In the era of big data, prediction has become a fundamental capability. Current prediction methods primarily focus on sequence elements; however, in multivariate time series forecasting, time is a critical factor that must not be overlooked. While some methods consider time, they often neglect the temporal distance between sequence elements and the predicted target time, a relationship essential for identifying patterns such as periodicity, trends, and other temporal dynamics. Moreover, the extraction of temporal features is often inadequate, and discussions on how to comprehensively leverage temporal data are limited. As a result, model performance can suffer, particularly in prediction tasks with specific time requirements. To address these challenges, we propose a new model, TE-LSTM, based on LSTM, which employs a temporal encoding method to fully extract temporal features. A temporal weighting strategy is also used to optimize the integration of temporal information, capturing the temporal relationship of each element relative to the target element, and integrating it into the LSTM. Additionally, this study examines the impact of different time granularities on the model. Using the Beijing International Airport station as the study area, we applied our method to temperature prediction. Compared to the baseline model, our model showed an improvement of 0.7552% without time granularity, 1.2047% with a time granularity of 3, and 0.0953% when addressing prediction tasks with specific time requirements. The final results demonstrate the superiority of the proposed method and highlight its effectiveness in overcoming the limitations of existing approaches.

In recent years, geographic knowledge graphs (GeoKGs) have shown great promise in representing spatio-temporal and event-driven knowledge. However, existing knowledge graph embedding approaches mainly focus on structural patterns and often overlook the dynamic evolution of entities in both time and space, which limits their effectiveness in downstream reasoning tasks. To address this, we propose a spatio-temporal evolutionary knowledge embedding approach (ST-EKA) that enhances entity representations by modeling their evolution through type-aware encoding, temporal and spatial decay mechanisms, and context aggregation. ST-EKA integrates four core components, including an entity encoder constrained by relational type consistency, a temporal encoder capable of handling both time points and intervals through unified sampling and feedforward encoding, a multi-scale spatial encoder that combines geometric coordinates with semantic attributes, and an evolutionary knowledge encoder that employs attention-based spatio-temporal weighting to capture contextual dynamics. We evaluate ST-EKA on three representative GeoKG datasets—GDELT, ICEWS, and HAD. The results demonstrate that ST-EKA achieves an average improvement of 6.5774% in AUC and 5.0992% in APR on representation learning tasks. In question answering tasks, it yields a maximum average increase of 1.7907% in AUC and 0.5843% in APR. Notably, it exhibits superior performance in chain queries and complex spatio-temporal reasoning, validating its strong robustness, good interpretability, and practical application value.