Study of time series clustering based on FCM

In this paper, Fuzzy Cognitive Map (FCM) is employed and discussed to analyze time series that represents fault dynamics. At first, the essential of FCM prototype is studied, and diversified data set that are produced from the prototype to construct candidate FCM model is proposed. The Particle Swarm Optimization (PSO) and Simulated Annealing (SA) learning algorithm is taken as training method based on computational intelligence. Secondly, analysis of representation for time series from real world application is carried out with the proposed FCM model to assess the quality of FCM design methodology and algorithmic performance. The dynamics of time series is depicted by fuzzy c-mean clustering, and the concept of the node in the candidate FCM model corresponds to the activation level of each cluster center. In the end, the parameters of node number in the candidate FCM model and steepness of activation function are taken into consideration in order to obtain the better result. The discussion and findings are offered and the results show the proposed method is desirable in the representation of time series both at numeric value level and linguistic level.