Accurate and Efficient Multivariate Time Series Forecasting via Offline Clustering

--Accurate and efficient multivariate time series (MTS) forecasting is essential for applications such as traffic management and weather prediction, which depend on capturing long-range temporal dependencies and interactions between entities. Existing methods, particularly those based on Transformer architectures, compute pairwise dependencies across all time steps, leading to a computational complexity that scales quadratically with the length of the input. T o overcome these challenges, we introduce the Forecaster with Offline Clustering Using Segments (FOCUS), a novel approach to MTS forecasting that simplifies long-range dependency modeling through the use of prototypes extracted via offline clustering. These prototypes encapsulate high-level events in the real-world system underlying the data, summarizing the key characteristics of similar time segments. By identifying prototypes during the offline clustering phase, FOCUS reduces the computational complexity of modeling long-range dependencies in the online phase to linear scaling. Extensive experiments across diverse benchmarks demonstrate that FOCUS achieves state-of-the-art accuracy while significantly reducing computational costs. Accurate and efficient multivariate time series (MTS) forecasting is of great importance in various real-world applications [3], [13], [16], [46], [65], [77], [78], [82], [83]. MTS data, characterized by its complex structure encompassing both temporal and entity dimensions [22]-[24], [47], [77], [85], is crucial for enabling precise predictions across different application scenarios. For example, in traffic flow forecasting, accurate predictions can optimize resource allocation [16], [17], [77], while in weather forecasting, efficient predictions can provide timely warnings to the public [35], [73]. A key challenge in achieving accurate MTS forecasting lies in modeling long-range dependencies [34], [52], [62], [92]. Modeling these long-range dependencies involves two primary steps: First, identifying potential events within the data, such as peaks in traffic or fluctuations in climate. Each event corresponds to a cluster of similar time segments, where each cluster shares a representative segment pattern, referred to as a prototype. Next, these prototypes and their corresponding events are analyzed to further explore their temporal relationships and interactions between entities.