A One-factor Granular Fuzzy Logical Relationship Based Multi-point Ahead Prediction Model

In the existing one-factor high-order fuzzy logical relationship (FLR) based forecasting models, each FLR reflects the influence from the premise trend on the consequent. The consequent of each FLR is related to only one observation at one moment. Therefore, such FLRs are used only in one-point ahead prediction but not multi-point ahead prediction. In order to deal with multi-point ahead prediction, the concept of granular FLRs is proposed in this paper. Each granular FLR reflects the influence on the consequent trend from the premise trend which is described by a granule built on a group of observations. Note that the consequent of each granular FLR is also a granule built on a group of observations. To select the available granular FLRs for forecasting, the k-medoids algorithm is employed to cluster the obtained FLRs. Based on the constructed granular FLRs and the clustering results, a new forecasting model is proposed. The superior performance of this model is illustrated by comparing with the other existing models in the experimental studies.