Abstract: Overheating of high-voltage cables can lead to insulation thermal breakdown, compromising the stability of power grid. However, existing research primarily focuses on traditional prediction models, which often fail to capture the complexity and dynamic characteristics of temperature data. To address this limitation, a novel cable temperature prediction model based on multi-scale Patch and convolutional interaction (MSP-CI) was proposed. Initially, input dimension was reduced using a channel resampling method, and a multi-scale patch branch structure was designed to decouple complex time series. Then, macroscopic information from coarse-grained patches and microscopic information from fine-grained patches were extracted through a combination of sequence decomposition and convolution interaction strategy. Finally, an attention fusion module was introduced to dynamically balance the weights of macroscopic and microscopic information, yielding the final prediction results. Experimental results on real high-voltage cable temperature datasets demonstrate that, compared to baseline models such as TimeMixer, PatchTST, and MSGNet, MSP-CI achieves an average reduction of 6.97% to 35.79% in mean squared error (MSE) and 6.36% to 42.84% in mean absolute error (MAE). These findings indicate that MSP-CI effectively enhances cable temperature prediction accuracy, providing a reliable basis for power dispatching operations.

Key words: high-voltage cable, temperature prediction, multi-scale, patch, convolutional interaction

摘要： 高压电缆长期过热可能导致绝缘热击穿，进而影响电网的稳定性。然而，当前研究主要集中在传统预测模型上，忽略了温度数据的复杂性和动态特征。为解决此问题，提出了一种基于多尺度Patch与卷积交互的电缆温度预测模型(Multi-Scale Patch and Convolutional Interaction, MSP-CI)。首先，采用通道重组采样方法降低输入维度，并构建多尺度Patch分支结构，以实现复杂时间序列解耦；其次，结合序列分解与卷积交互策略，分别提取粗粒度Patch的宏观信息与细粒度Patch的微观信息；最后，构建注意力融合模块，动态平衡宏观与微观信息的权重，得到最终预测结果。在真实高压电缆温度数据集上的实验结果表明，MSP-CI相较于TimeMixer、PatchTST、MSGNet等基线模型，在均方误差(MSE)上平均下降了6.97%~35.79%，在平均绝对误差(MAE)上平均下降了6.36%~42.84%。MSP-CI能有效提升电缆温度预测的准确率，为电力调度运行方式提供依据。

关键词: 高压电缆, 温度预测, 多尺度, patch, 卷积交互