Multi-sensor, multi-source information fusion presents significant challenges in complex real-world applications such as power consumption prediction, where existing methods often have limitations in capturing both spatio-temporal features and fully exploit complex relationships among multi-variate features simultaneously. In real-world scenarios, such as complex electrical power system settings, capturing both correlations is important, as spatio-temporal contains vital geographical information and complex inter-series relationships between features. To address these limitations, we propose AutoGRN for enhancing prediction accuracy and efficiency in multi-source spatio-temporal data fusion, with a focus on complex electrical power system settings. AutoGRN integrates a novel adaptive multi-channel attentive framework with copula-based dependency modeling, combining graph neural diffusion convolution and recurrent optimization. The framework automatically learns spatial features, capturing complex correlations among regions, while a sequence encoder extracts temporal patterns, ensuring the acquisition of time series characteristics such as seasonality and trends. High-dimensional spatio-temporal features are then fused through a specially designed multi-channel recurrent graph neural network, incorporating copula functions to model complex dependencies between variables. Extensive experiments on multiple real-world electricity consumption datasets demonstrate that AutoGRN achieves substantial advantages over state-of-the-art benchmarks in multi-variate prediction tasks, showcasing its potential for applications in various multi-sensor, multi-source fusion scenarios, particularly in complex systems requiring simultaneous analysis of spatial and temporal dynamics with intricate inter-variable dependencies. Code is available at https://github.com/AmbitYuki/AutoGRN.

中文翻译：

---

AutoGRN：一种自适应多通道图递归联合优化网络，具有基于 Copula 的依赖性建模，用于电力系统中的时空融合


多传感器、多源信息融合在复杂的实际应用中提出了重大挑战，例如功耗预测，在这些应用中，现有方法在捕获时空特征和同时充分利用多变量特征之间的复杂关系方面往往存在局限性。在实际场景中，例如复杂的电力系统设置，捕获这两种相关性非常重要，因为时空包含重要的地理信息和要素之间复杂的序列间关系。为了解决这些限制，我们提出了 AutoGRN 来提高多源时空数据融合的预测准确性和效率，重点是复杂的电力系统设置。AutoGRN 将新颖的自适应多通道注意力框架与基于 copula 的依赖建模相结合，结合了图神经扩散卷积和递归优化。该框架自动学习空间特征，捕获区域之间的复杂相关性，而序列编码器提取时间模式，确保获取季节性和趋势等时间序列特征。然后，通过专门设计的多通道循环图神经网络融合高维时空特征，并结合 copula 函数来模拟变量之间的复杂依赖关系。 在多个真实世界用电量数据集上的广泛实验表明，AutoGRN 在多变量预测任务中比最先进的基准具有显着优势，展示了其在各种多传感器、多源融合场景中的应用潜力，特别是在需要同时分析具有复杂变量间依赖关系的空间和时间动力学的复杂系统中。代码可在 https://github.com/AmbitYuki/AutoGRN 获取。