数据与计算发展前沿 ›› 2023, Vol. 5 ›› Issue (3): 160-168.

CSTR: 32002.14.jfdc.CN10-1649/TP.2023.03.012

doi: 10.11871/jfdc.issn.2096-742X.2023.03.012

• 技术与应用 • 上一篇    

基于LightGBM的夜经济用户级短期负荷概率预测方法

周文涛(),魏光涛,王泽黎*(),张晓晨,任立志   

  1. 国网北京朝阳供电公司，北京 100124
  • 收稿日期:2022-03-23 出版日期:2023-06-20 发布日期:2023-06-21
  • 通讯作者: *王泽黎（E-mail: zeli.cool@163.com）
  • 作者简介:周文涛，国网北京朝阳供电公司，高级工程师，主要研究方向为电力系统自动化。
    本文承担工作为：概率预测模型算法设计与实现。
    ZHOU Wentao is a senior engineer with State Grid Beijing Chaoyang Electric Power Supply Company. His main research interests are power system automation and big data analysis.
    In this paper, he undertakes the following tasks: design and implementation of the proposed probabilistic forecasting model.
    E-mail: wentao_zhou2022@163.com|王泽黎，国网北京朝阳供电公司，工程师，主要研究方向为电力系统自动化。
    本文承担工作为：概率预测模型算法设计、论文写作。
    WANG Zeli is an engineer with State Grid Beijing Chaoyang Electric Power Supply Company. His main research interests are power system automation, visualization, and machine learning.
    In this paper, he undertakes the following tasks: design of the proposed probabilistic forecasting model and paper writing.
    E-mail: zeli.cool@163.com
  • 基金资助:
    国家重点研发计划(2017YFB0202302);国家自然科学基金(62073133);国网北京电力公司实施项目(B702-03210001)

A User-Level Short-Term Probabilistic Load Forecasting Based on LightGBM in Night Economy

ZHOU Wentao(),WEI Guangtao,WANG Zeli*(),ZHANG Xiaochen,REN Lizhi   

  1. State Grid Beijing Chaoyang Electric Power Supply Company, Beijing 100124, China
  • Received:2022-03-23 Online:2023-06-20 Published:2023-06-21

RichHTML

7

PDF

46

摘要：

【目的】 为了度量夜经济中用户级短期负荷的不确定性，基于LightGBM（Light Gradient Boosting Machine）和KDE（Kernel Density Estimation）方法，本文设计了一种夜经济用户级短期负荷概率预测模型框架和预测方法。【方法】 首先，利用LightGBM对待预测用户的历史负荷与影响因素（如天气、日类型等）进行建模。然后，使用该模型预测用户的未来短期负荷，并将LightGBM模型所包含的树的输出作为概率预测的输入，利用核密度估计方法计算该用户未来短期负荷的概率密度及预测区间。【结论】 最后利用北方某城市的多个夜经济用户真实负荷数据进行了实验验证，实验结果表明本方法预测结果准确，鲁棒性高，且对夜经济多类用户均适用。

关键词: 用户级负荷预测, 负荷概率预测, LightGBM, 梯度提升决策树, 核密度估计

Abstract:

[Objective] To estimate the uncertainty of user-level short-term load in the night economy, this paper designs a user-level short-term load probability prediction method based on the light gradient boosting machine (LightGBM) and kernel density estimation (KDE). [Methods] First, the LightGBM is applied to model the historical load and influencing factors (such as weather, day type, etc.). Then, the trained LightGBM-based model predicts the user's future load. The predictions produced by the trees in LightGBM are used as input to the kernel density estimation (KDE) to generate the probabilistic density and prediction intervals of the user's future load. [Conclusions] The experimental results on the actual load of multiple night economic users verify the performance of this framework.

Key words: user-level load forecasting, probabilistic load forecasting, Light Gradient Boosting Machine, gradient boost decision tree, kernel density estimation