Scholarlink

This study proposes a residual-term-based outlier detection and correction method combined with a hierarchical fusion model of Temporal Convolutional Network (TCN) and Gated Recurrent Unit (GRU) for load forecasting. The additive decomposition model is used to separate trend, seasonal, and random components, with outliers identified via residual normalized scores and repaired by linear interpolation. SHapley Additive exPlanations (SHAP) values analyze seasonal feature importance differences to optimize inputs. The TCN-GRU model captures long-range dependencies and dynamic patterns effectively. Experimental results show that the proposed model outperforms TCN, TCN-LSTM, etc., with lower MAE, RMSE, MAPE, and a high R2 of 0.9967, verifying its effectiveness and superiority in load forecasting.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Hybrid TCN-GRU Model for Load Forecasting Using SHAP-Based Interpretable Feature Selection

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Fei Zhao,
  • Yong Li,
  • Chang Li,
  • Li Liu,
  • Qian Yu,
  • Yong Xu

摘要

This study proposes a residual-term-based outlier detection and correction method combined with a hierarchical fusion model of Temporal Convolutional Network (TCN) and Gated Recurrent Unit (GRU) for load forecasting. The additive decomposition model is used to separate trend, seasonal, and random components, with outliers identified via residual normalized scores and repaired by linear interpolation. SHapley Additive exPlanations (SHAP) values analyze seasonal feature importance differences to optimize inputs. The TCN-GRU model captures long-range dependencies and dynamic patterns effectively. Experimental results show that the proposed model outperforms TCN, TCN-LSTM, etc., with lower MAE, RMSE, MAPE, and a high R2 of 0.9967, verifying its effectiveness and superiority in load forecasting.