计算机工程与应用 ›› 2021, Vol. 57 ›› Issue (18): 275-280.DOI: 10.3778/j.issn.1002-8331.2012-0135

• 工程与应用 • 上一篇    下一篇

融合级联LSTM的短期电力负荷预测

娄奇鹤,刘虎,谢祥颖,马晓光   

  1. 1.华北电力大学,北京 102206
    2.国家电网有限公司,北京 100031
    3.国网电子商务有限公司,北京 100053
  • 出版日期:2021-09-15 发布日期:2021-09-13

Short-Term Power Load Forecasting Using Cascaded LSTM Models

LOU Qihe, LIU Hu, XIE Xiangying, MA Xiaoguang   

  1. 1.North China Electric Power University, Beijing 102206, China
    2.State Grid Corporation of China, Beijing 100031, China
    3.State Grid E-commerce Co., Ltd., Beijing 100053, China
  • Online:2021-09-15 Published:2021-09-13

摘要:

随着我国清洁能源的快速发展,分布式光伏电站得到了大力推广。在光伏电力系统运维中,电力负荷预测是影响分布式光伏电站发电、储能、传输等多个环节进行优化配置的关键因素。针对电力负荷的预测问题,提出了一种级联长短期记忆模型,将电力负荷预测划分为两个阶段:第一个阶段提取电力负荷的周期性特征,得到总体的变化趋势;第二个阶段提取负荷的波动性特征,对总体趋势进行修正,进一步提升预测的准确度。在某地区连续五年的电力负荷数据集上进行了实验验证,结果表明级联LSTM模型能够大幅降低预测误差。该模型可以为分布式光伏电站提供较为准确的负荷预测,能够为其智慧运维服务提供重要支持。

关键词: 电力负荷预测, 长短期记忆网络, 光伏电站, 人工智能

Abstract:

With the rapid development of clean energy in China, distributed photovoltaic power station has been vigorously promoted. In the operation and maintenance of photovoltaic power system, power load forecasting is the key factor that affects the optimal configuration of distributed photovoltaic power station, such as power generation, energy storage, transmission and so on. Aiming at the problem of power load forecasting, this paper proposes a cascaded long-short-term-memory model, which divides power load forecasting into two stages. In the first stage, the periodic characteristics of power load are extracted to get the overall trend. In the second stage, the fluctuation characteristics of load are extracted to modify the overall trend to further improve the accuracy of prediction. In this paper, the experimental verification is carried out on a five-year continuous power load data set, and the results show that the cascaded LSTM model can greatly reduce the prediction error. The model can provide more accurate load forecasting for distributed photovoltaic power plants, which can provide important support for its intelligent operation and maintenance services.

Key words: power load forecasting, long-short-term-memory network, photovoltaic power plants, artificial intelligence