Recently, significant advancements have been made in time-series forecasting research, with an increasing focus on analyzing the inherent characteristics of time-series data, rather than solely focusing on designing forecasting models.In this paper, we follow this trend and carefully examine previous work to propose an efficient time series forecasting model based on linear models. The model consists of two important core components: (1) the integration of different semantics brought by single-channel and multi-channel data for joint forecasting; (2) the use of a novel loss function that replaces the traditional MSE loss and MAE loss to achieve higher forecasting accuracy.On widely-used benchmark time series datasets, our model not only outperforms the current SOTA, but is also 10 $\times$ speedup and has fewer parameters than the latest SOTA model.

翻译：近年来，时间序列预测研究取得了显著进展，研究重点逐渐转向分析时间序列数据的内在特征，而非仅关注预测模型的设计。本文遵循这一趋势，在仔细审视先前工作的基础上，提出了一种基于线性模型的高效时间序列预测模型。该模型包含两个重要核心组件：（1）整合单通道与多通道数据带来的不同语义进行联合预测；（2）采用新型损失函数替代传统MSE损失和MAE损失，以实现更高的预测精度。在广泛使用的基准时间序列数据集上，我们的模型不仅超越了当前最先进（SOTA）水平，而且相比最新SOTA模型实现了10倍的加速，且参数量更少。