中圖分類號：TM715 文獻標志碼：A DOI: 10.16157/j.issn.0258-7998.257379 中文引用格式： 毛洪彬，舒征宇，雷明，等. 基于LASSO及GRU-注意力機制的配電網合環電流區間預測[J]. 電子技術應用，2026，52(6)：74-82. 英文引用格式： Mao Hongbin，Shu Zhengyu，Lei Ming，et al. Interval prediction of distribution network loop-closing current based on LASSO and GRU-attention mechanism[J]. Application of Electronic Technique，2026，52(6)：74-82.
Interval prediction of distribution network loop-closing current based on LASSO and GRU-attention mechanism
Mao Hongbin，Shu Zhengyu，Lei Ming，Ren Guanchen，Zhou Zihan
China Three Gorges University
Abstract： Against the backdrop of large-scale integration of distributed energy resources, such as photovoltaic and wind power, into the power grid, traditional methods for calculating loop-closing currents face challenges of insufficient accuracy and difficulty in characterizing the uncertainty of loop-closing characteristics. To address this, this paper proposes an interval prediction model for loop-closing currents that considers uncertainty. Specifically, the research process includes the following key steps. First, given the numerous influencing factors and complex data structure in loop-closing current prediction, this paper adopts the Adaptive LASSO (ALASSO) regression method to screen influencing factors and construct a multivariate dataset suitable for loop-closing current prediction. Second, to enhance the interpretability of the loop-closing current prediction model, the study introduces the Extreme Gradient Boosting algorithm to evaluate the feature importance of the influencing factors selected by LASSO, clarifying the contribution of each factor to the prediction results. Then, to address the temporal characteristics of loop-closing currents, this paper proposes a prediction model based on an attention-mechanized Gated Recurrent Unit (GRU) for time-period prediction of loop-closing currents. This model dynamically allocates weights through the attention mechanism to capture key features in the time-series data of loop-closing currents, thereby improving prediction accuracy. Finally, to more accurately represent the uncertainty of loop-closing current predictions, the study employs the Bootstrap method to scientifically calculate the confidence interval, thereby quantifying the potential fluctuation range of the prediction results. Subsequently, the calculated confid
Key words : adaptive LASSO；Bootstrap algorithm；attention mechanism；feature extraction；loop-closing current；uncertainty
為提升配電網合環電流預測的動態精度與泛化能力，本文提出一種融合注意力機制與混合神經網絡的概率化預測模型。該模型采用三層技術架構：特征提取與權重分析、時序建模及不確定性量化。首先，基于自適應LASSO（Least Absolute Shrink age and Selection Operator）回歸構建特征解耦層，通過動態調整懲罰項系數，篩選氣象、拓撲和動態負荷等高維特征，生成低冗余特征子集。其次，利用XGBoost算法評估特征重要性，量化各變量對合環電流的邊際貢獻，增強模型可解釋性。隨后，構建GRU時序建模模塊，借助門控機制捕捉長期依賴關系，并引入注意力機制對隱含狀態動態加權，突出關鍵時間窗口的電流響應特征。為進一步量化預測不確定性，采用Bootstrap方法分析預測誤差的統計分布，構建誤差置信區間，并將其與點預測結果融合，形成覆蓋不確定性的區間預測。最終輸出具備概率信息的合環電流預測結果，為電網運行決策提供更可靠的參考依據。
