Research On Multi-Modal Optimization And Ensemble Learning Based Fuzzy Cognitive Maps Learning Algorithms And Their Applications

Fuzzy cognitive maps(FCMs)are a kind of intelligent soft computing methods,which combine neural networks and fuzzy logic.FCMs are generally applied to model and analyze complex dynamical systems.To model and analyze complex dynamical systems,we use FCMs to model the units and the relationships between units in complex dynamical systems,and we need to learn the relationships between concepts in FCMs.In recent years,many scholars did research in FCMs and proposed evolutionary-based learning algorithms to learn FCMs.The evolutionary-based learning algorithms,which need not the experts' knowledge,can learn proper FCMs from historical data automatically.For the problem of FCM learning,we concluded three important essential factors,namely,the model of FCMs,the objective function and the optimization processing.We did a series of work targeting at each factor.The followings are the main work of this dissertation:1.An FCM learning algorithm based on ensemble learning and real-coded genetic algorithm is proposed.In this algorithm,we applied ensemble learning to FCMs and proposed the boosting FCMs(BFCMs).In the experiments,the proposed algorithm is applied to learn BFCMs from synthetic data with varying sizes and densities.In addition,the proposed algorithm is validated on the benchmark datasets DREAM3 and DREAM4.The experimental results show that the proposed algorithm is effective to the FCM learning problem and the gene regulatory networks reconstruction problem.2.An FCM learning algorithm based on the convergence of FCMs and multi-agent genetic algorithm is proposed.In this algorithm,we proposed another objective based on the convergence of FCMs,and optimized this objective to learn FCMs.In the experiments,the proposed algorithm is applied to learn FCMs from synthetic data with varying sizes and densities.The experimental results show that the proposed algorithm can learn FCMs with high accuracy.In addition,the proposed algorithm is validated on the gene benchmark datasets.The experimental results show that the proposed algorithm can reconstruct gene regulatory networks effectively.3.A niching-based multi-modal multi-agent genetic algorithm is proposed for the FCM learning problem.In this algorithm,the FCM learning problem is modeled as a multi-modal optimization problem.In the experiments,the proposed algorithm is applied to learn FCMs from synthetic data with varying sizes and densities.The experimental results show that the proposed algorithm can learn FCMs with high accuracy.In addition,the proposed algorithm is validated on the benchmark datasets DREAM3 and DREAM4.The experimental results show that the proposed algorithm outperforms other FCM learning algorithms obviously,which illustrates that the proposed algorithm can reconstruct gene regulatory networks effectively.