Key Technologies Of Reasoning Engine For Intelligient Decision

Combining rule-based reasoning(RBR)and case-based reasoning(CBR),hybrid reasoning is becoming a new research hotspot of artificial intelligence.Domestic and foreign scholars have done a lot of theoretical and application research in hybrid reasoning;however,most of the existing research mainly focus on specific problems and are defected in high coupling,poor portability and low performance from algorithm design to system application.Providing unified reasoning services for intelligent decision support systems in cloud computing as well as big data environment is the core research content of this paper.Multiple application scenarios,increasing domain knowledge base,massive data and high frequency reasoning request,such factors challenge the inference engine greatly in system architecture,hybrid reasoning scheduling mechanism,and distributed parallel reasoning technology.Based on Ministries research project "XX intelligent information processing support technology" as the background,the key technology of the inference engine is studied systematically and thoroughly.The main research work and innovative achievements are:(1)Propose a hybrid reasoning system framework that supports process orchestration and distributed reasoning Most of the existing hybrid reasoning systems integrate RBR and CBR in a fixed manner,and thus are lack of flexibility and versatility.The paper studies the dynamic characteristics of intelligent decision and the mental model of human reasoning ability,proposes a hybrid reasoning integration mechanism based on the idea of data driven,enables inference engine with the support capability for inference process orchestration,which can determine the integrated scheduling strategy of hybrid reasoning at run time,eventually overcomes the shortcomings of the fixed hybrid reasoning integration mechanism.Based on SOA architecture,designs a distributing parallel reasoning engine framework combining RBR and CBR,the hierarchical design of this framework decouples system function reasonably.The analysis shows that the reasoning engine framework has good adaptability in different application scenarios.(2)Propose a reasoning ability evaluation and adaptive scheduling mechanism based on knowledge base coverage In systems whose domain knowledge base are dynamic changing,the quantitative evaluation of the knowledge base is the key to evaluate RBR and CBR ability reasoning and to make reasonable scheduling.The existing knowledge evaluation indexes are not suitable to represent the information of the whole knowledge base.Based on the DIKW knowledge model,this paper establishes unified forms of concept expression for both rule knowledge and case knowledge,proposes a knowledge base coverage index which can describe the comprehensive degree of the knowledge base,introduces data cube as a computing tool for such index,and gives the calculation method and process of coverage for rule base and case base seperately.The experiments show that the reasoning capability evaluation and adaptive scheduling based on knowledge base coverage can effectively improve the efficiency of hybrid reasoning.(3)Propose a distributing parallel rule-based reasoning model dpRBR Massive data,high frequency reasoning applications promote traditional RBR technology developed into distributing parallel RBR.The existing distributing RBR system is insufficient in the aspect of the rule base segmentation algorithm,which brings the communication overhead of the distributed node and low efficiency in reasoning.This paper studies topological properties of the rule base,establishes directed hypergraph model of rule base,introduces Newman fast cohension aggregation method from complex network,proposes a rule base segmentation algorithm based on community structure aggregation and gives application examples.Based on the MapReduce model,the distributed RBR model dpRBR is designed,and the communication and scheduling mechanism of the distributed inference master and slave node is given.The experiments show that the dpRBR model has higher inference efficiency.(4)Propose a distributing parallel case-based reasoning model dpCBR The traditional CBR algorithm has the disadvantages of high computational complexity and high computational redundancy.The existing improving algorithms and distributing CBR method have many shortcomings in massive data,high frequency reasoning environment.This paper presents a distributed parallel CBR model dpCBR as a solution.Introduces the projection pursuit technology,establish a benchmark case in source case base and calculate one-dimensional projection distance,conducts pruning optimization on source case base during case retrieval based on one-dimensional projection distance,which greatly reduces complexity and redundancy of case match.The distributing parallel CBR model dpCBR is designed based on the MapReduce model,and the mechanism of historical data reuse is presented.The performance of dpCBR model is verified through simulation experiment.(5)Implements an inference engine prototype application example Based on the technologies mentioned above,implements a prototype RBR-CBR hybrid reasoning engine system,and gives examples of the application in air defense.The design scheme of prototype system is simple and efficient,as well as high flexibility and strong processing capability,which meets the design requirements and performance standard of the background project,and has been verified by application in platform,and achieved good results.The work of this paper is supported by Ministries research project,"XX intelligent information processing platform".Result and achievement of this paper are applied in this research project.This research project passed the inspection and came to acceptance of Ministries in June 2016.