| 引用本文: | 肖白,孙旭,张大弛,辛昊阔,姚狄,孔译辉,张晓华.基于DBSCAN和改进ConvLSTM的空间负荷预测方法[J].电力自动化设备,2026,46(2):169-175. |
| XIAO Bai,SUN Xu,ZHANG Dachi,XIN Haokuo,YAO Di,KONG Yihui,ZHANG Xiaohua.Spatial load forecasting method based on DBSCAN and improved ConvLSTM[J].Electric Power Automation Equipment,2026,46(2):169-175. |
|
| 摘要: |
| Ⅰ类元胞的形状和大小会随着电网的发展而改变，给空间负荷预测带来不利影响，且现有预测模型未充分挖掘不同空间分布的Ⅱ类元胞之间潜在的相互作用关系。为此，提出一种利用基于密度的带有噪声的空间聚类(DBSCAN)算法和改进的卷积长短时记忆神经网络(ConvLSTM)进行空间负荷预测的方法。通过DBSCAN分析历史Ⅰ类元胞负荷的异常数据特征，将密度较低且相对孤立作为数据集中异常值的剔除准则；在确定Ⅰ类元胞负荷的合理最大值后，利用网格化技术计算Ⅱ类元胞负荷的准实测值；改进ConvLSTM并构建空间负荷预测模型，该模型通过卷积操作改进门控机制层，选用能保留负荷数据特征的激活函数改进状态更新层，并选用强化学习过程的激活函数改进数据输出层；训练确定模型参数并实现SLF。算例分析验证了所提方法在实际应用中的有效性。 |
| 关键词: 空间负荷预测；DBSCAN ConvLSTM；时空预测；元胞；地理信息系统 |
| DOI：10.16081/j.epae.202510009 |
| 分类号: |
| 基金项目:国家自然科学基金重点资助项目(U22B20105)；国家重点研发计划项目(2017YFB0902205)；吉林省产业创新专项基金资助项目(2019C058-7) |
|
| Spatial load forecasting method based on DBSCAN and improved ConvLSTM |
|
XIAO Bai1, SUN Xu1, ZHANG Dachi2, XIN Haokuo2, YAO Di2, KONG Yihui3, ZHANG Xiaohua3
|
|
1.Key Laboratory of Modern Power System Simulation and Control & Renewable Energy Technology, Ministry of Education, Northeast Electric Power University, Jilin 132012, China;2.Power Economic Research Institute of State Grid Jilin Electric Power Co.,Ltd.,Changchun 130000, China;3.Yanbian Power Supply Company of State Grid Jilin Electric Power Supply Co.,Ltd.,Yanji 133000, China
|
| Abstract: |
| The shape and size of Type Ⅰ cells change with the development of power grid, which brings adverse impact on spatial load forecasting. Additionally, the existing forecasting models have not fully explored the potential interaction among Type Ⅱ cells with different spatial distributions. Therefore, a spatial load forecasting method based on a density-based spatial clustering of applications with noise(DBSCAN) and an improved convolutional long short-term memory neural network(ConvLSTM) is proposed. The abnormal data characteristic of historical Type Ⅰ cell loads is analyzed by DBSCAN, low density and relative isolation are taken as the elimination criterion of abnormal values in the data set. After determining the reasonable maximum value of Type Ⅰ cell loads, the grid-based techniques are used to calculate the quasi-actual values of Type Ⅱ cell loads. The ConvLSTM is improved, and a spatial load forecasting model is constructed, which improves the gate mechanism layer through convolution operation, selects the activation function that can preserve load data features to improve the state update layer, and selects the activation function from reinforcement learning process to improve the data output layer. The model parameters are trained and determined to implement spatial load forecasting. The example analysis verifies the effectiveness of the proposed method in practical application. |
| Key words: spatial load forecasting DBSCAN ConvLSTM spatiotemporal forecasting cell geographic information system |
