Research On Complex System Modeling Method Based On Interpretable Belief Rule Base

Complex system modeling is able to analyze,evaluate and predict itself according to the environment in which it is located in order to ensure the normal operation of the system and reduce the incidence of accidents.Damage to equipment caused by aging and other reasons during long-time operation of complex systems reduces the safety of complex systems and leads to many dangerous accidents.Therefore,it is necessary to model the complex system to ensure the safe operation of the complex system by modeling the actual state of the complex system.In addition,for complex systems with strong risk sensitivity and high reliability requirements,the modeling process should also focus on the interpretability of the model,and a model structure with interpretability can improve the trust of decision makers in the output results of the model,which has theoretical and practical application value.However,the existing complex systems can collect less observational data,other information needs to be added to improve the model accuracy.In addition to the observed data,the expert knowledge obtained from the analysis of the system mechanism is also an important source of system information.Therefore,there is the problem of how to combine multiple sources of information in complex system modeling to improve the accuracy and interpretability of modeling.To address these issues,this paper uses the belief rule base(BRB)to model different complex systems and focuses on the construction of complex system analysis models,evaluation models,and prediction models.The main contributions of this paper are as follows:(1)To solve the problem of reduced interpretability of the belief rule base during the optimization process,a self-growth belief rule base with interpretability constraints(SBRB-I)analysis modeling method is proposed.This method first designs two interpretability constraints based on the actual complex system to ensure that the interpretability of the optimized model is not compromised.Secondly,a self-growth learning strategy is proposed,which mainly integrates any available expert knowledge during the optimization process to improve the model’s interpretability and optimization performance,achieving effective coordination between expert systems and data-driven models and fully exploiting their respective advantages.(2)To solve the problem of interpretability evaluation of the belief rule base,an evaluation modeling method with an interpretable belief rule base(BRB-I)is proposed.This method rationally constructs the system through mechanism analysis in an interpretable way,including effective attribute extraction and model structure interpretability.In addition,an interpretability evaluation criterion based on Euclidean distance is proposed,and two interpretability strategies are designed according to this criterion to further improve interpretability in model optimization.(3)To reduce the complexity and improve the interpretability of the model,a deep belief rule-based model with interpretability(DBRB-I)prediction modeling method is proposed.This method first proposes a deep structure belief rule base model through trend analysis of the complex system,making the model highly scalable.Secondly,a reliable and transparent sensitivity analysis rule reduction method is proposed to reduce the size of the rule base and improve the model’s readability.Finally,through data statistical analysis and mechanism analysis of the complex system,three effective interpretability constraints are designed to improve the interpretability in model optimization.In summary,this paper proposes corresponding solutions to the problems of designing reasonable interpretability constraints,evaluating the interpretability of the model,and reducing model complexity in the BRB system: SBRB-I,BRB-I,and DBRB-I.The construction of complex system analysis models,evaluation models,and prediction models is achieved through the above three model structures,and the rationality of the proposed methods is verified from a theoretical perspective and verified in different complex systems.