| 引用本文: | 丁伟锋,周震震,谢振华,等.基于时序卷积神经网络和纵横交叉算法的低压台区负荷预测[J].电力系统保护与控制,2025,53(21):156-165.[点击复制] |
| DING Weifeng,ZHOU Zhenzhen,XIE Zhenhua,et al.Low-voltage distribution area load forecasting based on temporal convolutional network and crisscross optimization algorithm[J].Power System Protection and Control,2025,53(21):156-165[点击复制] |
|
| 摘要: |
| 精准的电力负荷预测对低压台区运维至关重要。为提升台区电力负荷预测精度，提出一种纵横交叉算法(crisscross optimization algorithm, CSO)优化卷积注意力机制(convolutional block attention module, CBAM)和时序卷积神经网络(temporal convolutional network, TCN)的低压台区电力负荷预测模型。首先，建立以时序卷积神经网络为基础的预测模型，提取电力负荷输入序列隐含的时间规律。其次，在模型输入侧引入CBAM模块，通过在通道和空间上与模型输入进行加权，提高模型对关键特征的敏感性。最后，为解决模型参数易陷入局部最优、模型泛化性不高的问题，提出使用CSO对CBAM-TCN的全连接层进行二次优化。以广东省某地两个典型低压台区实测电力负荷数据集进行仿真建模，结果表明所提组合预测方法性能优于其他对比模型，并对其有效性进行了验证。 |
| 关键词: 低压台区 负荷预测 纵横交叉算法 时序卷积神经网络 卷积注意力机制 |
| DOI：10.19783/j.cnki.pspc.241617 |
| 投稿时间：2024-12-04修订日期：2025-04-21 |
| 基金项目:国家自然科学基金青年科学基金项目资助(52207166)；中国南方电网有限责任公司科技项目资助(012000KK52200014) |
|
| Low-voltage distribution area load forecasting based on temporal convolutional network and crisscross optimization algorithm |
| DING Weifeng1,ZHOU Zhenzhen1,XIE Zhenhua2,XIAO Yaohui1,HUANG Heyan1,HE Sen1 |
| (1. CSG EHV Electric Power Research Institute, Guangzhou 510620, China; 2. School of Automation,
Guangdong University of Technology, Guangzhou 510006, China) |
| Abstract: |
| Accurate power load forecasting is crucial for the operation and maintenance of low-voltage distribution areas. To improve the accuracy of power load forecasting, this paper proposes a low-voltage load forecasting model that integrates a crisscross optimization algorithm (CSO) with a convolutional block attention module (CBAM) and a temporal convolutional network (TCN). First, a forecasting model is established based on TCN to extract the implicit temporal patterns of the input sequence of power loads. Second, a CBAM module is introduced at the model input side to apply channel-wise and spatial-wise weighting, thereby enhancing the model’s sensitivity to key features. Finally, to address issues such as local optima and limited generalization, the CSO algorithm is proposed to perform secondary optimization on the fully connected layer of the CBAM-TCN model. Using real power load datasets from two typical low-voltage substations in Guangdong province for simulation and modelling, the results show that the proposed hybrid forecasting method outperforms other comparative models and effectively validates its superiority. |
| Key words: low-voltage distribution area load forecasting crisscross optimization algorithm temporal convolutional network convolutional block attention module |
