Research On Fuzzy Cognitive Map Learning Method Based On Multi-Objective And Multi-Modal Optimization

Complex systems exist widely in all fields of our life.By mining the complex dynamic representation data of systems to explore the internal interaction mechanism and then model them can provide a necessary means for cognition,prediction and regulation of complex systems.Fuzzy cognitive map is a soft computing method combining fuzzy logic theory and cognitive mapping,which is widely used in the field of complex system modeling and analysis.However,the existing fuzzy cognitive map learning methods are easy to fall into local convergence,and have the limitation that it is difficult to accurately capture the relationships among concept nodes,and the generalization is poor in the highdimensional solution space.In addition,the existing learning methods often only focus on the numerical fitting ability of the fuzzy cognitive map,which leads to the excessive density of the obtained cognitive map network.To address these challenges,this thesis studies the fuzzy cognitive map learning methods from the perspectives of multi-disciplinary and multi-field integration,including multimodal optimization,multiobjective optimization,intelligent optimization and machine learning.The main research contents are as follows:(1)In order to solve the problem that the existing fuzzy cognitive map learning algorithms are easy to fall into local convergence and it is difficult to accurately capture the relationships among concept nodes,this thesis proposes a multimodal artificial bee colony algorithm to learn fuzzy cognitive map.Firstly,the fuzzy cognitive map learning problem is modeled as a multimodal optimization problem,then a complete set of fuzzy cognitive map learning scheme is designed from the perspective of multimodal optimization.Secondly,combining the artificial colony algorithm and the niching method,an optimization method with multiple convergence ability is proposed,the group cognitive guidance strategy and local search strategy are introduced to improve the performance of this algorithm.Finally,experiments are carried out on 8 real system datasets,and the results show that the proposed fuzzy cognitive map learning algorithm is superior to 6 baseline models and has higher accuracy and generalization.(2)In the large-scale fuzzy cognitive map learning problem,the solution space dimension is large and the relationships among concept nodes are complicated,which leads to the poor accuracy and low efficiency of learning methods.A multimodal covariance matrix adaption evolution strategy is proposed to learn large-scale fuzzy cognitive map in this thesis.Firstly,the fuzzy cognitive map learning problem is modeled into multiple multimodal optimization problems by combining divide-and-conquer strategy.Secondly,a multimodal optimization method is proposed by combining niching method and covariance matrix adaption evolution strategy,and incorporating cubic chaos operator.Finally,experiments are conducted on 15 large-scale gene regulatory networks and 12 synthetic datasets.The results show that the proposed method is superior to 8 baseline models and can learn large-scale fuzzy cognitive maps efficiently.(3)Most fuzzy cognitive map learning algorithms only focus on the numerical fitting ability of the model,which leads to the high density of the learned cognitive map network.And existing solutions are poorly equipped to deal with conflicting goals.This thesis proposes a multimodal and multiobjective optimization algorithm to learn sparse fuzzy cognitive map.Firstly,in order to improve the diversity of decision space,the fuzzy cognitive map learning problem is modeled as a multimodal multiobjective optimization problem.Secondly,a multimodal multiobjective optimization algorithm is proposed by combining the algorithm proposed in the previous research with multiobjective optimization strategy and iterative threshold search operator.A complete set of fuzzy cognitive map learning scheme is proposed from the perspective of multimodal and multiobjective optimization.Finally,the experiment is conducted on 5 real system datasets and 18 synthetic datasets.Experimental results show that this method can deal with conflicting objects efficiently and learn sparse fuzzy cognitive map accurately.