<p>Knowledge graphs can improve the performance of recommendation systems and provide explanations for recommendation results, which have been widely applied in the next Point-of-Interest (POI) recommendation. However, the current knowledge graph method for the next POI recommendation focuses on the static attributes of POIs, and only describes the spatio-temporal characteristics when the user transfers between POIs. To fully tap into user preferences for different POIs, we have done the following innovative work. (1) We construct a user preference knowledge graph with spatio-temporal characteristics, named UPSTKG, which expresses preference information from both individual user and global user perspectives. (2) We use local preference triplets in preference knowledge graphs to construct user preference graphs. And use GCN to obtain user preference vectors to replace common user vectors in the sequence, thereby strengthening the potential connection between users and different POIs. (3) We combine UPSTKG and user preference graph to propose the UPSTKGRec method for the next POI recommendation. To evaluate the effectiveness of UPSTKGRec, it is compared to six highly regarded techniques on three distinct benchmark datasets. Compared with the baseline, the average performance of indicators recell@5 and NDCG@5 has increased by 13.8% and 13.1%.</p>