower electricity prices or higher solar generation. Preferences are integrated throughout the network via Feature-wise Linear Modulation, while a dual-critic architecture and a diversity loss preserve preference-specific policy behaviors. We evaluate the approach against NSGA-II and Joined Heuristics on synthetic instances ranging from 10×5×5 to 15×15×15 jobs, operations per job, and machines using normalized hypervolume and inverted generational distance. While NSGA-II performs best on the smallest instances, the proposed approach becomes increasingly competitive as problem size grows and achieves the best results on the largest evaluated instances. These findings indicate promising scalability within the investigated problem range.