Acta Armamentarii

Figure/Table detail

Dynamic Prediction Model Based on Multi-level LSTM Network for Multi-agent Attack and Defense Effectiveness

DINGWei;MINGZhenjun;WANGGuoxin;YANYan

Acta Armamentarii, 2023, 44(1): 176-192. DOI: 10.12382/bgxb.2022.0192

参数		第4组仿真	第21组仿真	第35组仿真
	任务完成度	2.56	5.49	4.33
红方	作战效率	3.41	6.67	4.93
	作战成本	3.17	4.28	3.86
	任务完成度	1.48	3.60	3.32
蓝方	作战效率	2.77	5.36	2.69
	作战成本	2.50	4.15	3.63

Table 8 MRE of operational effectiveness parameters %

Other figure/table from this article

Fig.1 Multi-level characteristics of the established model
Fig.2 Temporal and quantitative characteristics of the established model
Fig.3 Overall framework of MAAD based on bottom-up emergence
Fig.4 Operational process of MAAD based on OODA loop
Fig.5 Dynamic environment of MAAD reconstructed using NetLogo platform
Table 1 Decision-making parameters concerning the attack and defense of agents
Fig.7 Schematic diagram of attack and defense population structure drawn by AutoCAD
Fig.8 Structure of LSTM networks
Table 5 Parameter settings of the model
Table 2 The utility part of multi-agent attack and defense process
Fig.6 Multi-agent distribution diagram for explaining the population structure parameters
Fig.9 Correlation logic and mapping relationship between multiple levels of MAAD
Fig.10 Data storage method of multi-level LSTM networks
Fig.11 Training prediction process of multi-level LSTM networks
Fig.12 Middle-value change of multi-level LSTM networks
Fig.13 Simulation of the attack and defense of red and blue agents using NetLogo platform
Table 3 Initialization of decision-making parameters in the N-th simulation
Table 4 Parameters concerning population structure and operational effectiveness of the MAAD in the N-th simulation
Fig.14 Data preprocessing: dividing the training data into mini-size batches
Table 6 Comparison between the tested and predicted output concerning operational effectiveness parameters
Table 7 Distribution of absolute errors concerning operational effectiveness parameters using frequency distribution histograms
Table 9 Comparison of performance indicators using different methods
Table 10 Comparison of the prediction results of 40 testing sets

模态框（Modal）标题

Figure/Table detail