Complex Dynamic Slow Independent Symbol Aggregation Approximation for Multimodal Temporal Fuzzy Cognitive Map

The current mainstream time series prediction methods exhibit commendable accuracy in prediction, but they often lack interpretability. One approach to address this issue is through the utilization of fuzzy cognitive maps for prediction. The multimodal fuzzy cognitive maps method stands out for its superior interpretability and adeptness in handling nonlinear and uncertain problems. However, the traditional modal segmentation method makes it difficult to effectively identify the modal information, affecting the method's effectiveness and accuracy. In this article, we proposed a novel multimodal fuzzy temporal cognitive maps time series prediction method based on complex dynamic slow independent symbol aggregation approximation. First, we constructed complex dynamic slow independent symbol aggregation approximation, which integrates slow features to obtain the modal information of the model. Then, the multimodal model parameters were determined by an optimization algorithm. Finally, a multimodal online matching mechanism based on slow feature tracking was proposed to address transitional states between modals. The validation based on two examples shows that the proposed method can effectively improve the prediction accuracy.