Scholarlink

<p>To address the issue of reduced prediction accuracy under complex load variations, this paper proposes an enhanced TCN-BiGRU short-term load forecasting model integrated with a Temporal Pattern Attention (TPA) mechanism. The model improves prediction performance through a multi-stage deep feature fusion framework: First, historical power load data are standardized and processed for outliers to construct a high-quality input dataset. Second, the dilated causal convolution structure of the Temporal Convolutional Network (TCN) is employed to efficiently capture local features and long-term dependencies in load sequences. Subsequently, a Bidirectional Gated Recurrent Unit (BiGRU) network is constructed, where hyperparameters such as hidden layer dimensions and learning rates are optimized using the Golden Jackal Optimization (GJO) algorithm. This approach not only overcomes the limitations of manual parameter tuning but also significantly enhances the network’s bidirectional temporal feature extraction capability. The optimized BiGRU mechanism then simultaneously mines forward and backward global correlation features in time-series data. Finally, the TPA mechanism dynamically adjusts the weight distribution of multi-scale temporal features to amplify the contribution of critical temporal patterns to prediction results. Simulation experiments on real-world power load datasets demonstrate that the GJO-optimized hyperparameter configuration enables the network to converge faster to its optimal state, thereby improving load forecasting accuracy.</p>

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

Short-Term Power Load Forecasting Based on GJO-TCN-BiGRU-TPA

  • 期刊论文

  • 发表日期：

    2025年9月19日

查看全文
  • Wei Wang,
  • Yudong Xu,
  • Yong Liu,
  • Lu Bai,
  • Guanliang Li,
  • Xiang Jing

摘要

To address the issue of reduced prediction accuracy under complex load variations, this paper proposes an enhanced TCN-BiGRU short-term load forecasting model integrated with a Temporal Pattern Attention (TPA) mechanism. The model improves prediction performance through a multi-stage deep feature fusion framework: First, historical power load data are standardized and processed for outliers to construct a high-quality input dataset. Second, the dilated causal convolution structure of the Temporal Convolutional Network (TCN) is employed to efficiently capture local features and long-term dependencies in load sequences. Subsequently, a Bidirectional Gated Recurrent Unit (BiGRU) network is constructed, where hyperparameters such as hidden layer dimensions and learning rates are optimized using the Golden Jackal Optimization (GJO) algorithm. This approach not only overcomes the limitations of manual parameter tuning but also significantly enhances the network’s bidirectional temporal feature extraction capability. The optimized BiGRU mechanism then simultaneously mines forward and backward global correlation features in time-series data. Finally, the TPA mechanism dynamically adjusts the weight distribution of multi-scale temporal features to amplify the contribution of critical temporal patterns to prediction results. Simulation experiments on real-world power load datasets demonstrate that the GJO-optimized hyperparameter configuration enables the network to converge faster to its optimal state, thereby improving load forecasting accuracy.