Multivariate prediction is a crucial tool for control system design, allowing for accurate prediction and control of system states by modeling the relationships between multiple input and output variables. However, existing models often rely on dimensionality reduction techniques to manage high-dimensional data, which can lead to the loss of valuable information. Combined prediction models (CPMs) have gained significant attention for their ability to integrate the strengths of multiple predictors, thereby improving prediction accuracy. Nevertheless, most current CPMs overlook the relationship between each predictor's output and the predicted sequences, resulting in suboptimal performance. To address these challenges, we propose an adaptive-network-based fuzzy inference system CPM (ANFIS-CPM) enhanced by an improved fuzzy C -means (FCM) clustering, termed ANFIS-CPM-FCM. Our approach defines a similarity metric to quantify feature relationships, enhances traditional FCM to automatically determine the optimal number of clusters based on a density clustering model, trains separate ANFIS models on each cluster, and aggregates predictions by considering the relationships between each predictor's output and the predicted sequences. Extensive comparative and ablation experiments on six publicly available datasets demonstrate that our ANFIS-CPM-FCM outperforms existing methods in terms of prediction accuracy and robustness, highlighting the benefits of integrating improved clustering with adaptive fuzzy inference systems.
Zhang et al. (Thu,) studied this question.