Abstract: Concerning the insufficient integration of local and global dependencies in existing time series models, a method for multivariate time series prediction namely PatchLG (Patch-integrated Local-Global correlation method) was proposed, which can better integrate both local and global correlations. The proposed method was based on three key components: 1) segmenting the time series into multiple patches, preserving the locality of the time series while making it easier for the model to capture global dependencies; 2) utilizing depthwise separable convolution and self-attention mechanism to model local and global correlations; 3) decomposing the time series into trend and seasonal components, and combining the predictions of these two components to obtain the final result. Experimental results on seven benchmark datasets demonstrate that PatchLG achieves an average improvement of 3.0% and 2.9% in Mean-Square Error (MSE) and Mean Absolute Error (MAE), respectively, compared to the optimal baseline method PatchTST, and possesses low actual running time and memory usage, validating the effectiveness of PatchLG in time series forecasting.

Key words: time series, multivariate time series prediction, depth-separable convolution, self-attention mechanism, local and global dependence

摘要： 为解决现有时间序列模型未能充分融合局部和全局依赖的问题，提出一种融合局部和全局相关性的多变量时间序列预测方法PatchLG（Patch-integrated Local-Global Correlation Method）。该方法基于3个关键部分：1）将时间序列划分为多个子序列（Patch），保持时间序列的局部性同时使模型更易于提取全局依赖；2）使用深度可分离卷积和自注意力机制建模局部和全局相关性;3）将时间序列分解为趋势项与季节项两个部分同时进行预测，并将预测结果组合起来得到最终预测结果。在7个基准数据集上的实验结果表明，PatchLG相较于最优基线方法PatchTST在均方误差(MSE)和平均绝对误差(MAE)两个指标上平均提升3.0%和2.9%，同时具有较低的实际运行时间和内存消耗，验证了PatchLG在时间序列预测的有效性。

关键词: 时间序列, 多变量时间序列预测, 深度可分离卷积, 自注意力机制, 局部与全局依赖