Page 73 - Read Online

Ma et al. Intell Robot 2023;3(4):581-95 I http://dx.doi.org/10.20517/ir.2023.33 Page 13 of 15

Table 2. Comparison of various indicators of different algorithms
Algorithms AUC ACC F1 Recall
Google-Net 0.8974 0.9029 0.8812 0.8765
Fast-CNN 0.9018 0.8917 0.8995 0.9121
GCN 0.9327 0.9470 0.9321 0.9274
GAT 0.9401 0.9123 0.9247 0.9338
GHNN-ACO 0.9531 0.9379 0.9409 0.9387

ture used in this paper can better identify the task progressive aspect and non-task progressive aspect, and
through local and global intelligent feature learning and recognition, this algorithm can better learn the main
characteristics and properties of task allocation, as shown in Figure 6.

Meanwhile, inthelinkpredictiontask, theGHNN-ACOalgorithmachievesefficientcollaborativecooperation
between intelligent agents. Through the path search of ACO algorithms, agents can find the optimal communi-
cation path, thus achieving more efficient task execution and cooperation. This kind of effective collaboration
between agents is crucial for emergency response, which enables the entire Multi-agent system to respond to
different emergencies more efficiently.

Moreover, although the GHNN-ACO algorithm has shown certain advantages, we have also identified some
potential limitations. This algorithm is sensitive to initial parameter selection and requires careful adjustment
to avoid falling into local optima. Accordingly, the algorithm in this paper has an accuracy rate of 95.31% in
assigning multiple tasks to multiple agents, as shown in Table 2.

In addition, due to the iterative updating of GHNNs and ACO algorithms, the computing cost is high, which
may pose challenges to the real-time application of large-scale Multi-agent systems. In future research, we
will strive to further optimize the GHNN-ACO algorithm. By improving the initial parameter selection and
optimizing the parallel computing ability of the algorithm, we hope to further improve the efficiency and
performance of the algorithm. In addition, we will also explore its application in other types of multi-agent
scenarios to expand its applicability and generalization.

5. CONCLUSIONS
In conclusion, the research results of this paper demonstrate that the GHNN-ACO algorithm performs well in
heterogeneousmulti-agentscenarios, providinganeffectivesolutionfornodeclassificationandlinkprediction
tasks in agent systems. This algorithm has potential application prospects in emergency response of multi-
agent systems and also offers a new idea for further research and optimization of multi-agent cooperation.
The algorithm in this paper achieves an accuracy rate of 95.31% in assigning multiple tasks to multiple agents.
We are optimistic about the application of the GHNN-ACO algorithm in heterogeneous multi-agent scenarios
and believe that it will play an important role in future intelligent agent systems.

DECLARATIONS
Acknowledgments
Dr. Ma expresses his gratitude to Prof. Duan for his remarks and discussion of multi-agent formation control
theory and paper structure issues.

Authors’ contributions
Made substantial contributions to the research, idea generation, and software development and conducted the
GHNN-ACO experiments. Wrote and edited the original draft: Ma Z