Research On Linguistic Rule-based Reasoning And Application In Uncertain Environments

With the development of artificial intelligence,it is the era of a new generation of artificial intelligence,namely cognitive intelligence,which not only requires machine learning based on big data and perception recognition,but also hopes that machines can imitate human reasoning and thinking to solve problems in uncertain environments.Linguistic value is the main tool used by human beings to obtain information or knowledge in the process of cognition,decision-making and execution,which can express the fuzziness and randomness in the uncertain environment.Knowledge representation and reasoning based on linguistic value is a very important research of artificial intelligence in the uncertainty.Therefore,this thesis studies the linguistic rulebased reasoning and application in uncertain environments.The main research can be summarized as follows:(1)Aiming at the problem of multi-granularity representation of multi-source linguistic information in uncertain environment,the knowledge representation and knowledge reasoning methods of multi-granularity linguistic value are studied based on the linguistic lattice implication algebra.The linguistic lattice-valued distribution assessment representation and the linguistic lattice-valued hierarchical structure are proposed to represent multi-granularity linguistic values.The multi-granularity linguistic lattice-valued conversion method is proposed to implement the granularity consistency of multi-granularity linguistic values,which can realize the transformation between linguistic variables into different granularities without the information loss.The multi-granularity linguistic value belief rule base and the corresponding reasoning method are established to realize the rule-based reasoning of multi-granularity linguistic information,and the validity and rationality of the method are illustrated by an example.(2)Aiming at the inconsistency and incompleteness of linguistic knowledge in uncertain environments,the intuitionistic fuzzy linguistic knowledge representation and knowledge reasoning method are studied based on the linguistic intuitionistic fuzzy lattice.In order to deal with the linguistic information based on both positive evidence and negative evidence,the linguistic intuitionistic fuzzy preference relation is established to determine the linguistic weights.For the inconsistent information caused by subjective uncertainty,a consistency checking algorithm is introduced to check and repair the consistency.For the incomplete information caused by objective uncertainty,an improved linguistic intuitionistic fuzzy preference relation is presented to complement the lost information.In order to realize the reasoning method of intuitionistic fuzzy linguistic value,two kinds of weighted operators are proposed to verify the rationality of weighted operation.The linguistic intuitionistic fuzzy rule base and the corresponding rule-based reasoning method are established.The validity and rationality of the method are illustrated by an example.(3)Aiming at the fusion of data-driven and knowledge-driven,an interpretable classification model is studied based on uncertainty knowledge representation and knowledge reasoning methods.To alleviate the low interpretability of machine learning,the convolutional rule inference network(CRIN)is proposed based on belief rule-based inference methodology using the evidential reasoning approach(RIMER).Considering the influence of global data distribution on attribute weights,the determination of each attribute may be affected by all other attributes,based on the sigmoid activation function,the optimized attribute weights are determined to make sure that each attribute weight is formed by all attribute values.Considering the influence of local rules input on final output in the network,the network framework and learning algorithm based on convolution strategy are proposed.The inference process in RIMER is used as the feedforward mechanism to ensure the interpretability of the network.Meanwhile,the gradient descent method is used as back propagation algorithm to adjust the parameters to establish a more reasonable belief rule base.The experimental results and comparative analysis demonstrate that the proposed CRIN has advantages in terms of interpretability and learning capability.In summary,the research on the representation and reasoning of linguistic knowledge in uncertain environments based on linguistic lattice implication algebra and linguistic intuitionistic fuzzy lattice provide a theoretical basis for learning algorithm based on rule-based reasoning.The established CRIN can not only realize the classification in uncertainty,but also combine knowledge representation and reasoning with machine learning,which suggests a new trend for interpretable network and uncertain classification issues in the fusion of data-driven and knowledge-driven.