Research On Knowledge Reasoning And Complex Network Node Evaluation Technology For Target System Analysis

The confrontation of modern warfares,manifested as a systematic contest between combat forces and resources.How to choose the target to attack to achieve effective injury to enemy system is the key to command decision.Therefore,using system thinking to analyze targets is related to the success or failure of military action.It is of great significance to carry out research on theories and methods related to the analysis of target systems.Most of the traditional target system analysis is based on the established target systems to analyze key targets and parts.However,in the fierce confrontation of modern warfare,the internal relationship of the target system is complex and dynamic,at the same time,the target system is affected by the “fog”of the battlefield.How to construct the target system and analyze the target system quickly,efficiently and accurately becomes more challenging.Therefore,this paper proposes a method that supports the dynamic iterative analysis of target systems,which involves using graph-based knowledge representation method to formally describe the target system knowledge,using default reasoning method to constructe the target architecture relationship model(TARM),and establishing a complex network dynamics model of the TARM to analyze the key nodes.The main work and innovations of this paper are as follows:(1)The flexible homomorphism and an efficient flexible homomorphic search algorithm for graph rules are proposedWhen using the graph-based knowledge representation method to describe the target system,it is difficult to establish a unified ontology support.To solve this problem,on the basis of traditional homomorphic reasoning,this paper proposes the flexible homomorphism that supports reasoning based on multiple concepts and relations,which have partial order structure.The graph rules with the flexible homomorphism improve the usage flexibility of themselves.The basic operation of using graph rules is homomorphic search,which is a typical NP-hard problem.In order to improve the efficiency of graph rules,this paper studys three techniques,which are to use reinforcement learning to optimize the homomorphic matching order of the graph rule nodes,to optimize concept and relation comparison sequence based on the statistical method and to use the node labels filtering alternative nodes.Then an efficient hybrid homomorphic search algorithm is obtained.The flexible homomorphism and hybrid homomorphic search algorithm provide knowledge support for automatic construction of TARMs.(2)The stochastic default reasoning method based on hierarchical structure prioritization is proposed.The opaque battlefield leads to the inferred TARMs with many possibilities.At the same time,as the battle progresses,the TARM will also change.Therefore,the construction of TARM is non-monotonic.Based on traditional default reasoning,this paper proposes a new stochastic default reasoning method based on hierarchical structure prioritization.In the new method,there is a graph structure priority on default rules.The reasoning based on the graph structure priority avoids missing extensions,which will occurr in the reasoning with the strict full order on defaults.In reasoning process,the new method randomly selects the default rule under the constraints of the graph structure priority,and the new method has good parallelism and performs in parallel reasoning.In the face of multiple possible reasonable extensions,the new method establishes the priorities based on expected accuracy,precision and recall between these extensions to determine robust semantics,which makes that the failure decision of the TARM is the least costly.(3)The default reasoning method based on recurrent neural network(RNN)is proposed.The stochastic default reasoning method is used to obtain all possible TARMs,which are complex,and the computational complexity is high.In response to this problem,this paper proposes to use the RNN to enhance the stochastic default reasoning method to improve the reasoning efficiency and pertinency.By classifying the historical reasoning data,the corresponding training data set is established to train the RNN.The trained RNN can recommend the rules for default reasoning under the constraint of the graph structure priority.It increases the efficiency of generating extensions by reducing the number of invalid rule usage.Compared with the stochastic default reasoning,the default reasoning guided by RNN has more pertinence,and it can more efficiently generate TARMs that meet the requirements.At the same time,for obtaining the training data of the RNN,this paper proposes a simplified processing method,which effectively reduces the complexity of training data processing.(4)The node evaluation method based on complex network dynamics model is proposed.Traditional node analysis methods based on complex networks mostly evaluate the importance of nodes based on topology information,ignoring the characteristics of nodes themselves.In the face of the complex network model of the TARM,the intrinsic characteristics of each node are distinct.To this end,this paper proposes a node evaluation method based on complex network dynamics model,which includes disturbance test and damage test,and the complex network dynamics model is transformed form the TARM.In the concrete implementation,they are all realized through dynamic simulation.The node importance is evaluated for the case that the node's own function is destroyed and can be recovered or can not be recovered.Since the disturbance test and the damage test include the network topology information and the characteristics of the node itself,and the operation mechanism of the target system is revealed.Therefore,the node importance assessment method proposed in this paper can more fully reflect the importance of different nodes in the target system.Based on the above research,this paper designs and implements the target system auxiliary analysis prototype system.The parallel computing framework based on virtual machines(VMs)technology is realized,which effectively improves the TARM reasoning efficiency.The conclusion-based node coding method based on the default rules is designed,and it effectively reduces the computing complexities of the accuracy,precision and recall of the extensions.The HTML-based graphical display is used to improve the human-computer interaction.Finally,based on the typical target system analysis case,the experiment results show that the proposed method in this paper is reasonable and effective.