29 May 2025 in IEEE Transactions on Cybernetics
Spatio-temporal forecasting plays a crucial role in the dynamic perception of smart cities, such as traffic flow prediction, renewable energy forecasting, and load prediction. Its objective is to understand the patterns of spatio-temporal changes under the interaction of various factors. Accurate spatio-temporal forecasting relies on sufficient high-quality data and powerful models. However, in reality, data is often sparse. In such cases, while adaptive graphs and large language models (LLMs) can maintain performance, they face issues of spatial spurious associations and hallucinations, respectively. These issues hinder the ability of the model to learn and infer cross spatio-temporal and cross-scale features effectively. To address this, we propose a novel model termed spatio-temporal causal intervention large language model (STCInterLLM). This model employs a newly designed causal intervention encoder to update spatial spurious correlations in the spatio-temporal adaptive graph. Subsequently, the novel chain-of-action prompting text is utilized to enforce the decomposition of the prediction process, thereby enhancing the causal representation of features while mitigating hallucinations in LLMs. Finally, a lightweight marker alignment module ensures the consistency between the encoder, prompting text, and LLM, enabling accurate forecasting of distinct scale spatio-temporal evolution patterns. Extensive experiments conducted on power distribution systems integrated with renewable energy sources and transportation systems encompassing diverse types of data, demonstrate that the proposed STCInterLLM consistently achieves state-of-the-art performance across significantly varied scenarios. Codes are available at https://github.com/lishijie15/STCInterLLM.
13 August 2025 in IEEE Transactions on Power Systems
Precise and efficient net load forecasting is crucial for Power Distribution Systems (PDS) to address the various challenges posed by the increasing penetration of Renewable Energy Sources (RES). Existing studies on net load forecasting often overlook the characteristic differences (i.e., variable heterogeneity) between loads and Distributed Generation (DG), and the differences in graph structures (i.e., spatio-temporal heterogeneity) caused by changes in node net loads. These studies typically assume constant node properties and use the same model with shared parameters to extract spatio-temporal features for different variables, which limits the representation of the features. To address these challenges, this study proposes a novel model named Dual-branch Experts Fusion Memory Network (DEFMN). Specifically, this model designs customized experts (feature branch) with independent parameters for loads and different types of DG (variable branch) to extract features according to the characteristics of various RES. We also employ shared parameters modules, including series embedding, meta spatial memory, and decoder, to fully capture the spatio-temporal correlations between loads and DG, effectively learning their variable heterogeneity as well as spatio-temporal heterogeneity. Additionally, we propose a new model named load-DG coupling model, which aims to construct a novel large-scale PDS with different RES penetration based on the IEEE 8500- node test feeder. Extensive simulations have been conducted on various challenging scenarios (i.e., PDSs with varying RES penetration), and the proposed DEFMN consistently achieves state-of-the-art performance in terms of accuracy (MAPE, etc.) and effectiveness in these scenarios. Codes are available at https://github.com/lishijie15/DEFMN.