面向不确定缺失信息的多模式自主切换态势评估方法研究

As an important aiding link of decision-making in information air combat, the intelligent construction of situation assessment system is very significant. In the face of large uncertain battlefield information, situation awareness and inference technology under current fixed mode often leads to inaccurate assessment results because of the lack of self-adjusting capability. Starting from adaptive adjustment of model and inference mechanism, a systematic air combat situation assessment model is established and an autonomous switching multi-mode inference method for uncertain missing information is proposed. Firstly, the effective information is chosen and evidence network is applied to establish the initial model of modular situation assessment. Then, on the basis of systematic analysis of uncertainty and missing degree of evidence, various inference switching modes are designed. The corresponding evidence processing methods are studied, and the dynamic adaptive generation algorithm of network parameters in each mode is proposed. Finally, in order to improve the rationality of inference results, an improved inference method based on temporal-spatial fusion mechanism is designed, and an off-line optimization method based on expert supervision mechanism is proposed for variable weight parameters. The successful promotion of this research will effectively solve the problem that the model and parameters in situation assessment are relatively fixed or largely dependent on knowledge base, and change the inflexible approach of single inference mechanism, which has important theoretical significance and practical value for autonomous and intelligent development of situation assessment technology under the complex information environment.

作为信息化空战的重要辅助决策环节，态势评估系统的智能化建设至关重要。现有固定模式下的态势感知和推理技术在面对大量多变的不确定性战场信息时，缺乏自调整能力常导致评估结果不准确。本课题以模型与推理机制自适应调节为出发点，建立系统化的空战态势评估模型并提出面向不确定缺失信息的多模式自主切换推理方法。首先，选取有效信息应用证据网络建立模块化态势评估初始化模型；然后，全面分析证据信息的不确定性及缺失程度，设计多种推理切换模式、研究相应的证据处理方法并提出各模式下网络参数的动态自适应生成算法；最后，为提高推理结果的合理性，设计基于时空融合机制的改进推理方法，并针对变权参数提出基于专家监督机制的离线优化方法。此项研究的成功推进将有效解决态势评估中模型及参数相对固定或过于依赖知识库的问题，改变推理机制单一的不灵活做法，对于复杂信息环境下态势评估技术的自主化、智能化发展有着重要的理论意义及应用价值。

现有的态势评估等信息推理系统模型和推理方法相对固定，在面对随机性不完备信息时，由于缺乏自适应调整能力难以给出持续合理的推理结果。针对此问题，研究并提出了一种能够自主通过信息的类型和不完备程度选择并切换证据输入模式、完成证据的有效修正并在连续推理过程中实现模型、参数乃至规则自适应调整的多源异构信息融合方法。该方法着眼于为不完备信息下的推理系统赋予人的分析和处理问题的思维方式，增强机器推理的智能性，对于扩展人工智能技术在军事、医疗等不确定性推理领域的应用具有重要的理论意义和实际价值。项目首先针对多模式自主切换证据网络的方法展开研究，在扩展了信度函数运算的基础上提出了包括基于信息类型的模块化证据网络建模方法、基于信息不完备程度的多种证据输入模式划分及证据修正方法、基于模式切换和变权原理的网络节点权重自适应生成方法、针对模块化证据网络的三步时空融合方法以及基于粒子群算法的模型参数优化方法等。然后，进行了方法的应用研究。提出了基于DDQN、DDPG改进算法的无人机机动决策模型建立方法，构建了无人机对抗仿真场景。在此基础上，针对不完备信息下的无人机态势（威胁）评估、空中目标识别以及车况安全性等推理问题，分别构建模块化网络模型，在连续推理过程中证实了方法在不完备信息下的可行性和有效性。